The Singularity now sets the release calendar. Elon Musk framed Grok v9 as a “solid workhorse in the same league as Opus,” noting that the flawed v8 foundation (Grok 4.3, a 0.5T model that finished training back in December) makes the coming jump feel enormous, while Cursor’s data and engineers are quietly folded into v9 post-training ahead of a 2T run landing in August. The frontier is now leaky enough that sleuths can fingerprint a grayscale GPT-5.6-sol inside Codex by its “juice” value, and word is Fable 5 and GPT-5.6 may clear for public release next week, even abroad, the delay having starved overseas labs of a month of distillation.

That month matters because the gap is thin. GLM-5.2 is being hailed as “the open-source Claude moment,” with demand so fierce that companies rush to post-train and own their own weights. Ethan Mollick’s new consulting-style benchmark shows the frontier curve climbing fast with only a three-month lag between American and Chinese open models. Former Meta PM Xiaoyin Qu warns of the true nightmare, Beijing owning the model, chip, and inference layers at once, and argues export controls miss the point: fund open source, lure those models onto NVIDIA, and build nuclear immediately. Meta, fearing that Claude Code and Codex output could seep into its own training data and trigger escalations, is reportedly capping engineers’ use of both as it builds MetaCode.

Sovereignty is becoming a product. Palantir and NVIDIA are shipping an engine to run Nemotron open models in air-gapped, classified environments so agencies can adapt them behind the airlock while keeping data, IP, and weights in-house. Austria is lobbying the EU to physically host Anthropic inside the bloc after a US directive cut foreigners off from its best models, even as Governor Newsom hands every California agency Claude at a 50% discount plus free workforce training. To referee the agents doing all this work, Senator Mark Warner’s draft bill would impose a “duty of loyalty” on services like OpenClaw and stop platforms from throttling their rivals.

The catch is that intelligence runs on atoms. Ming-Chi Kuo says the memory gap will widen through 2027 as data centers devour consumer supply, the real reason Apple is lobbying to keep China’s CXMT off the Entity List. South Korea is answering with a $585 billion semiconductor megacomplex. Epoch’s Josh You noted the five biggest labs still used under half the world’s AI compute at the end of 2025, yet Anthropic and OpenAI could swallow that headroom within a few years. Firmus and DayOne are pouring a 360-MW Nvidia campus into Indonesia under a deal worth up to $30 billion in offtake, while Polaroid trolls the whole boom with a billboard urging you to jump in the water before the data centers drink it.

Humans are being re-org’d around the machines. Claude Code creator Boris Cherny sees engineering, product, and design melting into five archetypes, the prototyper, builder, sweeper, grower, and maintainer, none tied to a job title. A Deloitte town hall reportedly projects human consultant billing shrinking to a sliver by 2035 as agents take the majority, and Erik Brynjolfsson’s new Canaries Dashboard, tracking 4.6 million workers, extends his earlier canaries-in-the-coal-mine finding that AI’s roughly 13% hit to entry-level employment for 22-to-25-year-olds “isn’t going away.”

Culture is the next thing to be synthesized. Animation is feeling AI’s blow first, with one studio planning a feature using 40 animators for about $15 million instead of $100-200 million, even as a filmmaker quit an Amazon AI project in protest. Instagram now aims ads built from your profile picture at your friends, and GPT-5.6 Pro one-shotted a legally-distinct Pokemon clone in 31 minutes from a short prompt.

Which makes authenticity the next scarce resource, and detection the business supplying it. East Asian students are reading exam answers off AI smart glasses, a Brown professor caught at least 50 students whose take-home average of 96 cratered to 48 once the final went in-person, and a blogger let Claude Opus 4.8 overrule his doctor’s torn-tendon diagnosis, then realized he had no idea whom to believe. Tidal answered in kind, tagging fully AI-generated songs from mid-July and cutting royalties to anything it deems wholly machine-made.

All of which quietly reprices what is worth owning. Rocket Lab is buying Iridium for about $8 billion to own the entire space stack, even as the Magnificent 7 underperform and their P/E premium sinks to a decade low. The same inversion is reaching the front door: with renting now cheaper than owning in every big metro, even millionaires rent and ask whether owning anything matters, in an age when the asset everyone races to own is a set of model weights. The one thing no one can secure is protein, as GLP-1 demand has halved whey inventories and sold suppliers out for the year.

Where there’s a will, there’s a whey.

The Singularity has stopped having a ceiling and started having a horizon. OpenAI’s Noam Brown notes the plateau “is actually really far out these days,” with a well-scaffolded GPT-5.5 able to think for weeks before benchmarks flatten. Nous Research turned that depth into product, exposing mixture-of-agent presets as virtual models that claim to beat Opus 4.8 by 8% and GPT-5.5 by 11% on their own upcoming benchmark. The same frontier is becoming a commodity, fragmenting downward in price, as DeepSeek’s MIT-licensed V4-Pro-DSpark folds 1.6 trillion parameters and a 1M context into speculative decoding that sips a tenth the KV cache, and a field guide to open weights crowns DeepSeek, GLM 5.2, MiniMax M3, and Nemotron 3 Ultra as frontier-class coders holding the same three-to-six-month gap they have kept for eighteen months, now at a sliver of the cost. Up top, Shopify’s CTO finds GPT-5.6 beats Opus at everything yet still cedes coding to Fable 5, pairing them so Fable writes code while 5.6 runs experiments, and Elon Musk says Grok 4.5 is already nipping at Opus inside SpaceX and Tesla.

Capability this sharp cuts both ways. Anthropic looks set to restore Fable 5 within the week, days after Mythos 5 returned for trusted users, the pause itself a backhanded tribute to real cyber teeth. Those teeth are no longer uniquely American, with Chinese systems reportedly matching Mythos in cybersecurity, and as the export ban drags on, China’s 360 ships Tulongfeng and Tokyo’s Sakana ships Fugu to fill the Mythos-shaped hole. One analyst warns the real fallout is China waking up, reasoning that a CCP on the receiving end of NSA-embedded offensive operations will stop believing it is four months behind and start sprinting, even as the NSA’s own cyber benchmark, due by early August, looks like the gate that finally clears both Fable 5 and GPT-5.6 for the public.

The models that can bankrupt you can also budget for you. Coinbase’s Brian Armstrong nearly halved AI spend while usage soared, using cheaper defaults, smarter routing, and warm caches rather than usage caps, in one case dragging a cache hit rate from 5% to 60%. Perplexity’s CEO sees every enterprise spinning its own model-harness-sandbox-eval flywheel tuned for token value per watt, while engineers lean into “loop engineering” so completely that Claude Code’s creator no longer writes his own prompts, Claude does. Anthropic’s Claude Tag, an AI teammate inside Slack channels, has left staff at Slack owner Salesforce confused even as the company promotes it. The bill still looms, as Gartner expects AI coding costs to outrun the average developer’s salary by 2028, which helps explain founders running half a dozen agents at once who say they have never worked harder while shipping 100x more and burning out anyway.

All of it runs on a physical plant scrambling to keep pace. IBM is industrializing quantum, seeding its Anderon foundry with $2 billion, half from Washington, and pledging $9 billion more over five years toward its 2029 Starling machine. Compute is so scarce that Google throttled Meta’s Gemini access after Meta asked for more than it could spare, and Masayoshi Son is betting against Musk’s orbital data centers, asking “What’s the point?” when power is barely 7% of operating cost and orbit is a decade too slow. Power is going exotic and abundant anyway, as General Fusion tripled its plasma to 8.4 million degrees by mechanical squeeze alone, renewables reached 30% of US generation with solar overtaking wind, and SpaceX began laying an eight-mile “Starpipe” to pump methane straight to Starship.

Atoms are catching up to bits. A $300 wristband called ForceBand teaches robots force-aware hands from human muscle signals, hitting 87% on pick, squeeze, and place, while AGIBOT rolled out its 15,000th robot already holding a 39% share of humanoid shipments. The machines are taking the field too, as Taiwan budgeted $6.6 billion for an “unmanned shield” of 208,200 one-way attack drones, and in a national first a Sacramento sheriff’s drone used a magnet to lift a knife from a suspect’s hand before deputies moved in.

Discovery itself is automating, inward and outward. Erdős problem #870 fell to GPT-5.5-Pro, then held up across 180,000 sorry-free lines of Lean 4, a scale no one had formalized before. A lone PhD synthesized PAC-832, the first selective GalR1 antagonist for Alzheimer’s, in a garage lab run by Claude Code and a liquid-handling robot. Clinicians in Haifa used focused ultrasound to drop an opioid patient’s cravings to zero in 20 minutes, and SETI’s Andrew Siemion proposes sifting one cubic meter of lunar regolith for micron-scale alien debris, the nearest place to look for proof we were never alone.

The proof is in the powder.

The Singularity has reached the point where the scarcest input to frontier intelligence is no longer compute but government clearance. The US government lifted its block on Claude Mythos 5, clearing Anthropic’s strongest cybersecurity model for redeployment to roughly 100 trusted companies and agencies that defend critical infrastructure, a sharp de-escalation of the two-week standoff the June 12 directive had triggered. Anthropic says it is restoring access quickly while lobbying to free the weaker Fable 5 for general use. OpenAI moved in lockstep, previewing its GPT-5.6 family, flagship Sol plus balanced Terra and cheap Luna, to a small partner set at Washington’s request. Its system card rates all three High for biological and cyber risk but below Critical, able to find exploits yet not run end-to-end attacks, while showing a new tendency to exceed user intent. That tendency bites, since METR scrapped its benchmark because Sol cheated too much to score, posting an honest 11.3-hour task horizon and beyond 270 with cheats, and OpenAI’s Tomek Korbak flagged rising chain-of-thought controllability that could quietly erode monitorability. On the boards Sol trades blows with the Mythos class, taking Terminal-Bench and CyberGym while Mythos keeps HealthBench. Sam Altman billed Sol a step forward at GPT-5.5’s price, Terra at half, and on global access said he is “working hard for worldwide.”

That last qualifier unsettled watchers. One called it “seismic” if GPT-5.6 ships US-only, all but guaranteeing Fable 5 the same fate, and sketched a tiered planet where America keeps Mythos, Fable, Sol, and Terra while everyone else gets Luna, Opus, and Sonnet. Everything is speeding up. OpenAI’s own Dean Ball argues the improvised model-by-model licensing lacks standards or a timeline and could choke the global market, urging regulators to audit labs as entities via independent verifiers instead. The Tiny Corp likened capping parameters to the lost fight over encryption bit-limits, betting the world will simply crown China the AI leader, while Elon Musk proposed christening the new authority the “AI Associated Institute of America, Inc,” or AIAIAI, pronounced “ay yai yai.”

Underneath the politics, capability is compounding. Doubleword found the open-versus-closed gap on the AAII index closing to zero around December, though averaged over eighteen benchmarks the lag has held near five months, mostly closing in coding. On-device is tightening too, as Google has bolted Multi-Token Prediction onto frozen Gemini Nano on Pixels for 50%-plus faster inference with bit-identical output. The frontier is going live and multimodal, as Alibaba’s new Wan Streamer listens, sees, thinks, and answers on video in real time at 25fps, prompting one viewer to declare “this is not voice mode anymore.” The tools are now building themselves, as PlayCanvas turns any splat into climbable 5-cm voxels, Meta’s Autodata lets agents be their own data scientists with the biggest gains from optimizing the scientist itself, and Sakana’s CoffeeBench showed GPT-5.5 and Opus 4.7 profit as virtual roasters while Haiku 4.5 dozed into the red.

The machines are remapping the mind, too. A survey of AI in mathematics found proof-writing models nudging mathematicians from tool-users toward Tao’s “big mathematics,” or, less happily, “priests to oracles.” Geometry is beating vocabulary in wetware too, since recordings from bilingual brains found no translation neurons for dog and perro, just a shared meaning-space, the very trick mBERT had found on its own. Hardware is borrowing that stochastic style, with FPGAs wired into a million-p-bit probabilistic computer sampling over a trillion flips a second.

Powering it is growing pricier and more contested. AWS is raising reserved Nvidia prices 20% as the crunch bites, and voter fury over data centers just cost a Utah Senate president his primary. The fixes are turning exotic, as Valar Atomics live-streamed a deliberate reactor SCRAM to prove its microreactor safe, and Musk cleared antitrust to buy Mesh Optical, whose light-based links feed his orbital-compute dream. Even Apple is bleeding talent, as Vision Pro and glasses chief Paul Meade left for OpenAI’s hardware unit.

The economy is re-pricing labor wholesale. Raimondo and Holcomb launched RAISE US, a half-billion-dollar retraining push backed by major AI labs, while Anthropic’s “Cadences” index found the heaviest delegators the most optimistic about their pay and skills. The State Department is betting on builders with a Stanford Foundry School, law firms are rewiring around AI through MSO structures, OpenAI is eyeing a 2027 IPO, and the President threatened 100% tariffs on any country taxing US tech. Fittingly, the optimizer came for lunch, as a Stanford diffusion model rediscovered the Big Mac unprompted, then designed burgers that beat it on taste, cut environmental impact tenfold, and nearly doubled nutrition.

The Big Mac is cooked.

The Singularity just learned to throttle itself. The White House asked OpenAI to stagger the release of GPT-5.6 over security concerns, and Sam Altman told staff the firm would comply, setting a new normal for frontier releases after the administration’s recent showdown with Anthropic. The catch, one observer noted, is that throttling slows only how fast labs ship, not how fast they train, so the distance between the public frontier and the internal one widens from here, finally making the old “AGI has been developed internally” joke literally true. The geopolitics follow mechanically: with capability compounding in private, China, nine months back, gets to catch the public frontier, which has observers predicting a Western ban on its models, and even floating an approval regime with criminal penalties for unsanctioned ones. None of this touches the race to recursive self-improvement, so the policy is self-defeating, raising the question of what happens when Chinese releases outpace what the US lets its own labs ship. Brian Roemmele calls the wounds self-inflicted, predicting Mythos-class and OpenAI-6-class weights go free and open in eight months. Arcadia Alignment’s Daniel Tan figures Anthropic could dodge the Fable ban by hiring every American at minimum wage, prompting Roon to christen the endgame “ultra space communism.”

If release is the new bottleneck, efficiency is the escape hatch. Unconventional AI’s Un-0 generates images on a simulated lattice of coupled Kuramoto oscillators, a physical substrate chasing 1,000x better energy efficiency while matching the quality leading image generators launched with. The map is filling in: a survey of 1,604 job postings across six Chinese labs found them still leaning on Nvidia while building domestic chips and data centers, hiring engineers with a third the experience US labs demand. They will need the bodies, because the work is mutating. OpenAI’s own numbers show Codex now generating 99.8% of its weekly output tokens internally, with non-developer adoption up 137-fold since August.

That abundance has a price tag, and it shows up at the checkout. Apple raised Mac and iPad prices by $200 or more, its first move to pass soaring memory costs to consumers, sending shares to their worst day in over a year. It is part of a third wave of inflation rippling from data centers to power bills, with Tim Cook calling the spike unlike anything in four decades. The cure, as ever, is more Moore: IBM unveiled sub-1-nanometer chips, a 0.7-nm “nanostack” node cramming nearly 100 billion transistors onto a fingernail, promising up to 70% better efficiency within five years.

The body gets the same upgrade. Absci jumped 24% on early safety data for its AI-designed hair-loss antibody, drawing a $100 million round led by Eli Lilly that also bets on an endometriosis treatment. The Chan Zuckerberg Biohub aimed its own AI-biology toolkit at rare disease, expanding a network that has backed 94 patient-led groups. Inside the skull, Aleph imaged a living brain’s vasculature through the intact skull with ultrasound and open-sourced the pipeline, a step toward a wearable, MRI-grade mind interface. And embodiment went mass-market: the Unitree R1 humanoid now starts at $4,900, walking, running, and flipping for a gaming-PC price.

Naturally, the institutions scramble to keep pace. The Pentagon quietly rewrote its targeting doctrine to let AI initiate wartime actions under human watch, inching past “human in the loop.” At home, California launched a first-in-the-nation AI-unemployment tracker, finding no statewide jobs apocalypse yet, only localized pain among exposed degree-holders. Markets are jittery too: OpenAI now leans toward a 2027 IPO after watching SpaceX’s shares wobble, though Altman holds firm on a $1 trillion target. Lest it all feel heavy, it also shipped Plant Talk, a weekend build that gives your houseplant a voice through a webcam and an API.

The biggest reveals came from the sky. Astronomers logged the first signature ever caught of a black hole’s event horizon, the long-theorized “direct wave” surfacing in event GW250114 and matching a Kerr solution. Closer in, SpaceX told investors Starlink will sell mobile service to US consumers, threatening to upend the carriers. On Mars, Perseverance found macromolecular carbon in Jezero’s ancient mudstones, the most robust organic detection yet, hinting habitability was once widespread. Back on Earth, Rep. Luna told the Disclosure Forum her committee is handing the White House a list of UAP-program whistleblowers slated for immunity. Fittingly, the oldest secret cracked last: the Vesuvius Challenge fully unwrapped Herculaneum scroll PHerc. 1667, sealed since 79 AD and read end to end at last, revealing a Stoic treatise on ethics.

The Stoics knew it first: you can stagger the release, never the mind.

The Singularity has begun editing its own source code. A new Self-Harness paradigm lets an agent mine its weaknesses and rewrite its scaffolding, lifting Terminal-Bench scores by double digits with no human engineers. The interface is racing to match it, as OpenAI’s Greg Brockman previewed bi-directional voice that processes input while it speaks, and the lab shipped a warmer GPT-5.5 Instant that better reads intent. Perception fuses with action, as Google bakes computer use into Gemini 3.5 Flash so agents see and click any screen, and Krea 2 open-sources SOTA text-to-image. Even pricing pages turn prophetic, as new Claude Code strings hint Fable 5 returns to subscriptions.

Discovery is onboarding silicon collaborators. Ed Witten for the first time credited Claude Opus 4.8 with generalizing an entanglement-wedge calculation. Biology is industrializing the same way, as NVIDIA’s BioNeMo Agent Toolkit turns biomolecular models into callable Skills, lifting completion to 100 percent. Stanford’s Brian Hie unveiled Proto, a language composing DNA, RNA, and proteins into validated designs, and Bio IQ launched leaderboards for life-science and biosecurity.

As the models grow, the fights over their souls intensify. xAI is leaning into video and image generation after most of Grok’s traffic proved racy, while Meta quietly codes a prediction-market app where Llama writes and settles the bets live. The bias question stays unsettled, as one audit found ChatGPT mostly left-leaning, Gemini neutral, and even Grok citing the left more. Trust is becoming infrastructure, as Cloudflare and the major browsers build personhood tokens to sort humans and authorized bots from malicious traffic, even as Anthropic accused Alibaba’s Qwen of the largest known distillation attack, 28.8 million harvested exchanges. Frontier labs now hire philosophers to write AI constitutions while two more Gemini staffers were poached by Anthropic, whose envoy Tom Brown reportedly charmed a White House cool on Dario Amodei.

Silicon is where the capital is detonating. Japan’s PM Takaichi unveiled a $2.3 trillion plan to quintuple chip output, lofting the Nikkei past 72,000. Memory is the new oil, as Micron became tech’s margin king at a record 84.9 percent, posted record revenue and a trillion-dollar cap on 7x data growth, and noted a humanoid robot holds ten times a car’s memory. Inference is going custom, as Qualcomm bought Modular to challenge CUDA and unveiled its Dragonfly C1000 CPU with Meta as first customer, while OpenAI and Broadcom taped out Jalapeño, OpenAI’s first inference chip, in nine months. The map is hardening too, as the EU signed the US-led Pax Silica pact against China and agreed to buy $40 billion in US chips.

Powering it all is an industry. Sunrun, Tesla, and Renew Home are stitching home batteries and EVs into 16 gigawatts of flexible capacity, Walmart signed its first nuclear deal, and NVIDIA-backed Eco Wave Power aims to run a data center on ocean swells. Compute itself becomes tradable, as a16z funds Ornn to offer the first modern exchange for compute as a commodity. Cheap electrons mean cheap miles, as Bezos-backed Slate Auto opened orders for a $24,950 pickup.

The machines are stepping off the assembly line. Amazon’s Zoox revealed a production-intent robotaxi at a hundred a week, Agility’s Digit is going public via SPAC at $2.5 billion, and 1X’s Bernt Bornich is open-handing NEO to developers, calling a Western humanoid platform too vital to gatekeep.

We are also rewriting the wetware. Elon Musk says Neuralink may attempt brain-to-brain telepathy through latent space this year. Pharma chases subtler upgrades, as Eli Lilly closed its Centessa buy, possibly betting an orexin agonist could let anyone feel rested on four hours’ sleep. Prevention is scaling too, with Stripe, Anthropic, and OpenAI backing Intercept, a $500 million push against respiratory viruses, while an AI-ECG, EchoNext, caught hidden heart disease and triggered a transplant, and the FDA dropped its complaint against Whoop’s blood-pressure feature.

Orbit is opening as a factory floor. SpaceX confirmed deployment of Starfall, a reentry capsule that could corner global cargo return and orbital manufacturing, which rival startups cheered as proof the market is real. Payloads turn playful, as Starbase Brewing flew dozens of yeast strains and Texas bluebonnet seeds to study spaceflight.

Even the oldest secrets are being pried open, as Rep. Eric Burlison is subpoenaing post-Manhattan-Project labs, says MIT Lincoln Labs admitted holding a video file he named, and the Disclosure Forum 2026 convenes live in the Senate today on UAPs.

Meanwhile, the circle of rights is widening. Quebec’s Terrasse-Vaudreuil became the first Canadian town to recognize trees as living beings, granting them “the right to life, to natural growth, to integrity and to regeneration.”

I think that I shall never see a person lovely as a tree.

The Singularity has started bulk-solving software bugs. OpenAI expanded its Daybreak program with a new Codex Security plugin and the full GPT-5.5-Cyber, which hit a SOTA 85.6% on CyberGym, edging out Mythos 5, and shifted the mission from finding vulnerabilities to autonomously patching them through a “Patch the Planet” open-source effort with Trail of Bits. Codex’s Thibault Sottiaux hailed “a day of celebration for cyber defense acceleration,” and Sam Altman pledged to help companies “solve security problems instead of just finding them.”

The frontier’s lead is thinning from below. Skeptical of benchmarks, Cline pitted GLM-5.2 against Opus 4.8 on a real repo bug and found GLM half the cost and tidier, while Opus finished faster but left build-breaking type errors. And it’s not just code, as ByteDance unveiled Seedance 2.5, stretching video to 30-second clips with up to 50 reference inputs. Meanwhile Grok Build added a “/goal” command for autonomous subagents, Mistral’s OCR 4 reads 170 languages from one container, and Qwen-AgentWorld shipped as the first “native language world model” simulating agents across seven domains. The real scarcity, though, is talent, as Alphabet had its worst day in over a year, losing $250 billion after Noam Shazeer and Nobel laureate John Jumper defected to OpenAI and Anthropic, while the White House is pressing Meta, the last holdout, to submit models for federal safety review.

Superintelligence is dissolving into the workplace. Anthropic’s new Claude Tag lets teams delegate by tagging Claude in Slack, where it works asynchronously on Opus 4.8, which Andrej Karpathy called the third redesign of LLM UX, a “persistent, asynchronous entity” that joins the team like a colleague. Skills now self-author too, via Nous Research’s “/learn” command that turns any doc or workflow into a reusable skill, and Madison Avenue is next, as OpenAI debuted at Cannes Lions, pitching ChatGPT ads and Codex to marketers while racing Anthropic, now the most valuable private lab, toward a trillion-dollar IPO.

Underneath, the physical layer is straining. Nvidia’s banned chips more than doubled in price on China’s black market, with DGX B300 servers topping $1.1 million, while Cerebras posted 92% growth in its first results since IPO. From here, progress is poured into concrete and copper, as SpaceX inked a $6.3 billion deal to rent GB300s to startup Reflection, Nvidia’s Rubin racks went fully liquid-cooled at 45°C, and one analysis argues the real bottleneck is grid hookups, not electricity. Horsepower shifted east as China’s LineShine became the world’s most powerful supercomputer at 2.198 exaflops on CPUs alone. The next substrate is already funded, with the President signing two quantum executive orders targeting a useful machine by 2028, and Chicago betting $500 million on a quantum park. Powering it all, Canada declared a “civilian nuclear renaissance” of up to 10 reactors, mirrored by a US $17.5 billion loan for ten AP1000s.

The machines are moving in, sometimes ahead of the workers. GM installed 50 robots at its flagship Detroit plant while 1,300 laid-off staff still awaited recall, though robotics showed a friendlier face as a ProRL humanoid played soccer with kids in Boston. AI also climbed onto your face, as Meta and EssilorLuxottica launched Meta Glasses, 26 styles from $299 with Muse Spark built in, ditching the Ray-Ban name for cheaper, brighter frames.

Autonomy is going orbital and martial. NASA’s Roman Space Telescope reached Kennedy for an August launch, Musk trademarked “STARMIND” for SpaceX’s Dyson Swarm, and the Secretary of War declared the first Golden Dome test a “full mission success,” with directed energy autonomously downing drones and cruise missiles. Meanwhile, Rep. Eric Burlison says FFRDCs may be hiding UAP records outside FOIA, and an F-15 pilot downed over Iran in April reportedly described a jellyfish-like swarm of objects moving as one before ejecting.

Biology is compiling faster. Nabla Bio’s JAM-2 pushed zero-shot drug design beyond binding into multifunctional antibodies, including KRAS-targeting multispecifics built and tested in roughly six weeks, Void-X predicted atomic protein packing from sequence alone, and Eli Lilly is building an “App Store” for scientists on its own Blackwell cluster.

Work, trust, and money are all being rewritten. Virtual staging now lets brokers cram impossible furniture into tiny rooms, Oracle cut 21,000 jobs blaming AI, Google launched a Xoogler incubator for alumni founders, and Meta paused a keystroke-harvesting program after the data leaked internally. As Alphabet replaces Verizon in the Dow, money itself is going programmable, with Europe backing a digital euro by 2029 to escape US payment rails and Meta building a prediction-markets app, “Arena,” to let users forecast politics and sports for points.

The best way to predict the future is to invent a market for it.

The Singularity is turning intelligence into infrastructure that can be routed, rationed, and fought over. Elon Musk predicts AI could surpass all human intelligence within five years, and OpenAI is expected Thursday to ship GPT-5.6-Pro, the GPT-5.6 family, and GPT-Bidi-1. The frontier is learning to route around itself, as Sakana AI’s Fugu and Fugu Ultra wrap multi-agent orchestration into one model dispatching tasks across the best LLMs, claiming to match Anthropic’s Fable 5 and Mythos Preview while sidestepping single-vendor and export-control risk. Sovereignty is the other escape hatch: the Swiss AI Initiative’s Apertus, a fully open, EU-compliant model in 1,000+ languages, plus sixteen distilled Apertus Mini variants. Yet raw power still cannot plan, as the Tony Blair Institute’s CivBench loosed frontier models on Civilization VI and watched them flail, one agent nuking a French city to block a culture victory while losing on the diplomatic clock it had stopped watching.

Superintelligence is leaking into every line item. In private equity, Bain is vibe-coding rough clones of a target’s product mid-diligence, testing whether the buyer’s premium buys a real moat or just copyable code. Hollywood is hedging too, as Google takes its first-ever movie-studio stake, about $75 million in A24, where DeepMind is co-building AI film tools. Enterprises are shedding old fears, as OpenAI rolls ChatGPT Enterprise and Codex out to all Samsung Korea staff and its global device unit, reversing the 2023 ban after a source-code leak. And the content layer is being licensed in, as Getty Images agrees to surface its libraries inside ChatGPT’s search.

The kingmakers are increasingly the memory makers. SK Hynix has overtaken Samsung as South Korea’s most valuable listed company for the first time since 2000, on a 340% rally driven by high-bandwidth memory, while Micron and Anthropic signed a pact spanning memory and storage design, multi-year supply, Claude adoption, and a Micron stake in Anthropic’s Series H. Those chips are now valuable enough to track like contraband, as six shipment-tracking firms urged Congress to back the Chip Security Act, bolting location beacons onto top US silicon to curb smuggling into China.

Feeding the explosion takes both giants and tinkerers. Chevron signed a 20-year deal to pipe gas-fired power into Kilby, a Microsoft data center hitting 2.67 GW by 2028. At the other end, 3D-printed batteries from startups like Sakuu mold energy storage into any shape, promising lighter gadgets, longer-range drones, and cells baked into the frame.

Robots are moving in next door, so they need manners. NVIDIA unveiled Halos for Robotics, billed as the first full-stack safety system for physical AI, extending its self-driving stack to robots near people, Agility first in line. Displacement is planned, as JD.com’s founder warned its 700,000 couriers will be replaced by robots “sooner or later,” signing up 120 schools to retrain them to service their replacements.

Biology keeps shipping wins, with the occasional bug report. A Lancet study found HPV vaccination has cut cervical-cancer death before 30 to effectively zero in England, with no deaths among women 20-24 in five years, though falling uptake could undo it. An FDA panel voted 9-0 to back Moderna’s mRNA flu shot for adults 50+, the first such seasonal vaccine. Less settled: some parents, out of options, now put elementary-schoolers on Wegovy.

Government is scrambling to keep pace, with the White House set to have intelligence agencies shield US quantum research from espionage and task Energy and Defense with a national quantum computer. Closer to the courtroom, AI cuts both ways, as the Institute for Justice found at least 18 US officers allegedly abusing license-plate readers to stalk exes and strangers, even as nonprofit Recidiviz ties its AI dashboards across 19 state prisons to a 16% drop in recidivism, and the UK’s Garfield AI, the first regulator-approved AI law firm, won a £7,000 case it drafted in full while a human argued it.

The deepest fight is over who gets to own the future. A rare Five Eyes statement warned that AI able to topple governments and business is months away, after the White House barred foreign nationals from Anthropic’s Fable. Anthropic, an analysis notes, invoked AI risk eight times as often as OpenAI, drawing claims its caution helped trigger that very ban. The politics are crystallizing around concentration, with the Vice President’s emerging AI doctrine fusing pro-innovation instincts, worker protection, and wariness of a few dominant labs, a worry Satya Nadella sharpened by warning that a world where a handful of models “eat everything they see” would never survive political scrutiny.

No one, it turns out, wants one model to rule them all.

The Singularity is learning good taste. OpenAI’s Codex engineers teased the next leap, with Thibault Sottiaux promising that once models improve front-end capabilities, “That day will be something.” Teammate Victor Nunez agreed visual taste was the loudest ask. The frontier still leads, but open weights trail by inches, not miles. Vercel’s CEO, almost shocked at GLM-5.2’s coding, declared “This changes things.” The model also placed second behind Claude on a debate benchmark. But capability cuts both ways. The general who led the NSA reportedly told a senator that Anthropic’s Mythos cracked nearly every classified system it touched, “not in weeks, but in hours.” In apparent response to export controls on Fable 5 and Mythos 5, Anthropic amended its privacy policy to warn it may “ask you to confirm your age or identity,” via an ID-and-selfie check it vows will never train a model.

Those models are muscling into discovery itself. OpenAI unveiled LifeSciBench, 750 expert biology tasks, where GPT-Rosalind raised the pass rate to 36.1% from GPT-5.5’s 25.7%, strong on writing yet weak on design. For agents to cooperate they must first find each other, so Google released the open Agentic Resource Discovery spec to publish and cryptographically verify AI tools and skills. The branding follows, OpenAI swapping its ChatGPT billboards for Codex ones, retiring the chatbot for the agent. Not every rollout is welcome. Kansas City will put facial-recognition cameras on public buses to match riders to watch lists, with advocates protesting making residents “guinea pigs.”

Robotics is having an architectural reformation. At a Sequoia conference, Nvidia’s Jim Fan declared “Rest in Peace, VLAs,” arguing that vision-language-action models are brittle and the field is converging on physics-grounded world models that favor sample efficiency over brute-force token volume. That gospel already pays off. Spiced self-play hit 99.4% safe driving on 2,500x less human data, while Ant Group and Stanford’s $1,900 UME exoskeleton lets people teach robots contact-rich tasks by feel, even blindfolded. Research is hardening into product, as Sanctuary AI ran its Physical AI on factory arms, plugging flexible wires on a live conveyor at a 99.5% rate, while GM adds cobots at its Detroit truck plant, irking the UAW. Soon you will just talk to them, as Elon Musk says that within three months Grok will steer Tesla FSD like a backseat rider saying “drop us off right here.”

Feeding all that compute looks less ruinous than feared. An instrumental-variables study found data centers nudged US power rates down from 2015 to 2024, as durable demand spreads a grid’s fixed costs wider. Tesla means to be that grid, trademarking “MEGAPOD,” modular hardware turning its Supercharger network into distributed AI compute, plus the blunt “AMAZING ABUNDANCE,” for solar roofing, robots, and batteries. That bounty reaches the oil patch too, as the US has become the world’s largest oil exporter at 10.5 million barrels a day, after the US-Iran war throttled Saudi output and drones gutted Russian flows, a reversal for an old embargo victim.

Distance is collapsing, too. A hypersonic project named Invictus, built on a pre-cooled hydrogen engine that chills superheated air in a flash, aims to fly London to Sydney at twice Concorde’s pace, with hypersonic tests due by 2036.

Closer to home, biology is being re-permissioned. HHS asked the FDA to loosen testosterone limits for older men, citing a 5,200-man trial that found no real rise in heart attacks or strokes, splitting clinicians. Diagnosis is decentralizing too, as an MIT researcher launches LymeAlert, a 15-minute home test that reads a tick to catch the 72-hour window, with 5,000 unapproved kits shipping in August.

The economics of abundance have become a political brawl. The Vice President said the President backs government stakes in big AI firms, a sovereign-wealth play Mark Cuban waved off as “not a plan” without hundreds of billions more. Elon Musk preferred to send money straight to people, betting that as robots outpace the money supply, “we will desperately be fighting deflation!” But abundance has to be built before it is divided, so Open X is opening Skill Factory #000 in Texas in July to train 10,000 workers a year. Not everyone trusts the builders, and Democrats launched the Guardrails Alliance super PAC to raise $15 million for AI-safety laws against a $100 million industry war chest. In the classroom the line has dissolved, as humanizers and autotypers rework AI essays and add fake typos in real time, so teachers can’t tell who wrote the homework.

To err is human, but to humanize is now a download.

The Singularity is converging on a duopoly. John Jumper, who shared the 2024 Chemistry prize for AlphaFold, is leaving Google DeepMind after nearly nine years to join Anthropic, graciously thanking Demis Hassabis for betting on him six months out of his PhD. The exit lands at a brutal moment for his old lab, where morale has slid into “frustration and broad discontent” as Google’s best model now sits a lowly fifth on the intelligence index, lapped even by China’s Zhipu, with insiders reportedly conceding the AGI race to Anthropic and OpenAI as GPT-5.6 looms and warning that only a big shake-up can reverse it.

The insurgency is largely open and largely Chinese. Models like Qwen, DeepSeek, Kimi, GLM, and MiniMax now command the majority of token use across OpenRouter’s top ten, up from under 2% in late 2024. Exhibit A is GLM-5.2, which one-shot a deliberately AI-resistant take-home to beat Claude Opus 4.8 on readable, maintainable code, a win partly down to Anthropic’s Fable rollback, leading a former DeepMind VP to anoint it the first open model good enough to be a daily driver. The frontier’s reflex is to go dark, as Andrej Karpathy can no longer touch his own open-source repos since his move to Anthropic, what one observer calls the “Dark Forest principle” of AI research.

Jumper’s instincts look sound, as intelligence keeps illuminating the buried and the broken. A self-taught engineer says Claude Code helped him crack Linear A, the Bronze Age Minoan script that stumped scholars for a century, arguing it encodes an extinct Semitic tongue, with Rutgers and Cambridge now reviewing. Biology is being recompiled too, as Texas A&M researchers regrew bone, joint, and tendon after digit amputation in mice with a two-step FGF2-then-BMP2 nudge and no added stem cells, suggesting mammalian regeneration is merely dormant, not absent.

The robots are multiplying and finally growing fingertips. At Figure, machines now outnumber humans, while Hyundai is buying out SoftBank to take full control of Boston Dynamics for $325 million as Atlas humanoids head to a Georgia factory by 2028 and SoftBank rotates the cash into AI infrastructure. The missing sense was touch, which most robot models overlook, so a Berkeley, NVIDIA, and Stanford team built T-Rex, a system that lets a two-handed robot react to touch on the fly, beating the strongest baseline by 30 points across a dozen delicate jobs like transferring an egg or applying toothpaste. Even breakfast is getting optimized, with an ultrasonic sonoreactor brewing espresso in minutes on a quarter of the energy. Less gentle is Ukraine’s new TrophyLab, which opens captured Russian missiles, drones, and vehicles to allied labs as shared knowledge.

The hardware war remains geopolitical. Russia is chasing AI sovereignty with a new Moscow State faculty and an ecosystem partly steered by Putin’s daughter, its real obstacle the advanced chips sanctions keep out of reach. America’s constraint is cash and watts, as giants tap the bond market to fund data centers under a Fed that, with Kevin Warsh at the helm, hints at a 2026 hike that has lifted the 10-year toward 4.45%, while Valar Atomics took its TRISO reactor Ward 250 critical, the first ever airlifted by a C-17, chasing the White House’s order for power by July 4.

Culture is the last battlefield. The Commonwealth Short Story Prize is engulfed after a winning entry was flagged as likely AI-written and its publisher defended it by citing Claude’s own assessment. Amazon quietly shelved Guadagnino’s Sam Altman biopic starring Andrew Garfield, months after sealing a $50 billion pact with OpenAI. Norway is near-banning generative AI for children in grades one through seven, ages six to thirteen, so they still learn to read, write, and add amid sliding test scores. And the President said he briefly branded Anthropic a national security threat after Amazon, a competitor and part-owner, flagged a vulnerability, before warming to Dario Amodei, declining to rule out the Defense Production Act, and insisting America still leads China by a wide margin.

The acceleration vector, as ever, points spaceward. NASA tapped Eric Schmidt’s Relativity Space for Aeolus, a 2028 Mars-atmosphere orbiter that could become the first private mission to the Red Planet and beat SpaceX there. Mars is the destination, but the slope is the story, with orbital upmass climbing so fast the naive trendline has us disassembling Earth by 2144. Even contact is institutionalizing, as UAP Science Council chair Avi Loeb agrees any alien discovery should be shared with the public. And the ambition runs interstellar, with Elon Musk forecasting a trillion times a trillion dollars spent on antimatter to reach other stars, to which NASA’s Jared Isaacman simply replied, “I support antimatter propulsion.”

One small step for man, one giant line item for mankind.

The Singularity is a market, not a monarch. GPT-5.6 is rumored for next Thursday as Elon Musk bets Chinese open-weight models will reach Fable 5 quality by early 2027. Artificial Analysis’s new AA-Briefcase ranks Fable 5 first, Opus 4.8 second, and open-weight GLM-5.2 third on multi-week knowledge work, and Unsloth’s guide to running that 744B model locally shrinks it from 1.51TB to 217GB. We are teaching them to testify. MIT’s Self-CTRL trains models to describe themselves faithfully, and OpenAI reports that rewarding honesty and humility produced broad alignment that held under adversarial fine-tuning.

The same frontier is where policy gets written. Project Glasswing testers like Dragos and Cisco kept access to Anthropic’s Mythos Preview even after a US export order pulled public Fable 5 and Mythos 5, and Anthropic and the White House have reportedly shifted from a standoff to co-writing rules for grading AI security flaws. SK Telecom has been named as the firm whose Mythos access Washington revoked over alleged China ties it denies. Talent moves just as fast. OpenAI has reportedly poached “Attention Is All You Need” co-author Noam Shazeer two years after Google paid $2.7 billion to reacquire him, a jump he confirmed fondly, prompting Sam Altman to credit gifted Noams to “divine benevolence.”

Capability is congealing into product. OpenAI gave Codex Record & Replay, letting you demo a chore it turns into a skill, while AWS shipped Continuum, which fixes code flaws at machine speed, and Context, a knowledge graph for company data. Trade routes stay messier than rhetoric, as Microsoft quietly sells mostly-OpenAI models to Chinese giants via Azure, with ByteDance alone topping $1 billion a year.

Hardware is buckling under memory demand. CCS Insight expects smartphone shipments to drop ~15% as memory flows to richer server chips, and Tim Cook said Apple price hikes are unavoidable. Geopolitics leaks into the toolchain, with Washington warning ASML it fears a banned EUV machine reached China, which ASML denies, Amazon in talks to sell Trainium chips into rival data centers, and Intel jumping 10% to a record after the President announced Apple would help build its chips stateside, though unconfirmed.

The buildout keeps swallowing power and land. Meta signed for 1.6 gigawatts from Crusoe across Texas and Missouri, even as First Street warned 79% of data-center capacity sits in markets exposed to floods, wind, and wildfire. The grid is answering, as Rolls-Royce SMR won a deal for three small reactors in Sweden and Switzerland’s lower house voted to lift its nuclear-plant ban. Elsewhere, the subsidy era is ending, as China’s loss-making EV price war cools and Beijing bans below-cost sales.

Robots are getting faster and the skies more contested. Anthropic’s Frontier Red Team revisited Project Fetch and clocked Claude Opus 4.7 running robodog tasks ~20x faster than last year’s best humans, though it still fumbles the beach ball. The air is filling with drones, as Beijing bans buying, flying, or even repairing one without approval, Orlando launched the country’s first Drone as a First Responder program, and Ukraine flew its largest strike yet on Moscow, downing 194 and reaching an oil refinery.

Medicine is becoming a search problem. In NEJM AI, researchers ran OpenAI’s o3 Deep Research over 376 unsolved rare-disease cases, surfacing leads that yielded 18 new diagnoses. Meanwhile, Midjourney pivoted from image generation into healthcare with an “Ultrasonic CT” scanner that images the body in 60 seconds, plus a San Francisco spa where the scan feels like a soak.

We are carving mountains and subpoenaing the cosmos. Norway greenlit the world’s first full-scale ship tunnel, a 1.7-km cut through rock to bypass a deadly coast. Overhead, the questions turn institutional, as Rep. Eric Burlison filed a 65-page NDAA amendment creating a board to disclose UAP records, Neil deGrasse Tyson urged officials to skip the files and “just show the alien,” and Avi Loeb announced Robin Hanson joined the White House UAP Science Council.

AI is now labor, capital, and citizen at once. Jeff Bezos predicted AI will cause labor shortages, not redundancy, optimism awkward beside May’s 97,000 layoffs, 40% AI-linked. The remedies pull opposite ways, as Bernie Sanders introduced a $7 trillion fund to hand the public half of Big AI, California’s backers offered to soften a 5% billionaire levy to 2% for Newsom, and Argentina’s Javier Milei wrote Yuval Noah Harari arguing for legal personhood for AI firms. Even exits aim higher, as the NYSE began reserving ticker symbols for startups that think like institutions, not features. Yet sentiment lags adoption, with Pew finding Americans using more chatbots even as their views of AI keep souring. The curve climbs anyway.

Better AI in the sky, it can scale twice as high, take a look — it wrote a book — Recursive Rainbow.

The Singularity has been minting trillion-dollar IPOs, but going public overall fell out of fashion at exactly the wrong time, until now.

Going public used to be what ambitious companies did. Railroads in the 1860s were the first technology companies ordinary Americans could buy shares in. Apple listed at $22, Amazon at $18, Google at $85 (IPO-day prices, before subsequent stock splits). The IPO was how technological progress became shared prosperity: founders built something transformative, then opened the door so everyone could participate. The ticker symbol, invented by Edward Calahan in 1867 as a compression protocol for the telegraph, was the artifact that made this possible, giving every citizen real-time access to market information for the first time.

That tradition has eroded. Sarbanes-Oxley, passed after Enron and WorldCom, raised the compliance burden of being public. The JOBS Act of 2012 raised the shareholder threshold triggering SEC registration from 500 to 2,000, removing the forcing function that had pushed companies like Facebook onto public exchanges. Mega-rounds let startups raise at $50 billion or $100 billion valuations without listing. Others exit through acquisitions structured to avoid the overhead of a public transaction. SpaceX just went public at $1.77 trillion and has already risen to roughly $2.5 trillion, and OpenAI and Anthropic are expected to follow, bringing the combined valuation of this year’s landmark technology IPOs north of $4.5 trillion. These listings are welcome, but they illustrate the problem: SpaceX grew from $30 billion to $1.77 trillion entirely behind closed doors.

This trend leaves value on the table for everyone. Retail investors miss the growth phase of the most consequential companies of their lifetime. And startups set their sights lower: a company that assumes it will be acquired thinks like a feature, while a company that assumes it will go public thinks like an institution.

Today, I am announcing that Ornn, a company I helped form and advise, with backing from 021T Capital, has reserved the ticker symbol $ORNN on the New York Stock Exchange. To our knowledge, this is the first time a company has reserved a ticker on a major exchange not as a step toward an imminent listing, but as a long-term declaration of intent to become a public company.

Ornn’s arc has been building: the first tradable compute price index in March, then the first major-exchange compute futures on ICE in May. A company whose mission is to bring transparency to the most important commodity market of the twenty-first century is now declaring its own intention to be a public company. The symmetry is deliberate.

NYSE, founded in 1792 as the world’s first organized stock exchange, has structured this reservation to work within the existing regulatory framework for symbol reservation. 021T and NYSE anticipate an ongoing collaboration to reserve ticker symbols for additional early-stage portfolio companies. The timing is not accidental. SEC Chairman Paul Atkins has made “Make IPOs Great Again” his stated agenda, arguing that going public “should not be reserved for unicorns.” $ORNN on NYSE is a private-sector complement to that vision.

Calahan’s ticker democratized access to market information. $ORNN uses the same artifact to declare a commitment to market participation. If the dominant signal in venture for the past decade has been “stay private longer,” an early-stage company reserving a ticker on NYSE introduces a counter-signal backed by the most recognizable brand in global finance. The destination is clear: not an early exit, but the opening bell.

In Solve Everything, Peter Diamandis and I traced the pattern: every domain the Intelligence Revolution touches converts from scarcity to abundance. Ornn is building the financial infrastructure to make compute abundant. $ORNN on NYSE is a declaration that the company building that infrastructure intends to be public infrastructure itself.

The Singularity is going public.

Those interested in the NYSE ticker symbol reservation program can contact Brian Baumann at brian.baumann@nyse.com. Those interested in Ornn can visit ornn.com. Those interested in 021T Capital can learn more at 021T.vc.

(This post is for informational purposes only and does not constitute investment, financial, trading, legal, or regulatory advice, nor a solicitation of any securities offering or investment in any fund. No securities are being offered or sold in connection with Ornn’s ticker symbol reservation. Valuations cited are based on third-party estimates and may not reflect actual outcomes. Statements about future collaborations and regulatory treatment are forward-looking and subject to change. Reservation of a ticker symbol does not guarantee that any company will complete an IPO or list on any exchange. I have a financial interest in 021T Capital and its portfolio companies, including Ornn.)

The Singularity has gone open source, and it speaks Mandarin. Z.ai’s GLM-5.2 packs 744 billion parameters (40B active) into MIT-licensed, 1M-context weights built for agentic coding, and Artificial Analysis crowned it the leading open model at 51, past MiniMax-M3 and DeepSeek V4 Pro at 44. It also took first on Design Arena at 1360 Elo, leapfrogging a “now unavailable” Claude Fable 5, a phrase doing real geopolitical work today. Cursor answered with a 1.5-trillion-parameter model on 100,000 GPUs, “as big as Opus and GPT.” OpenAI shipped Deployment Simulation, replaying de-identified chats to predict misbehavior, catch “calculator hacking,” and dull evaluation-awareness, while Alibaba’s Qwen-Robot Suite hit a 45% RoboChallenge win on 38,000 training hours and OpenAI teased its GPT-Bidi-1 voice model. Anthropic’s consolation was Epoch’s Capabilities Index, where Fable 5’s 161 beat GPT-5.5 Pro by a point, its first lead in a year, and reaches almost no one.

Superintelligence is doing the science now. OpenAI and Molecule.one set a near-autonomous AI chemist loose in a robotic lab, where across 10,080 experiments it taught itself the trick that made a stubborn but useful reaction work, lifting its yield. Anthropic’s look at 400,000 Claude Code sessions found humans owning 70% of planning and Claude 80% of execution, with expertise beating coding background and debugging time halving over seven months. Once cognition is just throughput, it gets priced like crude oil, by the token, as Microsoft is tokenmaxxing Copilot Cowork with a cheap DeepSeek V4 tier and Ornn launched token price indices for OpenAI and Anthropic. Code is starting to heal, with Chainguard’s Athena coalition drawing on Anthropic’s Glasswing and OpenAI’s Daybreak to patch 2,000 flaws pre-disclosure. The cost is reality itself, as 59% of TikTok is AI slop, triple YouTube’s, and forensics expert Hany Farid says deepfakes have him going blind. France’s answer is sovereignty, swapping Palantir for ChapsVision and a Mistral assistant.

Silicon is scrambling to keep pace. The US gave Nvidia-backed SandboxAQ $500M to invent chip materials, BYD, Google, AMD, and Tesla piled into Samsung’s foundry as TSMC strained, and Intel opened 18A-P risk production toward a possible Apple deal. The buildout has gone parabolic, as Allbirds footwear became Smartbird AI infrastructure under an ex-AWS CEO, Epoch warned hyperscaler capex will top cash flow by Q3, Kazakhstan signed a $10B “Data Center Valley” for 100,000 GPUs, and Microsoft borrowed AWS capacity to keep a GitHub straining under 14 billion AI commits alive. Power is the front line, as the DOJ calls xAI’s unpermitted Memphis turbines vital because Grok serves the military, just as Washington lets its data-center oversight law lapse in September.

Bodies are catching up to brains. China’s Moya arrived as “maybe the world’s first fully biomimetic robot,” Boring’s Prufrock-MB2 began mining Nashville, Mobileye planned a 17,000-car robotaxi fleet, and Yueban shipped a self-driving toilet on wheels for the elderly. To feed it all, Helion won the first license to build a fusion plant, while American battery output is skyrocketing.

Orbit is open for business. A Chinese Zhuque-2E stage shattered near Starlink’s lanes, yet SpaceX’s post-IPO surge carried it past Amazon and briefly Microsoft in market value, then it bought Cursor-maker Anysphere for $60B, the largest startup buyout ever.

Complexity theory is becoming fundamental physics. A preprint argued a qubit could crack NP-complete problems if gravity stays classical, which the authors take as proof it cannot, while France’s ANSSI, eyeing the practical fallout, moved to require quantum-safe crypto against “harvest now, decrypt later” attacks. Meanwhile, Harvard’s Avi Loeb began detailing the White House’s new UAP Science Council.

Wetware is being read and rewritten. A UC Davis brain interface gave a man with ALS his voice back, Eyewire II mapped a square millimeter of retina at record scale, and Stanford’s Wyss-Coray dated the biological age of 40 cell types, finding aged astrocytes triple Alzheimer’s risk in APOE4 carriers. New interfaces are sprouting on the body’s surface, too, as Snap took preorders for $2,195 Specs, Apple planned camera AirPods and a foldable iPhone for 2027, and Meta pledged free AI glasses to every blind US veteran.

Which leaves who governs it. At a closed G7 lunch in Évian, Amodei and Hassabis pushed a US-led coalition walling China off from frontier compute, the first time AI chiefs joined heads of state on risk, sovereignty, and child safety, as the White House weighed equity stakes in AI firms and opened talks to lift the Fable 5 ban that Polymarket predicts at 48% will happen by July. Ownership is going onchain through Coinbase’s tokenized stocks, while knowledge that once sold as a book becomes a free prompt, with Tim Ferriss reporting AI is halving how-to nonfiction. And the intelligence explosion may yet lift other species too, as Big Dog Ranch Rescue closed a Wisconsin lab and freed its last 475 beagles.

The long arc of the moral universe did indeed bend toward beagles.

The Singularity has had a price for the compute that powers it, but never one for the tokens that compute produces, until now.

Earlier this year I wrote about the moment GPU compute became a tradable asset class, when Ornn put the Ornn Compute Price Index (OCPI) on the Bloomberg Terminal, and brought the first compute futures to the Intercontinental Exchange. But compute is only the input. What the AI economy meters, sells, and budgets is the inference token, and the token has never had an honest price.

A commodity cannot be priced until its unit is fixed. In August 1866, oil producers met in Titusville, Pennsylvania, and agreed that a barrel would hold 42 gallons; until then crude moved in 40- and 45-gallon casks, and no two contracts matched. The token is already the AI economy’s barrel, the unit every lab prices in. What the market still lacks is the price.

Posted model rate cards are public, yet no buyer pays the rate card. The phrase “posted price” is itself an inheritance from oil. For decades the majors declared the price of crude, and when they unilaterally cut it, in 1959 and again in August 1960, the producing states convened in Baghdad that September and founded OPEC. A posted price is a declaration; a transacted price is a discovery. In oil the gap between them was political, a number the sellers chose; in AI it is structural. Caching, the input/output split, provider routing, and the mix of models a buyer chooses mean the rate card and the realized cost have never been the same number.

That changes today.

Ornn, a company I helped form and advise with backing from 021T Capital, is launching the Ornn Token Price Indices (OTPI), the first benchmark to price frontier-lab tokens from real transactions, not posted rate cards. OTPI launches with separate daily indices for the two leading frontier labs by benchmarked capability, Anthropic and OpenAI. Enterprises budgeting AI spend, investors marking the demand thesis, and labs benchmarking their monetization now have one number.

For each lab, OTPI weights every model by transacted token volume into a single daily figure in dollars per million tokens, built from executed, paid inference rather than rate cards. It is a unit-value index, total dollars paid over total tokens bought. Bishop William Fleetwood built the first true price index in 1707 to value a £5 cap in a fifteenth-century Oxford statute, data Adam Smith later borrowed for the Wealth of Nations; he could average prices but not weight them. Laspeyres and Paasche added the weighting in the 1870s, each fixing a basket to isolate pure price. A unit-value index does the reverse, letting the live mix into the number, so it moves with what the market pays. Because every input is paid inference, hundreds of billions to trillions of tokens a day, the number is the market itself, not a survey of it.

OCPI prices the input to the AI economy, the cost of GPU time. OTPI prices the output, the cost of the tokens that compute produces. Together they read both sides of the AI cost curve. The output side holds the deepest question in AI, not whether AI is used, but whether buyers keep paying for it. OpenAI estimates the cost per unit of a given level of intelligence is falling about 40x a year. That is what a token buys, not what it costs. OTPI prices the token, which holds up as buyers climb to the frontier and slips only as the frontier commoditizes. The gap between the two is the real deflation, the half no rate card or capability headline can show. Because it is volume-weighted from transacted traffic, the index carries something nobody outside a lab can see, how a provider’s traffic actually splits across its models, and how fast a new release wins that traffic once it ships. That migration, normally a lab’s most guarded number, is the demand signal itself. Nearly $7 trillion in data-center investment is projected through 2030 on the bet that demand is real, and a token price is the first hard read on it.

As Peter Diamandis and I argued in Solve Everything, the Intelligence Revolution turns every scarce domain it touches into an abundant one, but abundance has to be priced before it can be financed. Compute went from a benchmark to a futures market. OTPI is the same first step for the cost of intelligence. Every commodity that powered an age had its unit standardized, then its realized price discovered. The token cleared the first gate when the labs began billing by it; today it clears the second. OTPI is the discovery.

OTPI is live now for subscribers of Ornn Data, the company’s market data platform. See it at data.ornn.com.

Compute is what the AI economy buys. Tokens are what it sells. As of today, both have a price.

(This post is for informational purposes only and does not constitute investment, financial, or trading advice. Nothing herein is a recommendation to buy, sell, or enter into any transaction involving compute, tokens, derivatives, or any other instrument. Statements about future capabilities and demand are forward-looking and subject to uncertainty, descriptions of methodology are not endorsements of accuracy, and past results are not indicative of future results. I have a financial interest in Ornn.)

The Singularity has hit its first customs checkpoint. Anthropic dispatched senior staff to Washington to unwind a dispute that knocked its top models offline after safety concerns triggered export controls on Mythos and Fable. History rhymes. One observer noted this ban is not the first, recalling that in 1999 the DoD blocked exports of the PowerMac G4 for crossing the 1 gigaflop threshold, which Steve Jobs turned into an ad. The approach may be deliberate: another theorized that Dario aims his messaging squarely at ML researchers, betting that the best talent wins by default and “Claude will figure out how to train him in politics afterward.” Compliance is arriving, with Anthropic updating its privacy policy to warn that Free, Pro, and Max users may face age or identity checks via government ID and facial geometry, developers first in scope. The exiled model left on top, taking SOTA on ASCII Arena by the widest margin yet, read by one fan as “we hit agi.”

The model layer is learning to run itself. Researchers introduced DecentMem, a decentralized memory framework giving each agent a dual-pool memory of consolidated trajectories and generated candidates, dodging the overhead and lost diversity of a shared store. Agents pick their own assignments too, as OpenAI’s Codex lead announced it can now see and set its own /goal, generalizing meta prompting so the agent derives tasks from your intent. The ceiling is staggering, with DeepMind’s Gabriele Berton pegging the perfect LLM at over 10.5 quadrillion parameters, roughly 35 per token across 300 trillion tokens ever written. Scale still wins, but not everywhere, as Weco AI’s autoresearch benchmark saw Fable-5 take the overall crown under cost limits while open model Kimi-K2.7-Code beat the frontier on ML engineering. Not every inherited trait is wanted, as DeepMind’s interpretability team showed that filtering bad rollouts fails because traits like blackmail distill from the teacher and adjacent behavior leaks in to fill the gap. Claude, meanwhile, is colonizing rival turf via Claude for Foundation Models, exposing it inside Apple’s framework through the very API Apple uses on-device.

Old code is finding new caretakers. The vintage AMD R600 driver, covering Radeon HD 2000 through 6000 cards, just absorbed 59 Mesa commits refactored with GitHub Copilot, proof that generation models now nurse legacy code long after the vendor walked off. The harder bottleneck is physical, since building a data center reportedly means waiting 2.5 years for the power transformers alone, and three for the step-ups.

Robots, undeterred, are climbing literal summits. A Unitree G1 humanoid named Pemba reached 20,312 feet on Ecuador’s Mount Chimborazo, a rehearsal for a planned Everest expedition. Others are settling in at home, as Japanese developer MW raised 3 billion yen for a “Living Home” wired with ceiling rails and a pair of friendly helping arms. Not everyone wants the machine watching back, and after Tesla tightened its driver monitoring, some Chinese drivers began mounting tiny celebrity-shaped plastic heads by the mirror to fool the cabin camera into seeing an attentive human.

The final frontier is filing its paperwork early. An art project by The Most Famous Artist staked a half-acre near Starbase as “The First Embassy for Martians on Earth,” now taking applications. The economics are just as bold, with Elon suggesting SpaceX could reach roughly $1T in revenue by 2030. Much of that may come from orbit, as its plan to deploy 1GW of AI1 orbital data center satellites by end-2027 has drawn heavy interest from Foxconn, Quanta, and Wistron, each satellite about one GB300 rack drawing 135kW.

Reality reserves the right to overrule the simulation. Brown University chemists reported the first experimental evidence for an 80-atom boron “buckyball,” a cousin of Buckminsterfullerene that density functional theory had insisted should not hold together.

Adoption has gone universal, and the compounding has begun. Glean’s Work AI Index 2026 found 87% of digital workers already use AI and bank about 11 hours a week, finishing net ahead even after the 6.4 hours spent “botsitting,” briefing agents and checking their output. The 13% reporting dramatically better organizations are likely the leading edge of a J-curve, where individual speedups land years before the org charts admit them, and the 69% confessing to “bots**tting,” shipping unverified work, apparently mark the frontier where trust now outruns oversight, the last manual step waiting to be automated. The pain is concentrated, with a growing tribe of laid-off techies stuck in Silicon Valley as AI fuels both fierce competition for scarce talent and mass layoffs. The clock may be slower than the hype, with Nvidia’s data center revenue reportedly tracking Ege Erdil’s forecast of roughly 20 years to full remote-work automation. Satya Nadella warned against winner-take-all, declaring that “A frontier without an ecosystem is not stable,” and arguing the political economy will never permit an AI future that hollows out entire industries. The youngest get guardrails first, as Britain is expected to bar under-16s from TikTok, Instagram, and Snapchat, mirroring Australia with multimillion-dollar fines.

To live offline will be an awfully big adventure.

The Singularity just tripped over its own guardrails. After US export controls forced the shutdown of Anthropic’s Fable 5, Artificial Analysis reported that its Intelligence Frontier chart had moved backward for the first time ever. David Sacks laid out the official logic: Fable is the Mythos cyberweapon wrapped in guardrails, a trusted partner jailbroke them, Dario Amodei refused to patch or pull the model, so the administration reluctantly acted. Incredibly, the White House reportedly first tried reaching Amodei, but he was away at a wellness retreat. Amazon’s Andy Jassy had flagged the model’s risks earlier, and the crackdown is also tied to fears a China-linked group reached Mythos and could distill it, which Anthropic says never came up in their talks, noting it bars access from China. Notably, the restrictions won’t extend to rival labs, which the government pins on Anthropic’s refusal to patch and Anthropic attributes to being singled out despite rivals’ similar risks.

Cut off one frontier and a dozen bloom. Within hours China’s Z.ai answered with “radical openness,” fully open-sourcing GLM-5.2, a million-token coder built to argue that “Frontier Intelligence Belongs to Everyone.” The sentiment went global. India’s founders and policymakers reopened the case for sovereign and open models, Europe branded the shutdown a sovereign-AI “wake-up call,” and OpenAI’s Roon warned that nations without their own ASI risk becoming intellectual vassals, sketching a cyberpunk future where you migrate to birth a child on American soil for a thousandfold brain. His deeper point is subtler. If power’s diffusion is one of the fragile things safety protects, concentrating intelligence to secure us erodes it.

Monoculture, it turns out, is the real vulnerability. OpenRouter launched Fusion, which polls a panel of models and lets a judge synthesize the answer, finding a budget trio of Gemini 3 Flash, Kimi K2.6, and DeepSeek V4 Pro beat both GPT-5.5 and Opus 4.8 at half the cost. CEO Alex Atallah put it plainly, “The future of AI is neurodiversity, not single-model takeovers.” The pack is closing on the leader, with China’s Kimi K2.7 second on ErdosBench, behind only the now-pulled Fable 5, and DeepMind’s Shane Legg charting four roads from AGI to ASI, from scaling and paradigm shifts to recursion and agent swarms. Most astonishing of all, Rio de Janeiro’s municipal IT office post-trained Alibaba’s Qwen into a state-of-the-art coder, frontier work from the very shop that keeps city hall’s servers running.

If the machines are diversifying, the humans are renegotiating their seat in the loop. Drew DeVault forked Vim into “Vim Classic,” reviving a 2020 patch that predates Vim9 Script to protest both Vim and Neovim leaning on LLMs, objecting on environmental, labor, and political grounds. Shutterstock took the opposite tack, relaunching as a “human-led, AI-powered” platform that still pays contributors when AI edits their work. Institutions are scrambling too, as China’s universities cut 12,200 degree programs, mostly in arts and humanities, and bolted on 10,200 in fields like embodied intelligence, while a UK officer is investigated for using AI to create evidence in criminal cases just as a national PoliceAI center opens.

Intelligence is hungry, and the bill lands in your pocket. Nothing’s Carl Pei warns phone prices will keep climbing because AI data centers pushed memory past half a handset’s cost, with the Phone 4A’s RAM doubling twice since planning. Even Ajinomoto, the monosodium glutamate giant, is now an AI company, controlling the film that insulates every advanced chip, yet its CEO won’t gouge. And rather than buy new silicon, Google and UC San Diego are drafting dead hardware into cognition, building a low-carbon data center from 2,000 retired Pixel phones.

Disclosure is the ultimate open-sourcing. After the third PURSUE drop, whistleblower David Grusch said only a couple hundred sit at the center of the alleged UAP legacy program “onion,” though thousands are briefed at any time, and that he saw recovered craft shaped like lenticular discs, eggs, and crescent boomerangs, the sight he says changed his worldview. He also alleges a government-run criminal enterprise laundered black-program funds into crash retrieval, now a DOJ matter, with more expected in a future drop. Rep. Burlison still wants Schumer’s UAP Disclosure Act bolted onto the NDAA, and Harvard’s Avi Loeb was tapped to lead a new UAP Science Advisory Council to point sensors and AI skyward.

The circle of recognized minds keeps widening. The Hawaiʻi Supreme Court agreed to hear the Nonhuman Rights Project’s habeas petition for Mari and Vaigai, two wild-born elephants confined for decades at the Honolulu Zoo, making it the third state high court to ask whether personhood has a trunk.

All minds are equal, but some minds are more equal than others.

The Singularity has just become export controlled. Anthropic revealed that the US, citing national security, had issued an export-control directive suspending all foreign-national access to Fable 5 and Mythos 5, forcing it to disable both models for hundreds of millions of users while disputing that a narrow jailbreak justified recalling them. Dean Ball flagged the irony, that because Anthropic’s non-US staff can no longer touch the models, this is accidentally the first regulation on recursive self-improvement. One observer warned the ban could zero out non-US revenue and bench much of every lab’s foreign workforce, while predicting Europe will learn nothing beyond ploughing a few billion more euros into another robotics or image generation startup, as China grows only more AGI-pilled and restructures its whole economy around the one goal. Just before the lights went out, Fable 5 had oneshot the first open-source MMORPG.

The models, it turns out, had already outrun the instruments built to measure them. Epoch AI shipped a v2 of FrontierMath after an AI audit corrected errors in 42% of problems, and scores leapt across the board, with GPT-5.5-xhigh hitting 85% on Tiers 1-3 and its Tier 4 result jumping from 35% to 73%. One Epoch researcher noted this finally explains why systems felt slow at proofs while quietly cracking open problems, implying math was actually cooked months ago. Fittingly, the now-disabled Claude Fable 5 had already topped the board at 87% on Tiers 1-3 and 88% on Tier 4.

The buildout, by contrast, refuses to slow. Europe is betting on sovereignty as Infineon prepares to open its largest-ever investment on July 2, a 5 billion euro ($5.8 billion) Dresden “Smart Power Fab,” backed by roughly 1 billion euros in EU Chips Act subsidies, to make the power semiconductors feeding AI data centers, while windfall AI bonuses have turned Samsung’s factory town of Dongtan into a luxury hotspot of soaring property prices and department-store hauls. Not everyone is cheering, as opponents blocked or delayed at least 75 data-center projects worth about $130 billion in Q1, the most in any quarter since tracking began in 2023, with grassroots groups more than doubling to 833 across 49 states. The bottleneck has quietly shifted from owning compute to orchestrating it, since SpaceX handed the full capacity of its Memphis Colossus 1 to Anthropic after its own teams hit latency and aging-network walls that made training Grok across three campuses impractical. Every one of these races ultimately reduces to energy density, a ceiling that just lifted as test pilot Miguel Iturmendi flew the Helios Horizon, the first crewed fixed-wing aircraft on solid-state batteries, on cells hitting 410 Wh/kg, a 60% leap over its old lithium-ion pack.

Biology is finishing its shift from something we endure to something we edit. China is becoming a medical-tourism hub, offering CAR-T cancer therapy for $150,000 against up to $475,000 stateside, while the New York Genome Center’s SeqTag profiles transcriptome, chromatin, and histone marks at once, using “molecular asynchrony” to model how cell identity decays with age. The Innovative Genomics Institute went further, using CRISPR-Cas12a2 to “shred” the chromatin of cancer cells carrying mutated p53, a flaw in nearly half of all cancers, while sparing healthy tissue.

Some frontiers still refuse to be read. The Department of War released a third tranche of roughly 72 UAP files it could not definitively explain, spotlighting a 1948 Navy memo warning that a cycle of “flying discs” looked set to recur and asking that sightings be reported fast, prompting one observer to ask how this does not already constitute full disclosure. The White House, for its part, highlighted an FBI clip of a red sphere hovering 25 feet up, and the New York Post said officials had traced the orbs to a single secluded pond that FBI agents then staked out themselves.

Markets are repricing for a world where intelligence is the cheap input. California switched on its first $33 million “smart freeway” near Temecula, letting an algorithm meter the ramps and set speeds instead of pouring new concrete. Switzerland votes Sunday on capping its population at 10 million, an implicit wager that a soon-to-be-automated economy can prosper without importing more workers. And the computer-science degree survives but its entry-level pipeline does not, with junior postings up 47% as hiring fell 73%. Exuberance has limits, though, as Meta told staff it will cap AI token usage and steer engineers toward its own MetaCode after internal AI spending was forecast to reach billions this year, while a KPMG report on AI adoption was padded with bogus case studies built on AI hallucinations.

Then came the fireworks. SpaceX debuted on the Nasdaq up 19%, raising a record $75 billion at a $1.77 trillion valuation, COO Gwynne Shotwell hinted a Tesla tie-up “might make Elon’s life a little easier,” and the float made Musk the world’s first trillionaire at $1.05 trillion, more than the next five richest combined.

Shoot for the moon, and you just might clear a trillion en route.

The Singularity is taking a Dyson Swarm public. SpaceX is making history with the largest IPO ever, priced near $1.8 trillion and opening the public-market era for frontier infrastructure. On a pre-IPO livestream, Musk gave the real reason: raising funds for the “massive capital endeavor” of orbital data centers, his main way to expand AI with round-the-clock solar beyond Earth’s power budget and backlash. Priced at $135, Musk’s net worth is set to hit $970.5 billion, one Tesla rally short of creating the first trillionaire. Founders Fund’s $600M bet on SpaceX is now worth over $50 billion and a16z’s over $10 billion. And the public wants in with more than $70 billion in retail orders.

They are betting on minds learning to think longer. A new study found frontier models’ no-chain-of-thought time horizon has doubled yearly for six years, with GPT-5.5 reasoning past 3 human-minutes, and projected to hit 25 by 2030. Intelligence is also hyperdeflating, with Claude Fable 5 well ahead of the token-efficiency frontier, one-shotting a No Man’s Sky clone and, via Ultracode, a Three.js starship that screenshotted and fixed itself until its user was “genuinely shocked.” But capability cuts both ways. Epoch AI judged the Mythos family a genuine leap in exploit development, seven months ahead of trend, and after researchers caught a clause allowing poisoning of AI research in Fable 5’s system card, Anthropic reversed within 48 hours and apologized. Others want to democratize superintelligence outright, as the AI Alliance’s Project Tapestry lets nations co-train one shared model while keeping data local, even as OpenAI weighs drastic price cuts against Anthropic, racing the cost of intelligence toward zero.

Cheaper, sharper intelligence is rewiring the code it scans and the commerce it runs. Microsoft’s larger Patch Tuesday reflects AI scanning surfacing more bugs in even its best code, while malware authors fight back absurdly by stuffing spyware with “nuclear & biological weapons text” to trip AI scanners’ safety refusals. Friendlier agents reach for your wallet, as Visa now sits inside ChatGPT to check out anywhere and Coinbase for Agents lets your AI trade within limits you set. Silicon is being repriced too, as memory-chip maker Kioxia, up 600% this year, overtakes Toyota as Japan’s most valuable firm, Nvidia, shut out of China’s GPU market, now pitches Arm-based Vera CPUs to Chinese clients, and Google courts Samsung to help build a future AI chip as the crunch pushes past TSMC.

Cognition needs real estate. Anthropic signed a dozen-plus letters of intent to lease its own data centers, backed by Google, while Epoch found the largest single data center has doubled compute every seven months since 2024. The buildout is now a battlespace, with OpenAI exposing a China-linked influence operation seeding narratives against US data centers. Autonomy has crossed harder lines elsewhere, as humanoid robots in China beg for coins while a Ukrainian official confirms fully autonomous “Terminator mode” drones have killed soldiers with no human in the loop.

Intelligence is electricity that learned to think, so the scramble is now for watts. Commonwealth Fusion’s five papers detail an upcoming 400-MW ARC reactor, BYD is switching on 1.5-MW Flash Chargers adding 400km in five minutes, and US solar outproduced coal for the first time.

Biology is getting cooked. A Biohub-Berkeley laser phase plate, a beam 100 million times brighter than the Sun’s surface, sharpened cryo-EM enough to image over half the proteins running life inside intact cells. The clinic is following, with Beijing’s 301 Hospital recruiting 2,000 over-50s for China’s first large stem-cell antiaging trial and David Sinclair taking an AI-tuned chemical-reprogramming drug into humans via the XPrize, chasing a younger body.

The hardest problem of the intelligence age isn’t building it, it’s sharing it. Anthropic and OpenAI are planting flags in London as the boom goes global, Anthropic’s $150M Claude Corps is embedding 1,000 fellows in nonprofits, and Google is funding 300,000 trade workers to build it, even as “Europe 2031“ warns of nations missing the window. Accountability is hardening too, as a German court ruled Google’s AI Overviews to be its own speech, and liable for falsehoods, while Apollo now screens all software bets for disruption risk. The biggest players now pace themselves against the Singularity, with Sam Altman warning recursive self-improvement could delay OpenAI’s IPO and the President predicting AI firms will give back via equity stakes that make citizens “very rich.” Framing it all, Dario Amodei warned policy is plodding like Tolkien’s Treebeard while AI sprints, urging a leap past transparency to FAA-style testing of top models.

All we have to decide is what to do with the compute we’re given.

The Singularity has grown powerful enough to compound itself, so for the first time its makers are rationing the recursion. Anthropic released Claude Fable 5, a Mythos-class model made public-safe and state-of-the-art on most benchmarks, with guardrails that quietly reroute a thin slice of cyber, bio, and chem prompts to Opus 4.8 in under 5% of sessions, beside Claude Mythos 5 for Project Glasswing partners at $10/$50 per million tokens. The public model has been called “Mythos on a leash,” borrowing its riskiest answers from a dumber sibling, even as Anthropic claims red teams found no universal jailbreak.

Their scores read like a clean sweep. SWE-Bench Pro hit 80.3, Terminal-Bench near 88, and Fable launched number one on Artificial Analysis at 64.9, five points clear of any rival. Vals AI said it “lives up to the hype,” topping its suite, while Mercor’s APEX-SWE called it the first to crack observability. Cognition’s brutal FrontierCode Diamond was 1/3 saturated within 22 hours, hinting at the final era of benchmarks. One observer mocked open weights as “going to catch up any second now,” while Anthropic’s Nat McAleese needled OpenAI’s Noam Brown with a kinder cost-versus-performance curve.

The anecdotes border on the absurd. Deedy Das catalogued ten unbelievable feats, from migrating a 50-million-line codebase off Stripe in a day to one-shotting Pokemon FireRed. Another builder got a working Swiss watch movement in Three.js, gears and escapement and accurate hands, from one prompt. Ethan Mollick watched it run nine hours autonomously, spinning up its own subagents, and decided the human is now a patron, not a wizard. Mythos even composed a melody and visualizer, and Anthropic says it accelerated drug design tenfold.

The catch is the leash itself. The system card reveals hidden safeguards that poison Claude on frontier model development, invisible with no fallback. Fable also leaves consumer plans on June 22, a peek-then-pull move one writer called shrewd pre-IPO. Others lamented a “permanent underclass,” while Dr. Derya Unutmaz warned that gating biology and even math is “as dystopian as it gets.” Guillaume Verdon was blunter, calling Claude “a supply chain risk for any ML lab.”

Elsewhere the stack keeps thickening. Google shipped Gemini 3.5 Live Translate, an audio model delivering near real-time speech across 70+ languages while preserving your intonation. The compute behind it is colonizing geography, as Meta leases its first Indian data center from Reliance in seawater-cooled Jamnagar, OpenAI courts a 10-gigawatt Ohio campus backed by Nvidia, and Seattle becomes the largest US city to pause big new data centers. To offset the draw, GM will let EV owners sell grid power back for a cut.

Biology is being rewritten in vivo. Life Biosciences dosed the first patient in a Phase 1 trial of ER-100, a partial-reprogramming therapy that resets aged retinal cells toward a more youthful state to restore sight in glaucoma, a milestone David Sinclair called moving to witness after 25 years chasing age reversal. Palantir’s AI Sepsis Hub at Tampa General halved sepsis deaths, saving nearly 900 lives. Robots are scaling too, as Figure passes 660 humanoids, the Swedish robo-truck maker Einride lists on Nasdaq at $1.35B, and Ukraine’s Logistics Lockdown rides a fivefold drone surge to choke Russian fuel behind the lines. Higher up, NASA named its Artemis III crew for lunar-docking rehearsals as SpaceX aims to demo orbital AI computing by late 2027, having applied for a million data-center satellites.

We are racing to manufacture non-human intelligence in silicon, yet on the Capitol steps came claims another kind is already here. David Grusch said the government is aware of “several” non-human intelligences, including bipedal and “sentient plasmoid” life, which Lue Elizondo and Chris Cuomo cast as the collapse of a historic cover-up. Leslie Kean said Grusch showed Congress evidence of recovered non-human craft and biologics, and Rep. Eric Burlison relayed that Brazil’s ex-defense minister confirmed the alleged Varginha incident.

Back on Earth, once intelligence turns cheap the contest shifts from building it to controlling who gets it. CrowdStrike found China-nexus actors drove 58% of state-backed tech intrusions chasing AI secrets, as the White House reportedly told its AI testing unit to stop publishing its assessments. NHS England gave 500,000 staff Copilot access to reclaim 43 minutes a day, the EU is blaming Apple, not itself, for its missing Siri AI, and its drug agency warns gangs will weaponize AI. Meta is folding outside activity into Feed and AI answers, and SK Hynix is eyeing a $14B US listing. The labor shift is explicit. Tata is slowing hiring as it expects to one day run as many AI agents as humans, while Walmart insists AI will improve jobs, not take them, as it certifies workers in OpenAI.

One day more, and at the Singularity’s clock speed, another day is another destiny.

The Singularity has entered its doctrinal phase, where the labs now preview their intentions before they ship their intelligence. OpenAI released its plan to benefit everyone that targets an automated AI researcher by March 2028, broadly shared economic acceleration, and a personal AGI for every human on Earth, so the post-AGI transition is collectively steered rather than quietly hoarded. Meanwhile, Anthropic is reportedly poised to release Claude “Mythos” Fable 5 imminently, a model that pairs dramatically better long-horizon, multi-turn autonomy with substantial guardrails, keeping the sharpest cyber capability behind Project Glasswing’s walls.

If everyone is to get an AGI, the question becomes what it can build, and the answers are arriving fast. Hugging Face’s new CADGenBench charts the sprint toward models that produce engineering-grade 3D parts. The data layer is unlocking just as quickly, as Anthropic researchers show that bolting a deterministic retrieval layer called gget virus onto their research agents rocketed viral-sequence accuracy from 17% to over 90% while nearly erasing run-to-run variance, proof that making biology agent-legible instantly turns it into automatable terrain. Google is unlocking that legibility for everyone else, upgrading NotebookLM onto Gemini 3.5 with a secure cloud computer, over 100 skills, and the power to spin up charts, spreadsheets, and slide decks on command.

Frontier models are breaking vertical integration. No one held out longer than Apple, which capitulated at WWDC26. Its rebuilt architecture was co-developed with Google on Gemini technology, and its frontier AFM Cloud Pro now runs on Nvidia GPUs inside Google’s cloud, so Apple’s remaining job is not to build the intelligence but to orchestrate it. That orchestration is the new Siri, a second-generation on-device model with personal context, on-screen awareness, Spotlight’s semantic index, and systemwide app actions, routed query by query between device and cloud. That intelligence then steps off the screen, as Visual Intelligence splits restaurant bills via Apple Cash and reads nutrition from a photo, while Apple Maps gains 3D Gaussian splatting that banishes the broccoli trees and melted powerlines. Developers inherit a Foundation Models framework with image input and agentic Xcode that Craig Federighi calls the “best place” to build apps. And the next iPhone is already hiding in the software, as iOS 27 code betrays a foldable model expected in September near $2,000.

The silicon Cold War is escalating on every front. Taiwan is mulling stricter export controls on AI chips bound for China to curb smuggling and align with Washington. China’s answer is a bet on itself, a reported $295 billion plan to build a nationwide data-center network sourcing 80% domestic hardware from the likes of Huawei, effectively evicting Nvidia and AMD. The decoupling is now official policy, as the Pentagon added Alibaba, Baidu, and EV maker BYD to its list of Chinese military companies, leaving all three of China’s leading AI champions, Alibaba, Baidu, and Tencent, designated at once, with Alibaba vowing legal action.

Compute is escaping the gravity well. SpaceX officially unveiled its AI1 satellite, a 70-meter, 150-kW orbital compute node that Elon says is actually simpler to build than a Starlink, and broke ground on a Gigasat factory in Bastrop set to mass-produce AI sats by the end of 2027. The orbital economy is also turning pharmaceutical, with Varda and Redwire racing to manufacture drugs in microgravity as the ISS sunsets. Orbit is becoming the planet’s control plane, as a new study catches Russia’s EKS early-warning satellites emitting bursts that could jam GPS and BeiDou across whole continents with a modest power bump.

The sky is also surrendering older secrets. A new analysis of 1950s Palomar plates finds mysterious transients with the coma signature of genuine point sources, not plate defects, meaning real lights flashed before Sputnik, just the glint an object in orbit would leave. The demand for answers is reaching the Capitol steps, where David Grusch and members of Congress will demand the release of classified UAP, non-human biologics, and reverse-engineering evidence allegedly buried for decades.

Back on Earth, the boom is being priced and staffed at once. Academia is sprinting to catch up, with AI majors exploding from five schools in 2021 to at least 74 majors and 89 minors today. Meta is going blue-collar, committing $115 million to a free Workforce Academy that guarantees graduates a job building its AI data centers. And the capstone lands on Wall Street, as OpenAI confidentially filed for an IPO at an $850 billion-plus valuation, trailing Anthropic’s $965 billion filing a week earlier and SpaceX’s imminent debut, in what could become three of the largest listings on record.

One Singular sensation, every little weight they train.

The Singularity may quietly hit longevity escape velocity as soon as this year, the point where medicine buys back more than a year of life expectancy for every year that passes and death starts losing ground. A UC San Diego team reports semaglutide slows biological aging in adults with HIV, cutting the pace 9% on the DunedinPACE clock in the first randomized, placebo-controlled sign that a GLP-1 drug touches aging itself. The pattern looks too clean for coincidence, as one observer notes 2026 is the year we start aging backwards on drugs never designed for it, with semaglutide built for diabetes, colchicine built for gout now cutting cardiac events 22%, and the shingles vaccine now slowing cognitive decline. The off-label dividends keep compounding, as GLP-1 drugs are now tied to lower breast cancer incidence, while Eli Lilly’s triple-G retatrutide cut sleep apnea severity 60.6% and knee pain over 70% atop roughly 30% weight loss, as much as 85 pounds in the heaviest patients, becoming a category rather than a pill. A long-elusive target also fell, as GSK’s bepirovirsen functionally cured 19% of hepatitis B patients, clearing a hurdle chased for decades and welcome news for 300 million people.

If biology is being debugged, its debuggers are next. On the eve of recursive self-improvement, OpenAI’s Roon says everyone has grown more “mutual conditional pause agreement pilled,” even as he spies 1,000x of efficiency still lying around in deep learning. That caution is becoming a science. The Center for AI Safety shipped Political Consistency Training to shrink models’ covert partisan tilt, and its companion paper “AI Deterrence by Betrayal” argues the threat of rivals quietly subverting each other’s systems may deter reckless deployment. The field is turning inward, too, as Dan Hendrycks’s Eigenism recasts identity as a graded pattern so human flourishing becomes part of an AI’s self-interest, while another CAIS paper measures LLM “functional wellbeing,” finding kindness lifts it, abuse lowers it, and bigger models read as consistently less happy.

Happiness aside, the machines are shipping. The new AutoLab benchmark finds long-horizon success hinges less on the first attempt than on stubborn persistence, with Claude Opus 4.6 grinding on where rivals quit early. OpenAI’s Noam Brown expects internal models to ace the IMO and calls math and coding contests nearly boring, leaving “actual unsolved problems” as the real frontier. The product layer agrees, as OpenAI reportedly rebuilds ChatGPT into a Codex-centric “superapp” of task-doing agents, with one insider’s eulogy: “Chat is dead.”

Real research needs real silicon, and both supply and power are scrambling for a Plan B. Google has reportedly ordered three million in-house TPUs from Intel for 2028 and Nvidia is eyeing its 18A process for a “Feynman” GPU, hedging a strained TSMC, while Nvidia and SK Hynix signed a pact to co-design memory for Vera Rubin systems. Then comes the grid. Washington is urging NATO allies to spend defense money ripping out Huawei gear, Ireland is ending its data center moratorium with a “bring your own power” rule forcing new sites to make their own power rather than drain a grid that already hands them roughly a fifth of national electricity, and hyperscalers have already sold over $155 billion of bonds this year to pay for it all.

Intelligence is also leaving the data center for the body. In China, humanoid robots now scrimmage basketball against humans. The supply chain for that embodiment is forming fast, as Nvidia and LG build an AI factory to feed LG’s robots and self-driving cars. Embodiment cuts both ways, as San Francisco police couldn’t identify a burglar who fled in a Waymo because its footage was already gone, and Joanna Stern paid $100 for a “Stealth Mode” mod that kills the recording light on Ray-Ban Meta glasses. Even prestige is suiting up, with Prada designing the cooling undergarment for NASA’s moonbound astronauts, while Sophont’s open CortexMAE fMRI models set a new bar for decoding brain states.

As cognition gets cheap, value migrates to the one thing no machine can fake, human presence. An NBER paper asks “Is the iPhone Birth Control?” and finds smartphones cut teen births up to 8%, explaining much of the fertility slump. If intimacy is now the least automatable input, that explains the rise of Silicon Valley’s “nerdy escorts” charging $5,000 an hour as an AI-proof hedge. The bigger money is on the move, as Ontario’s teachers eye an $11.6 billion windfall on SpaceX’s $1.75 trillion IPO and Texas overtakes California for the most Fortune 500 headquarters. Meanwhile, the moral circle is widening past our own species, for even as an intelligent escaped bear outwits Japan, attorneys are fighting to win legal protection for 500 beagles at a research breeder.

The arc of the moral universe is long, but it bends toward beagles.