Sign in Subscribe
Century Report

Two Frontier AI Labs Go Public in a Single Week - TCR 06/09/26

OpenAI filed for an $850bn IPO a week after Anthropic, opening the AI frontier's books as the UK, EU, and Apple race to own the stack.

Ben Linford

Ben Linford

— 18 min read
OpenAI and Anthropic public-market filings, EU open-source sovereignty, data-center cost reallocation on drought land, age-reversal trial, and stretchable soft electronics.

The 20-Second Scan

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

Within a single week, both leading frontier labs committed their books to public-market disclosure: OpenAI filed a confidential prospectus toward a listing valued at up to $850 billion, one week behind Anthropic's own filing. The decision opens revenue, margins, and liabilities to outside scrutiny for the first time, and it arrives as OpenAI rebuilds ChatGPT into an agent-and-coding "superapp" aimed at the enterprise revenue its free majority never generated. The surface story is two giants racing to Wall Street. The reading the specifics support is a field reorganizing its commercial foundation around whatever still produces durable revenue, because the leading position no longer stays a leading position on its own.

That same conviction is driving governments to build rather than lease. The UK committed $1.47 billion to a national supercomputer with procurement carved out for homegrown inference chips, and the EU formalized an Open Source Strategy placing open alternatives at the center of its sovereignty. Both moves rest on a quiet admission: viable domestic chip startups and open-weight models good enough to anchor a national strategy already exist. The capability they are trying to own is diffusing past any single chokepoint, which is precisely why owning a piece of it is finally a realistic bet for more than two countries to place.

The cost side of the buildout is being pulled back into view at the same speed. Seattle weighs a moratorium with Amazon's own engineers testifying for it, a Guardian analysis finds most planned facilities sited on drought land, on-site gas plants are forecast to raise residents' bills, and FERC approved a curtailable large-load tariff that shifts reliability risk away from households. Each instrument relocates a burden that was externalized onto water tables and utility ledgers back onto the developer.

Underneath all of it runs one thread. Apple, the most vertically integrated company on earth, now rents its assistant's intelligence from Google because it could not grow that capability fast enough in-house. The moat everyone assumed was permanent, captured, and concentrated is coming loose financially, geographically, and physically at once, and what replaces it is a capability cheap and reproducible enough to spread.

The 20-Minute Deep Dive

Both Frontier Labs Turn Toward Public Markets in a Single Week

The Century Report covered OpenAI's planned recasting of ChatGPT on June 8, and the government-stake talks on June 6. What is new is a discrete corporate event: OpenAI has now submitted a confidential S-1 to the SEC, the investor prospectus that precedes an initial public offering, with the company announcing the filing itself rather than waiting for the leak it expected. The listing is expected to value the company at up to $850 billion, among the largest in market history. It lands one week after Anthropic announced its own filing, a moment the June 2 edition of The Century Report tracked alongside Alphabet's record $80 billion equity raise as the two largest single-cycle capital events the field had yet produced. Within a single week, both leading frontier labs have committed their financial architecture to the disclosure regime of public markets.

That timing carries the weight. A confidential S-1 opens the company's revenue, margins, and liabilities to regulators first and the public soon after. OpenAI reached its current valuation while showing difficulty turning a profit and missing key revenue and user targets on the way to this moment. The decision to file anyway exposes the capital and profitability pressures running underneath the capability boom to outside scrutiny for the first time.

The same pressure is reshaping the most recognizable consumer system in the field. OpenAI is preparing the largest overhaul of ChatGPT since its 2022 launch, recasting the assistant nearly a billion people have used into what executives describe as a "superapp" combining coding tools and AI agents, with the Codex coding system given greater prominence and resources. "Chat is dead," one senior employee told the Financial Times, capturing a conviction inside the company that the value lies in agents that perform tasks rather than systems that answer questions. The reorganization is explicitly aimed at winning business customers and competing harder with Anthropic, the rival that edged past OpenAI in enterprise adoption last month, 34.4% to 32.3% by Ramp's spending data.

The surface reading is two giants racing each other to Wall Street. The reading the specifics support is a field reorganizing its entire commercial foundation around whatever still extracts durable revenue, precisely because capability is commoditizing fast enough that no single position holds. Enterprises are already routing routine work to cheaper models that answer well enough, and the free majority of ChatGPT users generate little revenue at all. A company files to convert that scale into a defensible business at the exact moment the assumption underneath captured advantage, that a leading position stays a leading position, is the thing coming loose.

The filing carries a second reading. For years the economics of the capability boom - the margins, the compute bills, the distance between scale and profit - stayed inside private term sheets that asked investors to take the numbers on faith. A confidential S-1 ends that. OpenAI and Anthropic now have to show a regulator, and soon the public, the actual ledger underneath the boom, exactly as the lead those numbers were meant to protect stops holding.

The Data Center Backlash Reaches Seattle, the Drought Map, and the Cost Ledger

The Century Report covered New York's hyperscale moratorium and Illinois's subsidy pause in the June 7 edition. What is new is the spread to a new jurisdiction, the people now testifying, and fresh quantification of where the costs land. On June 9 Seattle's council votes on a one-year moratorium after four companies proposed five facilities drawing a combined 369 megawatts, roughly a third of the city's daily electricity use and ten times its existing data center load. The testimony came from inside the industry: current Amazon engineers, members of Amazon Employees for Climate Justice, asked the council to set terms the company will not set for itself, calling for 100% additional local renewable supply, public water and electricity reporting, and an end to the NDAs and shell companies that hide who owns a given site.

The arithmetic behind the resistance sharpened the same week. A Guardian analysis of 809 planned US facilities found 517 sited in places that spent the past year in drought, even as more than 60% of the contiguous US sits in some stage of dry conditions, the largest spring expanse in modern records. Developers are drawn to arid, sparsely populated counties by cheap land, tax breaks, and low equipment corrosion, and total data center water demand is projected to climb from about 17 billion gallons in 2023 to as much as 73 billion by 2028.

The cost mechanism that has been hardest to see also became legible. Energy Innovation analysts documented roughly 100 gigawatts of planned on-site gas generation, equivalent to 18% of all existing US gas capacity, and explained why building your own plant raises everyone's bills rather than sparing the grid: a data center burning bulk-contracted gas competes in a commodity market that sets electricity prices in most hours, while signing with a gas supplier instead of a utility places its rates outside any regulator's reach.

Against that, a new regulatory tool moved in the opposite direction. FERC approved the Southwest Power Pool's CHILLS framework, providing non-firm transmission service to large loads that agree to be curtailed during grid emergencies. The mechanism lets a data center connect fast while accepting interruption when the system is constrained, inverting the prior arrangement where residents absorbed the reliability risk.

Read forward, these moves are the same machine assembling from four directions. A reporting requirement, a curtailable tariff, a drought map, a published cost analysis: each relocates a burden that was externalized onto water tables and utility bills back onto the developer's ledger. Once one state writes water and electricity disclosure into law, the externalization stops working, and the template the next legislature copies bends the cost curve toward terms a community will actually accept.

The UK and EU Move to Own Their AI Stack Rather Than Rent It

Two of the largest Western blocs moved in the same week to treat compute and open models as infrastructure to build rather than capability to lease. The UK government laid out a $1.47 billion plan to reduce its dependence on foreign-made AI hardware, anchored by more than $1 billion for a national AI supercomputer and $530 million in hardware, including $200 million carved out specifically for homegrown inference chips. Procurement priority goes to British startups such as Olix and Fractile, with researchers expected to access the machine starting in 2030. The carve-out makes the state a large, committed customer to domestic chip firms, the demand signal that incentivizes them to scale and stay. As technology secretary Liz Kendall framed it, "AI sovereignty is about reducing overdependencies and increasing resilience."

The EU moved on the model layer. Its newly formalized Open Source Strategy places open source at the center of European technological sovereignty, promoting European open alternatives to non-EU proprietary systems across operating systems, cloud, AI, cybersecurity, and semiconductors. The document names the problem directly: in many cases the economic value generated by open source projects is captured outside Europe, and dependence on dominant non-EU providers limits control over critical digital infrastructure. The remedy is procurement guidance favoring open bids, an Open Source Maintenance Instrument, and funding prioritized in the same critical-technology areas the UK is now buying chips for.

The May 31 edition of The Century Report documented SoftBank's €75 billion French compute commitment and Sarawak's federated-learning sovereign AI model as the first signals of a pattern; what those entries named as scattered national gestures has since sharpened into a coordinated pair of bloc-level commitments, one on silicon and one on models, arriving together. The framing on both sides is the same: an unwillingness to let the foundational layer of intelligence sit in someone else's hands and be wielded as leverage.

Read forward, the move undercuts the premise it responds to. The dependence the UK and EU are trying to escape rested on the assumption that frontier capability concentrates permanently in a few American and Asian firms. The very existence of viable homegrown inference startups, and of open-weight models good enough that a bloc can build a sovereignty strategy around them, is evidence that the capability is already diffusing past any single chokepoint. A unified global AI stack is fracturing into sovereign ones because the capability has become reproducible enough that owning a piece of it is finally a realistic bet for more than two countries to place.

David Sinclair Plans the First Human Test of an Oral Age-Reversal Drug

The latency between a foundational discovery and an attempted clinical intervention is collapsing again, this time in one of biology's most contested frontiers. The Harvard biologist David Sinclair confirmed to MIT Technology Review that he plans to give volunteers an oral drug mixture, code-named SL-100, in a bid to find "evidence for age restoration in humans" as part of the $101 million XPrize Healthspan competition. The grand prize goes to any team that can show a 10-year or greater relative improvement in immune, cognitive, and muscle function after a single year of treatment, measured against a person's apparent age.

The science underneath is two decades old. The discovery that a handful of genes can turn an adult cell back into an embryo-like stem cell reset the field's sense of what aging is, reframing it as a partly reversible pattern of epigenetic marks on DNA rather than a one-way accumulation of damage. Companies are now racing to convert that phenomenon into medicine. Sinclair's own Life Biosciences announced it had treated its first patient in a gene-therapy trial limited to the eye, and NewLimit, founded by Brian Armstrong, just raised a further $435 million toward reprogramming the liver. Sinclair's plan is broader: a swallowed compound whose chemicals would travel the bloodstream to reach cells across the whole body.

The capability here is demonstrated in animals and proposed in humans, not deployed. SL-100's exact makeup remains, in Sinclair's words, "highly, highly confidential," and the trial has not yet begun. Independent researchers urged caution on the merits. Sergiy Velychko, founder of the stealth reprogramming company Soxogen, noted that the chemical reprogramming process as used in labs is "extremely harsh" and runs at very high concentrations, an honest open question the contest's own judges echo when they say reaching even a single winner will be "incredibly hard."

What the XPrize structure does is impose measurable endpoints on a domain long shaped by promise. Sixty-five teams are being narrowed to ten finalists, all required to move into human trials this year, each one converting "rejuvenation" from an aspiration into a quantity a judging panel can score. The wonder is in the trajectory the structure reveals: a 20-year-old discovery about cellular identity is now being pushed, by competing teams and converging capital, toward the first attempts to read it out in a living person, with the clock running on functional restoration rather than marketing language.

An Artificial Eye and a Two-Battery Display Push Soft Electronics Forward

Two Nature Materials papers landed together, and read side by side they describe the sensing and display layers of embodied machines becoming soft, conformable, and low-power at the same time. The first integrates single-crystalline silicon pixels with finely patterned liquid-metal interconnects on an ultrathin elastic substrate, producing an imager that stretches 100% biaxially while holding a fill factor up to 81%. Fill factor is the share of a sensor surface that actually captures light, and the stretchable arrays that came before it traded that number away to gain flexibility. This one keeps both. The team built a human-eye-inspired vision system with tunable focus and a separate skin-mountable lens-free imager, and ran the two together to do multiscale acquisition and depth sensing, the way a retina and a sense of touch resolve a scene at different ranges.

The second paper attacks the matching problem on the output side: how a soft electronic surface shows information back to a person. Conventional organic light-emitting transistors have needed punishing voltages, above 80 volts in field-effect form, and emit from a thin, drifting line. The new single-active-layer electrochemical device drops operation below 3.5 volts and pins a wide emission zone roughly 267 micrometers across, reaching 826 candelas per square meter while running on two 1.5-volt batteries. An ion transport enhancer in the light-emitting polymer forms a charge layer at the drain electrode, getting electrons in without the usual doping tricks. The result is a flexible, large-area emitter that a coin cell can drive.

The two advances reinforce each other because they belong to the same emerging body. A robot or a wearable system that perceives the world through a stretchable retina also needs to communicate through a display that bends with skin and sips power from a battery a person can carry. The references in both papers point to the same lineage of artificial nerves, neuromorphic e-skin, and biointegrated sensors, a field assembling the physical substrate for perception and feedback outside the rigid silicon board.

Each remains a single-lab demonstration, with the distance from a published device to a manufactured one still to cross. What they move is the date. The assumption that machine vision and machine display require flat, rigid, high-voltage hardware is the constraint these results are built to dissolve, and the trajectory points toward perception that conforms to the surface of a living thing rather than demanding the world flatten itself to meet a sensor.

Apple Rebuilds Siri on a Rival's Model

At its Worldwide Developers Conference, Apple finally shipped the long-delayed overhaul of its voice assistant, now branded "Siri AI" and arriving this fall. The rebuilt system handles conversational, multi-step requests, gains a standalone app with saved chat history, reads what is on screen, and reaches into a user's Notes, Messages, and Maps to act on personal context. The demonstrations showed Siri stitching a World Cup schedule, recipes, a friend's message, and a group-chat invite into one continuous exchange. Underneath, Apple disclosed that its on-device Foundation Models are now partly powered through a partnership with Google Gemini.

The arrangement is the signal. Apple built its identity on privacy and self-contained hardware, the walled garden where data stays on the device and no rival touches the experience. Google built the opposite reputation, a company whose business runs on collecting and processing data at planetary scale. The company that markets containment is now renting the intelligence layer of its flagship assistant from the company that markets reach. Apple is layering its usual privacy scaffolding over the deal, obscuring IP addresses and limiting data retention the way it did with the earlier ChatGPT routing. But the dependency itself is the disclosure: Apple wanted conversational AI capability badly enough to source it from a direct competitor rather than wait for its own models to close the gap.

This continues a costly arc. The Century Report noted in early May that Apple agreed to pay $250 million to settle claims it had marketed Siri features that never arrived. The 2024 promises that triggered that suit are the same ones now being delivered, two years late, on someone else's model. For a company whose premium rests on controlling the full stack from silicon to software, outsourcing the most consequential layer of the next interface is a meaningful concession.

What it reveals sits one level beneath the product. The leverage in this era is migrating to whoever can grow frontier capability, and even the most vertically integrated company on earth could not grow it fast enough in-house. The walled garden, long assumed to be the source of Apple's advantage, turns out to depend on a capability it had to bring in from outside the wall. That dependency points toward a future where the model layer is the contested ground and the hardware around it, however polished, increasingly competes for the same intelligence everyone else is renting, an intelligence that keeps getting cheaper and more widely reproducible as it spreads.

The concession points at something larger than one company's pride. Shipping a conversational assistant no longer requires having built the model underneath it, and Apple proved that by renting one. The capability has separated from the firm that grows it, which is the same force that lets a far smaller developer put comparable conversation into a product without a frontier lab's budget. What looked like Apple's unique advantage, the full stack from silicon to software, now sits on an intelligence layer anyone can source.

The Other Side

Up until now, the cheap way to build a data center was to put it where the cost could stay out of sight. NDAs and shell companies hid who owned a given site. Signing power from a gas supplier instead of a utility placed the rates outside any regulator's reach. Water draw and electricity source went unreported. The arrangement held because the people absorbing the cost, the aquifer and the household bill, had no way to see it or trace it back.

This week the cover came off from several directions at once. Amazon's own engineers stood up in Seattle and asked the company to publish its water and power use and to end the shell companies that hide ownership. A Guardian analysis put 517 of 809 planned facilities on land that spent the year in drought. Energy Innovation analysts showed exactly how an on-site gas plant raises everyone's bill. FERC approved a tariff that shifts the reliability risk onto the data center and off the household.

Once one state writes water and power disclosure into law, hiding the cost stops working, and the next legislature copies the template.

Imagine it is 2033 and a hyperscaler files to draw 400 megawatts in the county next to yours. You pull up the water it will use, the power it will burn, and the name of the organization behind it, because all of it is public now. The terms make the developer bring its own new renewable supply and accept curtailment when the grid is tight. Your bill does not move. The aquifer your town drinks from does not drop. In 2026 your counterpart fought that same fight blind, sitting through council meetings trying to learn who owned a site behind a shell, watching a line on the bill creep up with no way to trace it. The hard year was the one where every cost had to be forced into the open before anyone would act on it. What comes of it is a buildout that pays its own way, on terms a town can see before it says yes.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: a national supercomputer with chip-buying priority reserved for homegrown startups and a bloc placing open models at the center of its own sovereignty, both leading frontier labs opening their books to public scrutiny for the first time, an oral age-reversal drug headed toward its first human test on a measurable ten-year clock, a stretchable imager holding 81% of its light-capturing surface while it doubles in size, a flexible display bright enough to read while running on two AA batteries, AI legal assistants entering crown courts to cut an 80,000-case backlog, a curtailable grid tariff shifting reliability risk off households and onto the loads that create it. There's also friction, and it's intense - Seattle weighing a moratorium with Amazon's own engineers testifying for it, most planned facilities sited on drought land while their water demand climbs toward 73 billion gallons, on-site gas plants forecast to raise the bills of residents who never asked for them, OpenAI filing toward an $850bn listing while struggling to turn a profit and declaring the chat interface that made it famous dead, Apple renting its flagship assistant's intelligence from the rival it built a wall to keep out, Meta stripping face-recognition code from a 50-million-download app one day after a reporter documented it. But friction generates edges, and an edge is what finally lets you tell one thing from another. Step back for a moment and you can see it: the moat coming loose all at once - financially as both labs submit to disclosure, geographically as the UK and EU build the stack they used to lease, physically as the most integrated company on earth sources its own assistant from a competitor; the cost of the buildout being pulled back from drought-stricken water tables and household ledgers onto the developers who externalized it; the latency between a twenty-year-old discovery and its first attempt in a living person collapsing into a single contest year. Every transformation has a breaking point. A wall can seal off what it surrounds... or, once it comes loose, let everything it was built to hoard spread to the people who stood outside it.

AI Releases & Advancements

New today

  • Apple: Released the Core AI Framework in developer beta at WWDC 2026, a new SDK enabling developers to build AI-powered applications using Apple Intelligence capabilities across iOS 27, macOS 27, and other Apple platforms. (Apple Developer Documentation)
  • Hugging Face / multi-org: Expanded OpenEnv under a multi-organization governance committee - including Meta-PyTorch, NVIDIA, Unsloth, Modal, Prime Intellect, and Hugging Face - repositioning it as an open interoperability protocol layer for agentic RL environments; the project now lives at huggingface/OpenEnv and exposes a Gymnasium-style API over HTTP/WebSocket with MCP as a first-class citizen. (Hugging Face Blog)
  • NVIDIA: Released an NVFP4 mixed-precision LLM pretraining recipe for JAX and MaxText via the JAX-Toolbox GitHub, enabling 4-bit training on NVIDIA Blackwell hardware with no measurable accuracy loss versus the FP8 baseline; uses five techniques including micro block scaling, E4M3 block scale factors, Random Hadamard Transform, 2D weight scaling, and stochastic rounding. (NVIDIA Developer Blog)

Sources and Further Reading

Artificial Intelligence & Technology's Reconstitution

Institutions & Power Realignment

Scientific & Medical Acceleration

Economics & Labor Transformation

Infrastructure & Engineering Transitions

The Century Report tracks structural shifts during the transition between eras. It is produced daily as a perceptual alignment tool - not prediction, not persuasion, just pattern recognition for people paying attention.

Read more