Two Trillion-Dollar Bets, One Constraint: What Anthropic’s Filing and Google’s $80bn Raise Reveal About the AI Endgame

If there is one thing this week has shown, it is that these are no longer companies in any ordinary sense. Anthropic and Alphabet, Google’s Parent are monolithic forces large enough that their internal decisions now set the terms for an entire industry, and on the same day, each told the market how it intends to win the next decade of artificial intelligence, in language that turned out to be opposite.

Anthropic confirmed on 1 June that it had confidentially submitted a draft registration statement on Form S-1 to the US Securities and Exchange Commission, the first formal step toward a public offering that would value the maker of Claude at close to $1 trillion. Within the same news cycle, Alphabet said it would sell $80 billion in stock, including a $10 billion investment from Warren Buffett’s Berkshire Hathaway, to fund the compute infrastructure its AI business now requires.

One giant is preparing to raise money by selling a piece of itself to the public. The other is raising money to build the physical capacity it says it cannot expand quickly enough. Both are responding to the same pressure, and they have chosen contradictory ways to meet it.

That pressure is compute, the combination of chips, data centres, power and supply chain that determines how much intelligence a company can serve to customers, and it has become the single constraint governing every serious decision in frontier AI. The entities that secure it set the pace for everyone else. The ones that misjudge it, in either direction, risk either starving their own growth or spending themselves into ruin.

What makes this week worth examining is that two organisations of almost unfathomable scale looked at the identical problem within hours of each other and reached conclusions that cannot both be described as caution. A near-trillion-dollar valuation on one side and an $80 billion capital raise on the other are not the moves of firms competing for a market. They are the moves of institutions large enough to shape one.

Anthropic is selling restraint as a strategy, not a limitation

The filing arrived less than a week after Anthropic closed a $65 billion Series H round that lifted its post-money valuation to roughly $965 billion, co-led by Altimeter Capital, Dragoneer, Greenoaks, Sequoia Capital, Capital Group, Coatue and D1 Capital Partners. The trajectory behind that number is among the steepest in the history of private markets, with the annualised revenue run rate reaching $47 billion in May, up from roughly $9 billion at the end of 2025, growth the company has attributed largely to enterprise demand for Claude and its coding products.

The speed caught Anthropic itself off guard, and Dario Amodei, Co-Founder and CEO, has been candid about the scale of the miss. “We planned for a world of 10x growth per year. In Q1 2026, we saw 80x annualised growth per year in revenue and usage,” he told the company’s developer conference in May. The strain that followed was, by his own account, a direct consequence of that gap. “That is the reason we have had difficulties with computing,” he said.

The confidential mechanism matters to what the public can currently know, because filing a draft S-1 under standard SEC review procedures lets Anthropic begin the process without disclosing revenue, margins or risk factors, which means the near-trillion-dollar figure rests for now on private financing rather than audited public accounts. Amodei kept the company’s own statement deliberately narrow. “This gives us the option to go public after the SEC completes its review,” he said, a sentence that promises nothing about timing or price and reflects the latitude a confidential filing is designed to preserve.

What separates Anthropic from its rivals is not the ambition but the spending discipline underneath it, and Amodei has been unusually explicit about why he refuses to match competitors dollar for dollar on infrastructure, casting the largest compute commitments as a route to insolvency rather than dominance. “If my revenue is not $1 trillion, if it’s even $800 billion, there’s no force on Earth, there’s no hedge on Earth that could stop me from going bankrupt if I buy that much compute,” he said.

The point was that he could not commit to $1 trillion a year of computing in 2027, because a growth rate of five times rather than ten would be enough to end the company. He has reduced the whole question to a distinction between profit and loss that the rest of the industry has been slower to internalise. “I think I’d rather have the largest revenue than the largest data centre, because one is black and the other is red,” he said. The line captures a philosophy his rivals have been reluctant to adopt, that a position which generates cash is worth more than one which merely consumes it impressively, however much the second photographs better in a press release.

Google is buying the exact capacity Anthropic will not overcommit to

Alphabet’s $80 billion raise is the deliberate mirror image of that restraint. The company said the capital would fund investments in its AI compute infrastructure to meet what it described as unprecedented customer demand, and its framing of the problem was the precise inverse of Amodei’s. Where Anthropic guards against overbuilding, Alphabet told investors that demand from enterprises and consumers was running at levels meaningfully exceeding its available supply, and that scaling its investment was how it intended to secure the foundational infrastructure for the growth ahead.

The raise is structured as $30 billion in underwritten offerings, $40 billion in staggered open-market sales and the $10 billion Berkshire stake, and it sits on top of an already enormous commitment, with the company guiding to capital expenditure of $180 billion to $190 billion in 2026 and signalling a significant further increase in 2027.

Sundar Pichai, CEO of Alphabet and Google, has left little doubt about where the binding constraint lies. Asked what kept the company’s executives awake, he answered not with competition or regulation but with two words. “Compute capacity,” he said, before laying out the physical limits behind them. “Be it power, land, supply chain constraints, how do you ramp up to meet this extraordinary demand for this moment?”

The question showcases the $80 billion as a response to scarcity rather than ambition, and his wider case for the spending rests on the conviction that AI is already paying its way across the business rather than promising to do so later. “AI is the most profound platform shift of our lifetimes,” Pichai said.

In Alphabet’s most recent results, he was more specific about where that shift was registering, telling investors that “our AI investments and full-stack approach are lighting up every part of the business.” The figures attached to those words support the reading that this is demand-led rather than speculative, with Google Cloud revenue having grown 63% in the most recent quarter and the company’s first-party models processing more than 16 billion tokens per minute through direct customer use of its interface.

The scale of that commitment is easier to grasp set beside the wider field. Alphabet, Microsoft, Meta and Amazon are together expected to spend more than $700 billion on AI-related capital this year, a figure Wall Street analysts expect to climb above $1 trillion in 2027. Alphabet has reached the debt markets repeatedly to fund it, with a bond issuance exceeding $30 billion in February and a further raise of roughly $11 billion in European currencies, on top of a $25 billion sale the previous November.

The $80 billion equity raise is therefore not an isolated event, but the latest instalment in a sustained build, and the Berkshire investment lends it particular weight, given the reluctance of Buffett’s conglomerate to participate in technology bets it does not consider durable.

The same constraint, read as opposite kinds of risk

Held against each other, the two bets are a disagreement about which danger is greater. Amodei treats overcommitment as the existential threat, accepting that Anthropic may not be able to serve all the demand it generates rather than risk the kind of spending that cannot be unwound if growth slows, and he has gone as far as criticising unnamed rivals for committing capital because it sounds impressive rather than because the returns are understood.

The strongest evidence for his position is that Anthropic has outpaced OpenAI on revenue while spending materially less on compute, which suggests restraint has not yet cost it the lead and may have sharpened it. Pichai treats underinvestment as the greater risk, on the logic that compute capacity is a direct driver of future revenue and that a giant which fails to build now forfeits demand it can never recover, since a customer turned away by a supply shortage does not wait politely for capacity to come online.

Also, Alphabet’s full-stack ownership of chips, data centres and models, a vertical integration that lets it convert capital into capacity at a cost structure a model developer renting compute from others cannot match.

The asymmetry between the two is worth drawing out, because it explains why each can act so differently without either being reckless. Anthropic does not own its infrastructure outright and has signed compute agreements with Amazon, Google, Nvidia and Microsoft, much of which is not expected to come online until late 2026 or into 2027, which leaves it exposed to the timing and pricing of suppliers but spares it the capital risk of building.

Alphabet owns the stack from silicon upward, which exposes it to the capital risk Amodei fears but frees it from dependence on anyone else’s roadmap. Each giant has, in other words, chosen the risk that best suits its own structure, and the contradiction between them is less a contest of judgment than a reflection of two fundamentally different shapes of business arriving at the same frontier.

Neither reading is obviously wrong, which is what makes the contrast instructive rather than a question of who has miscalculated. Anthropic’s model wins if demand growth proves volatile and the companies that overbuilt are left servicing infrastructure they cannot fill. Google’s model wins if demand continues to compound and the constraint stays on the supply side, where owning the physical layer becomes the decisive advantage. The two giants are, in effect, pricing different futures, and the market will eventually pay out on only one of them.

The valuations assume a future that has not yet arrived

The IPO race supplies the urgency around all of this. Anthropic’s confidential filing landed in an already crowded 2026 listing season that includes SpaceX, which is targeting a valuation well above $1 trillion, and OpenAI, which is preparing its own filing after a March round valued it at $852 billion. Anthropic’s move puts it procedurally ahead of its closest rival, and the appetite is vivid enough that reporting from several outlets has described venture investors offering to sell personal assets for access to shares, a detail that says as much about the mood of the market as about the company.

That enthusiasm deserves a sober counterweight, and it has one. The investor Michael Burry has publicly questioned whether trillion-dollar AI valuations can be justified, and the confidential nature of Anthropic’s filing means the public cannot yet test the $965 billion figure against audited financials, only against a private round and a run rate the company has disclosed on its own terms. The valuations on both sides rest on the assumption that demand compounds for years rather than quarters, and that assumption remains unproven however persuasive the present numbers appear. What the week established is not which giant is right, but that the entire industry has now organised itself around a single variable. Compute, rather than model quality or talent, is the resource that will decide the order, and the two best-placed institutions in artificial intelligence have just made opposing wagers on how much of it to own.