The Shape of the Money
The persistent structural paradox defining the two titans of AI.
Yesterday’s piece documented OpenAI’s sprint to match a capability that Anthropic has been building for over a year. Both companies are now building AI that acts inside applications rather than generating text about them, and six launches in eight days confirm that the two labs have arrived at the same conclusions about the future of their products.
But as the capabilities of their tools approach parity, everything else about these rival titans is rapidly diverging. In the span of three weeks, OpenAI closed the largest private funding round in history and signed a classified-use agreement with the Pentagon. Anthropic simultaneously lost its military contracts and was designated a supply-chain risk, then launched a $100 million enterprise push backed by private equity talks.
In January, this publication argued that OpenAI and Anthropic had chosen fundamentally different financial strategies. What we are seeing now is a concrete expression of those strategies. How each company is financing itself is now shaping its trajectory more than anything it ships.
Consolidated and locked in
On February 27, OpenAI announced a $110 billion funding round, the largest private technology financing in history. Amazon led with $50 billion, and Nvidia and SoftBank each committed $30 billion. The round valued OpenAI at $730 billion before the new capital, or $840 billion after it. No other investors participated in the initial close, though the company indicated that additional financial investors are expected to join as the round progresses. Microsoft, OpenAI’s longest-standing backer, was absent; both companies issued a joint statement affirming that their existing partnership remains unchanged. The concentration of the round is its most significant feature. Three investors with billions at stake in the AI infrastructure buildout now hold positions that function less as equity investments than as strategic partnerships denominated in dollars.
The Amazon relationship illustrates the pattern most clearly. The $50 billion investment arrived alongside an expansion of OpenAI’s existing AWS agreement by $100 billion over eight years, a commitment that makes Amazon Web Services the exclusive third-party cloud distribution provider for OpenAI’s enterprise platform, Frontier. OpenAI also agreed to consume at least two gigawatts of computing capacity powered by Amazon’s in-house Trainium chips. Nvidia supplies the chips on which OpenAI’s models train and run, and SoftBank brings a global distribution network, but neither relationship approaches the Amazon deal in scope. OpenAI’s cloud, its sales channel to enterprise customers, and a substantial share of its compute now run through a single investor.
The Pentagon deal followed the same logic. Anthropic overestimated its leverage in negotiations with the Department of Defense and lost its place on classified military networks. OpenAI signed a classified-use agreement the same week that it closed its $110 billion round, adding the U.S. military to a short list of patrons that already included the company’s three largest investors.
The deal carried a public cost. ChatGPT uninstalls surged 295% the day after the deal was announced, and Claude rose to number one on the App Store by March 1. Altman conceded that the announcement had been “opportunistic and sloppy” and revised the agreement’s surveillance language, adding prohibitions that converged toward the terms that Anthropic had originally demanded. The classified-use agreement itself stood, and OpenAI’s relationship with the Pentagon joined Amazon, Nvidia, and SoftBank as a fixture of the company’s institutional architecture.
The Pentagon contract also carries a product obligation: the military needs the capabilities that Anthropic was previously providing, and OpenAI must now deliver them. The product blitz documented in yesterday’s piece is partly explained by this obligation. A customer operating in classified environments needs AI that acts inside applications, not AI that describes what should be done. The funding round and the Pentagon deal share more than a calendar week. OpenAI’s infrastructure runs through Amazon, its chips come from Nvidia, and its product roadmap now answers in part to the Pentagon.
Diversity is their strength
The costs of Anthropic’s government setback are accumulating. The company faces a six-month phase-out of all federal use of its technology, and the military contracts that made Claude the first frontier model on classified networks are gone. The supply-chain-risk designation widens the loss beyond the Pentagon. Defense contractors must now certify that they do not use Anthropic’s models in military work, effectively excluding the company from an entire sector of potential customers. Anthropic has sued to challenge the designation, but its financial consequences are already making themselves felt.
Anthropic has not waited for the courts to recover. The Claude Partner Network, announced the same week, commits $100 million to making Claude the default AI platform for large enterprises. The program builds on the fact that Claude is currently the only frontier model available across all three major cloud providers, a position that Anthropic reached organically rather than by means of a single massive deal.
Anthropic is also reportedly in talks with Blackstone and Hellman & Friedman to create an AI joint venture. Private equity brings capital and commercial infrastructure without the strategic entanglements that accompany hyperscaler investment. Neither firm would route Anthropic’s cloud through a single provider or tie its product roadmap to a hardware vendor. Consistent with its launch of the Claude Partner Network, Anthropic is replacing lost government revenue through channels that do not reproduce the concentrated dependencies that define OpenAI’s position.
The Pentagon experience demonstrated how quickly a concentrated dependency can unravel, and it would appear that the lesson has not been lost on Anthropic’s leadership. Multi-cloud distribution, a $100 million enterprise commitment, and private equity capital amount to a deliberate strategy of building a commercial base broad enough that the loss of any single relationship is painful but survivable. However, this strategy demands both speed and commitment as Anthropic’s federal revenue is disappearing fast while its new commercial partnerships will need time to mature.
Pick your poison
The product convergence described above and in yesterday’s piece has a consequence that extends beyond the two labs themselves. As ChatGPT and Claude approach functional parity, enterprise customers are gaining the freedom to choose between them based on whom they wish to buy from rather than which tools they need. Upstream cloud infrastructure, vendor commitments, political exposure, and long-term flexibility will become increasingly important factors in any given company’s choice of AI platform.