The $73 Billion Hedge
Google and Amazon just committed $73 billion to the company whose success would prove their own AI models are not enough.
On April 24, 2026, Google committed up to $40 billion in Anthropic: $10 billion upfront at a $350 billion valuation, with $30 billion contingent on performance milestones, plus five gigawatts of computing capacity delivered through Google Cloud over five years beginning in 2027. The compute comes via Tensor Processing Units and custom chips designed with Broadcom, infrastructure that begins coming online next year. Four days earlier, on April 20, Amazon had expanded its own commitment to Anthropic to $33 billion: $5 billion in fresh equity priced at the same $350 billion pre-money valuation, up to $20 billion in additional investment tied to commercial milestones, layered on top of $8 billion in prior staged investments deployed between 2023 and 2025. As part of the Amazon deal, Anthropic committed to spend over $100 billion on AWS over the next decade. Combined: $73 billion in committed capital from two companies that are simultaneously Anthropic’s largest investors, its infrastructure providers, and its direct competitors.
The thesis is uncomfortable for both Google and Amazon, but the arithmetic makes it inescapable: you do not spend $73 billion on a competitor unless you doubt your own product.
Google built Gemini. Google DeepMind, with over 2,000 researchers, is the world’s largest AI research organisation. Google has spent tens of billions developing frontier models, integrating them into Search, Workspace, Cloud, YouTube, Android, and every major product in its portfolio. Google controls the Tensor Processing Unit, arguably the most advanced AI training hardware in production. And Google just invested $40 billion in the company whose Claude models directly compete with Gemini across every benchmark, every enterprise use case, and every developer workflow. Amazon has built its own AI stack through AWS, launched Bedrock as its managed AI service, developed custom Trainium chips specifically for AI training, and positioned itself as a comprehensive AI infrastructure provider. Amazon has its own foundation models. And Amazon just committed $33 billion to a company whose success at scale would demonstrate that Amazon’s own AI capabilities are insufficient for the market’s most demanding customers.
The historical parallel that illuminates the logic, and that both companies would prefer not to draw, is Microsoft’s $150 million investment in Apple in August 1997. Apple was ninety days from bankruptcy. Microsoft dominated personal computing. The investment was not charity and it was not a bet on Apple’s products. It was a strategic hedge: Microsoft needed Apple to survive to avoid an antitrust monopoly ruling, and it wanted to ensure that Microsoft Office remained the default productivity suite on Macintosh regardless of the competitive dynamics between Windows and Mac OS. The investment looked irrational at the time. It proved brilliant. Apple survived, rebuilt under Steve Jobs, and eventually became the most valuable company on earth. Microsoft’s $150 million stake appreciated into billions. The two companies found an equilibrium that allowed both to thrive for decades.
Google and Amazon are making the same structural bet at a vastly larger scale and with a critical additional dimension. They are not merely investing in a competitor to hedge their own position. They are providing the compute infrastructure that competitor depends on, which means they capture value regardless of which models win the market. If Claude becomes the dominant AI model, Google and Amazon collect cloud revenue from running Claude. If Gemini or Amazon’s own AI offerings win, the Anthropic investments produce equity returns that more than compensate. The only scenario in which both investments lose is one where neither Anthropic’s models nor the cloud providers’ own models dominate, an outcome that implies AI itself underperforms as a technology category, which neither Google nor Amazon believes. The hedge is nearly perfect. It is also nearly unprecedented in its scale and in the directness of the competitive relationship it straddles.
The deal structures reveal the strategy with precision that corporate communications cannot obscure. Google’s five gigawatts of compute commitment is not just an investment. It is a lock-in mechanism. Every gigawatt of compute Anthropic consumes through Google Cloud generates cloud revenue for Google regardless of whether Gemini or Claude wins the model competition. Amazon’s structure mirrors this exactly: five gigawatts of compute capacity for Claude training and inference, deployed across multiple US, Asian, and European data centre sites through 2036. Both investors are paying Anthropic to be their customer. The investments function simultaneously as equity bets, infrastructure revenue guarantees, and strategic options on the outcome of the AI race. If you own a meaningful stake in the company most likely to challenge your own products, and that company is contractually obligated to spend $100 billion on your cloud platform over the next decade, you have constructed a position where you profit regardless of who wins.
Guan Xiaoming, a senior research fellow at the Beijing Academy of Artificial Intelligence, published a May analysis arguing that this dual infrastructure lock-in has no precedent in technology history. The closest analogy he identified was the railway era, when competing railway companies sometimes invested in each other’s lines to guarantee traffic regardless of which route won. But even that analogy breaks down because railways did not simultaneously compete on the same routes they were investing in. Google and Amazon are doing something genuinely new: competing with Anthropic in the model market while guaranteeing Anthropic’s success generates infrastructure revenue for them, while simultaneously booking Anthropic’s rising valuation as their own profit.
The profit illusion is where the financial reporting becomes genuinely misleading. Fortune reported on April 30 that nearly half of Alphabet’s record $62.6 billion profit in Q1 2026, approximately $28.7 billion, came from updating the value of its equity holdings in private companies, primarily its Anthropic stake. When Anthropic closed its Series G round on February 12, 2026 at a $380 billion post-money valuation, it triggered mark-to-market accounting gains across every investor’s books. Google’s stake appreciated dramatically. That appreciation was recorded as income. It appeared in the profit figure that analysts cited when discussing Google’s “AI-driven earnings growth.” Similarly, Amazon’s Q1 2026 net income included $16.8 billion in pre-tax gains from its Anthropic stake, representing more than half of Amazon’s pre-tax income for the quarter. Amazon’s $8 billion historical investment in Anthropic is now worth more than $70 billion according to the company’s own valuation marks.
The circularity embedded in this arrangement deserves careful examination. Google and Amazon invest in Anthropic. Their investments drive up Anthropic’s valuation during funding rounds. The higher valuation triggers mark-to-market gains on their income statements. Those gains inflate their reported profits. Inflated profits are attributed to “AI momentum” by analysts. Higher reported profits justify higher stock prices for Google and Amazon. Higher stock prices give them more capital to invest in AI, including further investments in Anthropic. And the cycle continues until someone examines what the underlying operational AI margins actually look like absent the Anthropic appreciation. The answer, as Fortune’s reporting implied without stating explicitly, is that the “AI profit boom” celebrated in Q1 2026 earnings coverage was partially a valuation story rather than an operational story. Remove the Anthropic appreciation and the underlying AI businesses at both companies are less profitable than the headlines suggested.
Anthropic’s own trajectory adds a philosophical dimension to the financial one. The company was founded in 2021 by Dario and Daniela Amodei and other former OpenAI employees who left because they believed OpenAI was prioritising commercial interests over AI safety research. Anthropic positioned itself as the “safety-first” AI company, the responsible alternative, the laboratory that would demonstrate that commercial success and careful development were compatible. Five years later, Anthropic is valued at $350 billion in primary rounds and up to $1 trillion on secondary markets. It generates $30 billion in annualised revenue, having grown from $1 billion at end of 2024 to $9 billion at end of 2025 to $30 billion by spring 2026. Anthropic itself characterised this growth rate as “crazy.” The company has more than 1,000 business customers each spending over $1 million annually, a number that doubled in two months. It is reportedly in early discussions for an IPO that could come as early as October 2026. The safety-first company has become the most commercially successful AI company on earth, having passed OpenAI in revenue while spending a fraction of the training budget.
The paradox is not that Anthropic makes money. Companies that build excellent products should make money. The paradox is that Anthropic’s commercial success now depends on two investors whose strategic interests are in structural conflict with Anthropic’s competitive independence. Google wants Anthropic to succeed (investment returns, cloud revenue) and simultaneously wants Anthropic to fail (competitive threat to Gemini, reduction of Google’s AI narrative). Amazon wants Anthropic to succeed (equity appreciation, AWS revenue from the $100 billion commitment) and simultaneously wants its own AI products to win market share from Claude. These are not theoretical conflicts of interest. They are structural. Every strategic decision Anthropic makes, every pricing decision, every product launch, every partnership, every step toward IPO, is made in the context of two investors who are also its competitors and its sole infrastructure providers. Anthropic cannot meaningfully threaten to leave Google Cloud or AWS because its entire operational infrastructure depends on both. And Google and Amazon cannot meaningfully threaten Anthropic because doing so would destroy the value of their own investments.
Guan observed that the ownership web extends far beyond the Anthropic triangle. Microsoft invested approximately $13 billion in OpenAI over several years, though the exclusive partnership ended on April 27, 2026. Google invests in multiple AI startups through its venture arms. Amazon funds AI companies through its accelerator programme and its AWS credits-for-equity arrangements. The AI industry is not a competitive market in the traditional sense. It is an oligopoly connected by cross-investments where the three major cloud providers fund each other’s competitors while competing with those same companies. Google invests in Anthropic, which competes with Gemini. Amazon invests in Anthropic, which competes with AWS AI services and Bedrock. Microsoft invested in OpenAI, which competes with all of them. Each cloud provider also competes with the others for the compute contracts that the AI companies depend on. Traditional competitive analysis, the kind that assumes firms maximise their own market position against rivals, breaks down completely when the competitors are also each other’s shareholders, customers, and infrastructure providers.
Senators Elizabeth Warren and Ron Wyden launched formal inquiries into these arrangements in spring 2026. Their letters to Google, Microsoft, Anthropic, and OpenAI expressed concern that the partnerships “discourage competition, circumvent antitrust laws, and result in fewer choices and higher prices for businesses and consumers using AI tools.” The specific worry is monopolisation of limited compute resources: when Google and Amazon control the infrastructure that AI companies depend on, and simultaneously own equity stakes in those companies, and simultaneously compete with them in the model market, the competitive dynamics that theoretically drive innovation and lower prices are distorted beyond recognition. The senators asked pointed questions about whether the compute commitments create de facto lock-in, whether the equity stakes grant board influence or veto rights, and whether the arrangements amount to acquisitions structured to avoid merger review.
The antitrust question is legitimate but may exceed the reach of existing legal frameworks. Traditional antitrust law addresses horizontal consolidation, where competitors merge, and vertical integration, where companies acquire their suppliers or distributors. The Google-Amazon-Anthropic arrangement is neither horizontal nor vertical. It is diagonal: the same entities are simultaneously investor, competitor, infrastructure supplier, and financial beneficiary of the entity they compete with. No antitrust framework was designed to analyse a market structure where a company profits from its competitor’s success through equity appreciation, captures revenue from its competitor’s operations through infrastructure fees, and competes with that same entity in the product market, all simultaneously. The FTC and DOJ have tools for horizontal mergers. They have tools for vertical integration. They may not have tools for whatever this is.
The variable that determines whether the arrangement is stable or unstable is Anthropic’s degree of infrastructure independence. If Anthropic remains permanently dependent on Google Cloud and AWS for its compute needs, the conflicts of interest are permanent but manageable: all three parties profit, competition is somewhat muted but continues, and the antitrust concern produces disclosure requirements but no structural remedies. If Anthropic achieves infrastructure independence, perhaps through its own data centres, perhaps through an IPO that provides capital for independent compute procurement, the investments become purely financial stakes and the competitive dynamics return to normal. And if antitrust regulators intervene to force structural separation, the entire arrangement unwinds in ways that reduce the capital available to all parties but restore competitive clarity.
Guan’s final observation in his May analysis cuts to the strategic core of what the $73 billion reveals. A hedge, by definition, is a bet against your own position. Google is hedging against the possibility that Gemini loses the model competition. Amazon is hedging against the possibility that its own AI stack is insufficient. Together, the two largest cloud infrastructure providers on earth are telling the market, through their capital allocation decisions rather than their earnings calls, that they are not confident their own AI capabilities will be enough. The $73 billion is the most expensive admission of uncertainty in corporate history. And the question it poses for the industry extends beyond Google and Amazon: if the companies with the most resources, the most data, the most engineering talent, the most compute capacity, and the longest research track records are not confident enough to bet exclusively on their own models, what does that say about the predictability of frontier AI development? What does it say about the durability of any competitive moat? And what does it say about the confidence that anyone should have in the trajectory of an industry whose own incumbents are hedging against their own failure at a scale measured in tens of billions?
ANNEX: WHAT DOES THE WORLD LOOK LIKE IF THE HEDGE DEFINES THE INDUSTRY?
Google and Amazon have collectively committed $73 billion to Anthropic while simultaneously competing with Anthropic in the AI model market. The four scenarios below sum to 100 percent and represent Scenarica’s probability-weighted assessment of how this unprecedented structural arrangement evolves through 2028.
Comfortable Coexistence -- 35%
If you are an institutional investor building a long-term position in cloud infrastructure stocks or evaluating the AI sector’s competitive dynamics for portfolio construction, this is the scenario where the current equilibrium persists. Anthropic, Google, and Amazon find a stable arrangement where Anthropic dominates certain verticals, primarily developer tools, enterprise AI, and safety-critical applications, while Google dominates consumer AI through its product ecosystem and Amazon dominates cloud-native AI services through Bedrock and its managed offerings. The cross-investments produce returns for all parties. Cloud revenue from Anthropic’s compute consumption grows for both Google and Amazon. Antitrust scrutiny produces transparency requirements and disclosure obligations but no structural remedies. The ownership web persists because it works well enough for everyone involved.
Variable to watch: Anthropic’s market share in enterprise AI relative to Google Cloud AI and AWS AI services. If all three grow simultaneously without significant market share loss among them, this scenario’s probability rises to 40 to 45 percent. Stable coexistence requires that the total AI market expands fast enough to accommodate all players without forcing zero-sum competition.
Anthropic Breaks Free -- 25%
If you are advising Anthropic on its IPO strategy or evaluating the long-term independence of the company, this is the scenario where revenue growth and public market capital give Anthropic sufficient leverage to reduce infrastructure dependence. Anthropic’s revenue trajectory, currently $30 billion annualised and accelerating, gives it the financial resources to contract or build independent compute capacity. An IPO, potentially as early as October 2026, provides access to public market capital without the strategic strings attached to Google and Amazon funding. The investments become purely financial stakes: Google and Amazon benefit as equity holders but lose the infrastructure lock-in they paid $73 billion to secure. Anthropic becomes a fully independent company that happens to have two large shareholders who are also competitors.
Variable to watch: any announcements by Anthropic regarding independent data centre construction, compute procurement outside Google Cloud and AWS, or partnerships with alternative infrastructure providers. An Anthropic IPO filing would also shift this scenario’s probability to 30 to 35 percent, as public market capital reduces dependence on strategic investors.
Antitrust Intervention -- 25%
If you are a competition policy analyst or a litigator tracking technology antitrust enforcement, this is the scenario where the Warren-Wyden inquiries produce consequences. Legislation or FTC action limits cross-ownership between AI model companies and their cloud infrastructure providers. Google and Amazon are required to divest some or all of their Anthropic stakes, or to restructure the compute commitments to remove lock-in provisions. The competitive landscape becomes clearer: companies are either competitors or investors, not both simultaneously. But the AI companies lose access to the capital and subsidised compute that the current arrangements provide. Innovation may slow as capital becomes more expensive and less available.
Variable to watch: FTC or DOJ public statements on AI industry cross-ownership, any legislative proposals emerging from the Warren-Wyden investigation, and the timing of any enforcement action relative to Anthropic’s potential IPO. If a formal antitrust investigation is opened before Anthropic files for IPO, this scenario’s probability rises to 30 to 35 percent.
Model Commoditisation -- 15%
If you are modelling the long-term value of proprietary AI models or evaluating whether Anthropic’s current revenue growth is sustainable, this is the scenario where the thing being hedged stops mattering. Open-source models from Meta’s Llama, Mistral, and potentially others become competitive with proprietary models across most commercially relevant tasks. The quality gap between Claude and free alternatives narrows to the point where enterprise customers no longer pay meaningful premiums for proprietary access. Anthropic’s revenue growth stalls. Google’s and Amazon’s $73 billion in committed capital depreciates as the moat of proprietary model quality erodes. The hedge fails not because the wrong horse won, but because the race stopped mattering.
Variable to watch: enterprise adoption rates of open-source models relative to proprietary APIs. Track Llama 4’s enterprise deployment numbers and Mistral’s revenue growth. If open-source models capture more than 30 percent of enterprise AI workloads by mid-2027, this scenario’s probability rises to 20 to 25 percent.
Sources:
Bloomberg, “Google Plans to Invest Up to $40 Billion in Anthropic,” April 24, 2026.
TechCrunch, “Google to invest up to $40B in Anthropic in cash and compute,” April 24, 2026.
CNBC, “Google to invest up to $40 billion in Anthropic as search giant spreads its AI bets,” April 24, 2026.
CNBC, “Amazon to invest up to another $25 billion in Anthropic as part of AI infrastructure deal,” April 20, 2026.
Dataconomy, “Amazon Deepens Anthropic Deal With $25 Billion New Investment,” April 21, 2026.
Fortune, “Half of Google’s and Amazon’s ‘blowout AI profits’ came from a stake in Anthropic,” April 30, 2026.
The Next Web, “Amazon Q1 revenue hits $181.5B but $16.8B Anthropic gain inflates net income,” Q1 2026.
Senator Elizabeth Warren, “Warren, Wyden Launch Investigation into Google, Microsoft Partnerships with AI Developers,” 2026.
Axios, “Google’s $40B Anthropic move is Big Tech’s latest huge AI bet,” April 24, 2026.
SaaStr, “Anthropic Just Passed OpenAI in Revenue. While Spending 4x Less to Train Their Models,” April 2026.
Disclaimer: This report is published by Scenarica Intelligence for informational purposes only. It does not constitute investment advice, a solicitation to buy or sell any financial instrument, or a recommendation regarding any particular investment strategy. Scenarica Intelligence is not a registered investment adviser or broker-dealer. All scenario probabilities and assessments represent the analytical judgment of Scenarica Intelligence and are subject to change without notice. Past performance of any asset or strategy discussed does not guarantee future results. Readers should conduct their own due diligence and consult with qualified financial advisers before making investment decisions.
Scenarica Premium: The full Scenarica suite includes Geopolitics, Economy, Bitcoin, AI, and Sunday Edition.
Scenarica Intelligence
We don’t predict the future. We price it.