There's a line buried in almost every analysis of Google's new Anthropic investment that deserves more attention than it's getting: Google is a direct competitor in AI models, and it's also Anthropic's single most important infrastructure supplier. Both things are true simultaneously, and neither cancels the other out.
Google plans to invest up to $40 billion in Anthropic, committing $10 billion now at a $350 billion valuation, with a further $30 billion contingent on Anthropic hitting certain performance targets. The deal also significantly expands Anthropic's access to computing capacity. Google Cloud will provide a fresh 5 gigawatts of capacity over the next five years, with room to scale further — building on an earlier partnership with Google and Broadcom to access multiple gigawatts of TPU-based compute beginning in 2027.
The headline number is large enough to demand context. This isn't an outlier. It's the latest move in a capital arms race that's reshaping how frontier AI gets built, funded, and controlled.
What makes Anthropic worth this kind of money right now isn't just the Claude models that enterprise teams have been quietly integrating into their workflows over the past year. Anthropic's annualised revenue went from $1 billion at the end of 2024 to $9 billion at end of 2025, then crossed $30 billion in early April 2026 — approximately 1,400% year-over-year growth, a trajectory Axios described as unlike anything in American business history.
The number of Anthropic customers spending more than $1 million annually doubled in less than two months. Claude Code, the company's developer-facing coding assistant, has been the primary accelerant — hitting $2.5 billion in annualised revenue in February alone, more than doubling since the start of the year, driven by enterprise adoption across finance, legal, healthcare, and software development.
The Google deal didn't arrive in a vacuum. Anthropic has faced widespread complaints about Claude use limits in recent weeks and responded with a series of infrastructure deals — including a partnership with cloud computing provider CoreWeave for data centre capacity. Amazon separately committed up to $25 billion in fresh investment, with Anthropic pledging to spend more than $100 billion on AWS technologies over the next decade and securing up to 5 gigawatts of compute capacity on Amazon's custom Trainium chips.
In the span of roughly a week, Anthropic locked in two of the largest investment commitments in the history of private technology — from two companies that also compete with each other, and with Anthropic itself, for AI dominance. The structure of the deals tells you something important: this isn't just venture funding. It's infrastructure pre-purchasing, dressed up as equity.
The Mythos factor deserves more scrutiny than it's received in the financial coverage.
Anthropic's new frontier model, Mythos, was released to a limited group of partners through a cybersecurity initiative called Project Glasswing. Partner organisations previewing it include Amazon, Apple, Broadcom, Cisco, CrowdStrike, the Linux Foundation, Microsoft, and Palo Alto Networks — selected because the model has capabilities Anthropic considers too dangerous to release broadly.
Mythos Preview fully autonomously identified and exploited a 17-year-old remote code execution vulnerability in FreeBSD, allowing root access from an unauthenticated user anywhere on the internet — with no human involvement after the initial request. The model demonstrated the ability to identify and exploit zero-day vulnerabilities in every major operating system and every major web browser. Anthropic notes that over 99% of the vulnerabilities found haven't been patched yet, which is why public disclosure has been tightly managed.
The model has already fallen into unsanctioned hands, according to Bloomberg — a detail that complicates the clean safety-first narrative Anthropic has built its brand around, and that will certainly come up if the company proceeds with a public offering.
"The firm is essentially attempting to sell its most powerful tool while simultaneously restricting how it can be used — a strategy that will certainly be under intense scrutiny during a future IPO roadshow." — Euronews Business
That scrutiny may arrive sooner than expected. Anthropic is reportedly considering an IPO as soon as October, with investors now willing to back the company at $800 billion or more — more than double its $350 billion February valuation. Initial reports put the target raise at $60 billion, which would rank among the largest IPOs in history.
The global picture here is more complex than the US-centric coverage suggests.
Google's relationship with Anthropic dates to 2023, when Alphabet invested $300 million for approximately 10% of the company. Months later it added $2 billion more. Prior to the current deal, Google's total investment exceeded $3 billion and it held roughly a 14% stake. The new commitment dramatically deepens that position — but in Europe, it's already drawing regulatory attention. The EU AI Act came into full force this year, and cross-shareholding arrangements between competing AI model providers and cloud infrastructure suppliers are precisely the kind of structure European competition authorities have flagged for review. Amazon's parallel Anthropic investment faces similar scrutiny — the FTC has been tracking the broader pattern of big tech firms acquiring strategic minority stakes in frontier AI labs. In Asia, the dynamic is different: South Korean and Japanese enterprises have been significant enterprise adopters of Claude models via AWS, and the deeper Google Cloud integration will expand TPU access for those customers, potentially accelerating adoption in markets where Nvidia GPU availability remains constrained.
Here's the contrarian read, and it's worth raising honestly: the compute-as-currency structure of these deals creates a dependency problem that the investment framing obscures.
The deal shows that a lab can be valuable not just because users love its models, but because cloud providers see strategic reason to lock it into their infrastructure — and that creates leverage for the provider and dependency risk for the customer. Anthropic has now committed to spending more than $100 billion on AWS over ten years, and is receiving gigawatts of Google TPU access in parallel. The company's independence — its ability to negotiate chip prices, switch providers, or hold competitive tenders — narrows with each deal it signs.
There's also a competitive structure question that doesn't get asked enough. Google's Gemini models are competing directly with Claude for enterprise accounts. Google Cloud is simultaneously Anthropic's biggest chip supplier and a distribution partner for Claude models. Google Cloud provides access to Anthropic's Claude models through its cloud division, which competes with Amazon Web Services and Microsoft Azure — meaning Google benefits whether Claude wins customers on Gemini's own platform or whether those customers go to Claude via Google Cloud infrastructure. It's a hedge so elegant it almost looks accidental. It probably isn't.
"The next phase of AI competition will be shaped less by short-term model launches and more by who can guarantee the next five years of compute." — Progressive Robot analysis of the Google-Anthropic deal
For enterprise technology decision-makers, the immediate practical questions are less about Anthropic's valuation and more about what comes next in the supply chain.
The compute concentration risk is real. If your AI stack is Claude-heavy and AWS-deployed, you now have financial exposure to a deal structure in which Anthropic has pre-committed a century of cloud spend to a single provider. That's not inherently bad — Trainium chips are genuinely competitive with Nvidia H100s for inference workloads — but it does mean your vendor's infrastructure choices are now locked in for a decade in ways that could affect pricing flexibility and failover options.
Three things worth watching closely in the next ninety days:
Whether the EU's Directorate-General for Competition formally opens a proceeding on the Amazon and Google minority stakes in Anthropic — the Microsoft-OpenAI arrangement prompted exactly this kind of review, and the structural parallels are explicit
Anthropic's IPO timeline and whether an October 2026 listing holds; if Mythos-related security incidents accelerate, the company may face pressure to delay until the model's risks are better characterised for public market disclosure
Claude's enterprise usage limits and whether the new compute capacity actually translates into reliability improvements by Q3 — because at $30 billion in annualised revenue, the gap between what customers expect and what the infrastructure delivers is already a reputational problem, not just a technical one
The $40 billion figure will age quickly as a news peg. The structural question it surfaces — who controls the chips, who owns the equity, and how those two things interact when competitive interests diverge — is going to matter for years.
