Google to invest up to $40B in Anthropic in cash and compute

At a glance:

• Google (Alphabet) will invest up to $40 billion in Anthropic, starting with $10 billion now at a $350 billion valuation and another $30 billion contingent on performance targets.
• Anthropic’s latest model, Mythos, has been released to a limited group of partners and is described as its most powerful model to date with significant cybersecurity applications; broader access is restricted due to potential misuse.
• Google Cloud will supply a fresh 5 gigawatts of compute capacity over the next five years, expanding an existing partnership that includes TPU-based capacity and ties to chipmaker Broadcom.

Strategic stakes in the AI compute race

The AI industry’s frontier is increasingly shaped by who can secure the physical infrastructure required to train and deploy ever-larger models. OpenAI has moved aggressively to lock in multi-hundred-billion-dollar capacity across cloud providers, chip suppliers, and energy, including an expanded deal with Cerebras this month. Against that backdrop, Anthropic has been scrambling to ease constraints around Claude use limits and to underwrite the enormous costs of scaling its systems. Earlier this month, Anthropic struck a data-center deal with CoreWeave and this week secured an additional $5 billion from Amazon as part of a broad agreement that could see Anthropic spend up to $100 billion for roughly 5 gigawatts of compute capacity over time. These moves highlight how model competitiveness is now tightly coupled to long-term infrastructure procurement.

Google’s position is distinctive because it functions both as a direct competitor in foundation models and as a critical infrastructure supplier to Anthropic. Anthropic relies heavily on Google Cloud for chips and infrastructure, including tensor processing units (TPUs) that are considered among the best alternatives to Nvidia’s in-demand processors. This dual role allows Google to fund Anthropic’s growth while simultaneously steering its compute spend toward its own cloud ecosystem, blurring the lines between partnership and platform lock-in in an industry where capacity is scarce.

Mythos, security constraints, and unsanctioned access

Anthropic’s latest model, Mythos, was released to a limited group of partners this month and is described by the company as its most powerful model to date, with significant cybersecurity applications. Because of potential misuse, Anthropic has restricted broader access while it works with select organizations to evaluate and address those risks, though the model has already fallen into unsanctioned hands. Running Mythos at scale is also likely to be expensive, compounding the pressure on Anthropic to secure cost-efficient compute and to calibrate safety controls against commercial demand.

The selective-release strategy reflects a wider industry pattern in which frontier models are debuted under controlled conditions to manage risk and performance feedback before wider deployment. For Anthropic, this approach intersects with business necessity: the company must demonstrate cutting-edge capability to justify outsized capital commitments while containing liabilities that could arise from misuse. How effectively Anthropic can secure and govern Mythos will influence both its valuation trajectory and its bargaining power with investors and cloud partners.

From TPUs to Broadcom: the infrastructure stack

Anthropic’s relationship with Google predates this week’s announcement. Earlier this month, Anthropic announced a partnership with Google and chipmaker Broadcom, which designs custom AI chips for Google, to access multiple gigawatts of TPU-based computing capacity beginning in 2027; a subsequent Broadcom securities filing put that figure at 3.5 gigawatts. That deal laid the groundwork for a vertically integrated stack in which custom silicon, cloud services, and model deployment are aligned under long-term timelines.

The new Google investment expands that arrangement, with Google Cloud now providing a fresh 5 gigawatts of capacity over the next five years, with room to scale further. This not only secures Anthropic’s training and inference runway but also deepens Google’s footprint in the high-margin market for AI-optimized compute. By tying capital to cloud capacity, Google converts what might otherwise be a competitive tension into a recurring-revenue flywheel, while Anthropic gains predictable access to specialized hardware that is in short supply across the industry.

Valuation, IPO timing, and investor momentum

Anthropic’s valuation stood at $350 billion as recently as February, and investors have since been eager to back the company at $800 billion or more, according to Bloomberg. The company is also reportedly considering an IPO as soon as October, a timeline that would place it among the most significant public debuts in AI if executed. That potential IPO adds urgency to the current fundraising and infrastructure deals, as public-market scrutiny will require clearer disclosures around compute costs, safety practices, and competitive differentiation.

The up-to-$40 billion commitment from Google, combined with Amazon’s $5 billion injection and the $100 billion compute agreement, signals that Anthropic is assembling a coalition of hyperscalers to underwrite its growth. This structure dilutes single-cloud dependency but also complicates governance, as different partners may have competing expectations around model roadmaps, safety standards, and regional deployment. Managing these relationships while scaling Mythos and future models will be a defining test for Anthropic’s operational maturity.

Competitive dynamics and what to watch

While Google and Anthropic are aligned on infrastructure today, their long-term interests may diverge as model capabilities improve and market share shifts. Google is investing in a direct competitor to its own Gemini models, and any significant advance by Anthropic could pressure Google’s pricing, product roadmaps, and enterprise positioning. Conversely, if Anthropic falters or faces regulatory constraints, Google’s compute-heavy bets could weigh on its cloud margins without delivering commensurate strategic upside.

What to watch next includes Anthropic’s ability to expand Mythos access without triggering high-profile misuse incidents, the timeline and terms of its anticipated IPO, and how Google’s 5-gigawatt capacity commitment is priced and scheduled over the next five years. Regional deployment choices, energy procurement strategies, and the balance between custom silicon (TPUs) and Nvidia alternatives will also shape the economics and competitive contours of this alliance as the AI race intensifies.