Google is preparing a major capital push into Anthropic, committing an initial tranche now and promising more if the AI startup meets performance milestones — a move that underscores how control of large-scale computing is reshaping the race for advanced models. The deal, which links fresh funding to multigigawatt cloud capacity, has immediate implications for how quickly new, powerful systems can be trained and rolled out safely.
According to details disclosed by Anthropic, Alphabet’s cloud arm will invest $10 billion immediately at a company valuation of about $350 billion, with the potential to increase that commitment up to $40 billion if predefined targets are reached. The agreement also expands Google Cloud’s allocation of specialized hardware to Anthropic, aimed at supporting the large compute loads modern models demand.
Anthropic unveiled its latest model, called Mythos, earlier this month to a small group of partners. The company says Mythos represents a step-change in capability and has uses in cybersecurity, but access is tightly controlled because of concerns over misuse. Despite those restrictions, copies of the model have appeared outside authorized channels, highlighting the tension between rapid development and secure deployment.
Europe’s budget battle looms as Germany’s Merz pushes for EU overhaul
MRVL stock rises 8% after beating Q4 earnings, AI chip demand accelerates
The broader context is an industry-wide scramble for what matters most to generative AI: compute. Running and training next-generation models requires enormous, continuous processing power, and companies are locking up long-term capacity through partnerships with cloud providers, chip makers and data-center operators. OpenAI, for example, has pursued large-scale, cross-industry arrangements to secure this same kind of infrastructure.
Anthropic has been diversifying its infrastructure and capital sources in recent weeks. It agreed to a data-center capacity deal with CoreWeave and received an extra $5 billion from Amazon under an arrangement that ties Anthropic to substantial future compute purchases. Separately, Anthropic, Google and Broadcom announced a plan for TPU-based capacity beginning in 2027; filings by Broadcom indicate that figure may be in the low gigawatt range.
Under the new Google package, Anthropic will gain access to an additional 5 gigawatts of Google Cloud TPU-style compute over the next five years, with flexibility to scale further if needed. That level of capacity is significant: it can materially lower the time and cost required to train very large models and to run them at production scale.
- Immediate cash: $10 billion investment from Google at approximately $350 billion valuation.
- Contingent funding: Up to $30 billion more if Anthropic meets agreed performance milestones.
- Compute commitment: 5 GW of TPU-like capacity from Google Cloud across five years, scalable.
- Other partners: Recent deals with CoreWeave, a $5 billion injection from Amazon, and a Broadcom collaboration tied to 2027 capacity.
- Product status: Mythos released to select partners amid access restrictions and reports of unauthorized distribution.
- Exit plans: The company has reportedly discussed the possibility of an IPO later this year.
Why this matters now: securing large pools of compute is the practical bottleneck between laboratory breakthroughs and broadly available AI services. Firms that control long-term compute supply can accelerate model development, influence deployment pace, and shape which companies set technical and safety standards.
For users and businesses, the deal may have downstream effects on product availability, pricing and competitive dynamics. If Anthropic scales Mythos successfully, enterprises could gain access to more advanced capabilities — but the costs of running such systems may also push companies to favor a smaller set of cloud providers with the specialized hardware required.
There are also safety and policy questions. Restricted launches, like the one for Mythos, aim to let partners test misuse vectors before wider release; leaks suggest that containment remains difficult. The concentration of compute and capital among a handful of large tech firms and cloud vendors could increase regulatory scrutiny as governments assess market power, national-security concerns and the societal impact of more powerful models.
Anthropic’s market valuation and financing ambitions speak to investor appetite for AI platforms: reports indicate some backers have discussed valuations substantially above the current figure, and an IPO timeline has been floated for later this year. Whatever path the company chooses, the Google arrangement signals that access to sustained, high-end compute — not just algorithms — will be decisive in the next phase of generative AI development.
