Google, Anthropic discuss multibillion-dollar cloud deal: sources

Anthropic is in discussions with Google about a potential cloud computing deal valued in the high tens of billions of dollars, according to people familiar with the matter.

The discussions, which are not finalized, would involve Google providing additional cloud computing services to Anthropic.

Google is already an investor and cloud provider for Anthropic, which was founded in 2021 by former OpenAI employees in San Francisco and is known for its Claude large language models.

Neither company commented on the ongoing talks.

After news of the discussions, Google shares rose more than 3.5% in extended trading, while Amazon, another Anthropic investor and cloud provider, saw its shares slip about 2%.

Anthropic recently closed a US$13 billion funding round led by Iconiq Capital, with Fidelity Management and Research Co. and Lightspeed Venture Partners as co-leads, nearly tripling its valuation to US$183 billion including dollars raised.

Google has invested about US$3 billion in Anthropic to date, while Amazon has committed roughly US$8 billion, and Anthropic is a key AI customer of Amazon Web Services, making major use of Amazon’s custom AI chips.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Google weighs a big compute pact that could reshape AI model costs

Anthropic is weighing a compute pact with Google in the high tens of billions, far above Amazon’s $8 billion and past Google funding near $3 billion. That scale hints at a shift in its setup or prep for work beyond Claude 4, the latest Claude language model.
After raising $13 billion to a $183 billion valuation, the lab is still chasing huge compute blocks rather than saving cash, which could signal investor pressure to prove aggressive scaling or real needs that current deals do not cover.
Google appears ready to supply that capacity to counter the Microsoft OpenAI stack on models with cloud. It is also offering Tensor Processing Units, Google‑designed chips for AI training and inference, to third‑party clouds like Fluidstack 1 2. That move helps justify wider TPU bets tied to Anthropic work.

Hardware-tuned models become the next edge for infrastructure firms

Colocation operators are third‑party facilities where companies rent space and power for their servers. Google is placing TPUs in sites like Fluidstack’s New York build 1, giving operators a pitch in regions where it runs 29 data center sites in 11 countries 3.
Amazon Web Services is AWS, Amazon’s cloud computing arm. Labs will hedge chips across AWS and Google TPU fleets, which boosts demand for multi‑chip options over Nvidia‑only builds 4. Smaller cloud service providers, CSPs, already do this 4.
Chip startups plus semiconductor investors get a signal from splitting work across AWS custom silicon and Google TPUs. Efficiency wins like Google’s 30x jump from first to seventh‑gen TPU matter more than lock‑in 5. That creates room for niche processors with better performance per dollar.