Google launches AI chips with 4× boost, lands Anthropic multibillion deal
Google rolled out a fresh batch of AI processors this week, and the headline-grabber is the claim of about four times the raw performance of the older TPU generation. At the same time the company announced a multibillion-dollar partnership with Anthropic - the startup that built Claude - which looks like one of the biggest AI-centric deals we’ve seen this year. The chips are impressive, sure, but what’s really catching my eye is how the two firms intend to scale their workloads.
Anthropic is set to lean on Google’s cloud, tapping a huge pool of compute that could, in theory, reshape its model-training roadmap. Analysts are already running the numbers and it seems the pact might give Anthropic access to roughly a million of the new units, plus the needed networking and storage fabric. That matters because training ever-larger language models isn’t just about raw speed; cost, efficiency and how comfortable engineers are with the platform all play a role.
This context, I think, helps explain Google’s thinking.
The company specifically cited TPUs' "price-performance and efficiency" as key factors in the decision, along with "existing experience in training and serving its models with TPUs." Industry analysts estimate that a commitment to access one million TPU chips, with associated infrastructure, networking, power, and cooling, likely represents a multi-year contract worth tens of billions of dollars -- among the largest known cloud infrastructure commitments in history. James Bradbury, Anthropic's head of compute, elaborated on the inference focus: "Ironwood's improvements in both inference performance and training scalability will help us scale efficiently while maintaining the speed and reliability our customers expect." Google's Axion processors target the computing workloads that make AI possible Alongside Ironwood, Google introduced expanded options for its Axion processor family -- custom Arm-based CPUs designed for general-purpose workloads that support AI applications but don't require specialized accelerators.
People are already wondering if the promised 4× boost will actually show up in real-world workloads. Google’s seventh-generation TPU - Ironwood - lands together with more Arm-based compute choices, which seems to signal a shift from pure model training to serving billions of users. The company points to price-performance and efficiency, and leans on its long-standing TPU know-how to lock in a multibillion-dollar deal with Anthropic.
Analysts have spotted a pledge for up to one million TPU units, but the fine print and the day-to-day effect on jobs are still fuzzy. Google Cloud calls the kit its most powerful AI infrastructure yet, yet it’s hard to say how fast customers will jump on board when alternatives are already out there. All in all, the move feels like a clear push to lock Google into the AI serving arena, but whether those hardware gains will spread across the industry remains an open question.
More in Business & Startups
GPU Loans Depend on Microsoft Contracts to Secure Income Guarantees
The financing model behind high‑end graphics processors has a hidden backstop that most investors...
Salesforce adds 6,000 enterprise customers in three months amid AI risk concerns
Salesforce’s recent surge—6,000 new enterprise accounts in just three months—has drawn attention...
HCLTech partners with OpenAI as AI and cloud drive steady IT growth
Why does a deal between HCLTech and OpenAI matter now? While the broader market buzzes about...
AI shifts hiring: companies favor generalist engineers over specialists
Why are hiring managers rethinking their talent playbooks? The rise of AI tools is nudging firms...
Rivian showcases new AI chip and silicon inspection labs in office tour
Rivian is opening its doors to a behind‑the‑scenes look at the hardware that could define the next...
Common Questions Answered
What performance improvement does Google's new Ironwood TPU claim over the previous generation?
Google states that the seventh‑generation Tensor Processing Unit, named Ironwood, delivers a four‑fold increase in raw performance compared to its prior TPU line. This boost is intended to accelerate both model training and large‑scale serving workloads.
How does the multibillion‑dollar deal with Anthropic relate to Google's AI chips?
Anthropic will use Google's cloud infrastructure, including access to up to one million Ironwood TPUs, as part of a multi‑year agreement valued at tens of billions of dollars. The partnership leverages the chips' price‑performance and efficiency, which were cited as key reasons for Anthropic's choice.
What aspects of price‑performance and efficiency were highlighted as reasons for Anthropic's TPU adoption?
The article notes that Anthropic specifically cited the TPUs' superior price‑performance ratio and operational efficiency, along with Google's extensive experience in training and serving models on TPUs. These factors helped secure the large‑scale cloud commitment.
What additional compute options accompany the Ironwood TPU in Google's latest hardware rollout?
Alongside the Ironwood TPU, Google introduced expanded Arm‑based compute options, signaling a broader strategy to support diverse AI workloads beyond just tensor processing. This diversification aims to improve flexibility for serving billions of users.