Anthropic and Google Cloud are closing a historic deal – a million chips and 1GW of computing power for AI
Imagine a world where one computer thinks like an entire city...
This is more or less what Anthropic and Google are building now.
Yesterday they announced a huge agreement, perhaps the largest in the world in the AI field:
Anthropic will gain access to up to one million TPU (Tensor Processing Units) units – Google's dedicated chips for training artificial intelligence models.
Together, they will accumulate computing power of more than 1 gigawatt (1GW) by 2026, enough power to run tens of thousands of models like Claude simultaneously.
A multi-billion dollar deal: a huge step towards the future of AI. The scope of the agreement is estimated at tens of billions, and it is expected to support annual revenues of about $7 billion for Anthropic.
The natural comparison is with OpenAI's "Stargate" project with a capacity of 33GW.
But note: Anthropic chose a different approach:
Not speed, but precision. Not size, but efficiency.
An approach that focuses on perfect execution and measurable results, not just grand statements.
The multi-cloud strategy: the winning combination. Instead of relying on one cloud, Anthropic wisely operates with a multi-cloud model, which allows it to choose the right provider for each task:
Google: Google TPU – used to run Claude models in real time.
Amazon: Amazon Trainium – designed for training new models quickly and at low cost. Nvidia: NVIDIA GPU – for inference and advanced research.
This combination creates a super-flexible system, where each component does what it does best.
The result?
Higher performance within the same budget.
This is a huge advantage in a world where every second of computation costs real money.
Google's Ironwood project: the new generation of chips
Google, for its part, is not staying behind.
It provides Anthropic with its TPU Ironwood chips, its seventh and most powerful generation to date, with significant improvements in performance and energy efficiency.
Google Cloud CEO, Thomas Kurian, said: 'This expansion represents deep trust in the capabilities of the TPU and the performance it has delivered over the years.' In other words, Google sees Anthropic not just as a customer, but as a strategic partner leading the next generation of AI.
Rapid growth at levels hard to imagine. Anthropic is not just building servers; they are building an empire.
The data is astonishing: Over 300,000 companies already use Claude systems (a 300-fold increase in two years!).
The number of enterprise customers paying over $100,000 per year, jumped 7-fold in one year.
A new product called Claude Code (an autonomous code agent) reached revenues of $500 million in just two months.
The company boasts that it is “the fastest growing product in the history of artificial intelligence.”
A predator's strategy: both Google and Amazon, a dual partnership that empowers Anthropic.
Anthropic chose not to put all its eggs in one basket.
Alongside Google, it continues to cooperate closely with Amazon Web Services (AWS), which invested $8 billion in it, compared to $3 billion from Google.
Amazon's cloud (AWS) not only invests, it is the infrastructural heart of Anthropic.
Its supercomputer, called Project Rainier, is based on Amazon's Trainium 2 chips, which are considered one of the most efficient solutions in the world in terms of cost-performance ratio.
Thus, Anthropic benefits from the best of both worlds: Google's speed and optimization, alongside Amazon's power and flexibility.
A word about Google's side
Google is bringing a huge growth engine to its business. Until today, it developed chips mainly for itself; now it's a business generating billions, just like that, with a bang!
Google is simply a monster; there's no other word for it.