Google unveiled Ironwood Thursday - its most powerful AI chip yet. The seventh-generation Tensor Processing Unit will be widely available in coming weeks.
Ironwood is designed for the heaviest AI workloads, from training large models to powering real-time chatbots and AI agents. Google says it's more than four times faster than its predecessor.
AI startup Anthropic plans to use up to 1 million Ironwood chips to run its Claude model. That's a massive deployment showing enterprise confidence in Google's custom silicon.
TPUs give Google an advantage as hyperscalers rush to build data centers and AI processors can't be manufactured fast enough to meet demand.
"Of the ASIC players, Google's the only one that's really deployed this stuff in huge volumes," said Stacy Rasgon, semiconductor analyst at Bernstein. "For other big players, it takes a long time and a lot of effort and a lot of money. They're the furthest along among the other hyperscalers."
Amazon Web Services made its first cloud AI chip, Inferentia, available in 2019, followed by Trainium three years later. Microsoft didn't announce its first custom AI chip, Maia, until late 2023.
Google has been developing TPUs for over a decade and made them available to cloud customers in 2018.
Google's TPU success has caught Nvidia's attention. When OpenAI signed its first cloud contract with Google earlier this year, the announcement spurred Nvidia CEO Jensen Huang to initiate further talks with the AI startup and CEO Sam Altman, according to The Wall Street Journal.
Unlike Nvidia, Google isn't selling chips as hardware. It provides access to TPUs as a service through its cloud.
Google's cloud business has become a major growth driver. Third-quarter cloud revenue jumped 34% year-over-year to $15.15 billion, beating estimates. The company ended the quarter with $155 billion in business backlog.
"We are seeing substantial demand for our AI infrastructure products, including TPU-based and GPU-based solutions," CEO Sundar Pichai said on the earnings call. "It is one of the key drivers of our growth over the past year."
Google doesn't break out TPU business size within its cloud segment. But D.A. Davidson analysts estimated in September that a "standalone" business consisting of TPUs and Google's DeepMind AI division could be valued at about $900 billion.
That's up from a $717 billion estimate in January. Alphabet's current market cap exceeds $3.4 trillion.
Customization is Google's major differentiator. TPUs offer efficiency advantages over competitive products.
"They're really making chips that are very tightly targeted for their workloads that they expect to have," said James Sanders, analyst at Tech Insights.
Rasgon said efficiency will become increasingly important because "the likely bottleneck probably isn't chip supply, it's probably power."
Tuesday, Google announced Project Suncatcher - exploring "how an interconnected network of solar-powered satellites, equipped with our Tensor Processing Unit (TPU) AI chips, could harness the full power of the Sun."
Google plans to launch two prototype solar-powered satellites carrying TPUs by early 2027.
"This approach would have tremendous potential for scale, and also minimizes impact on terrestrial resources," the company said. It would "test our hardware in orbit, laying the groundwork for a future era of massively-scaled computation in space."
Google's decade-long TPU investment is paying off as custom chips become crucial in the AI race. While Google continues buying Nvidia GPUs, its custom silicon gives it advantages competitors can't match.
Being "the only one that's really deployed this stuff in huge volumes" puts Google years ahead of AWS and Microsoft in custom AI chip development. That head start matters as power constraints become the limiting factor for AI infrastructure.
Anthropic committing to 1 million Ironwood chips validates Google's approach. Major AI companies are willing to build on Google's custom silicon rather than relying solely on Nvidia.
The $900 billion standalone valuation estimate for TPUs and DeepMind shows how valuable this business has become. What started as internal workloads over a decade ago now drives significant cloud revenue growth.
Rasgon's point about power being the bottleneck rather than chip supply highlights why efficiency matters. Google's custom chips designed for specific workloads use power more efficiently than general-purpose alternatives.
The space satellite project might sound futuristic, but it addresses real constraints. If power becomes the limiting factor for AI computation, solar-powered satellites with TPUs could provide massive scale without terrestrial resource impact.
For investors, Google's TPU success shows the value of long-term hardware investment. While Microsoft and Amazon play catch-up on custom chips, Google's decade head start has created meaningful competitive advantage.
