In September, Amazon said it would invest up to $4 billion in Anthropic, a San Francisco start-up that works on artificial intelligence.
Shortly thereafter, an Amazon executive sent a private message to an executive from another company. He said Anthropic won the deal because it agreed to build its AI using special computer chips designed by Amazon.
Amazon, he wrote, wanted to create a viable competitor to chipmaker Nvidia, a major player and kingmaker in the all-important field of artificial intelligence.
Last year’s surge in generic AI exposed just how dependent big tech companies had become on Nvidia. They can’t build chatbots and other AI systems without a special type of chip, which Nvidia has mastered over the past several years. They’ve spent billions of dollars on Nvidia’s systems, and the chip maker hasn’t kept up with demand.
So Amazon and other industry giants – including Google, Meta and Microsoft – are building their own AI chips. With these chips, tech giants can control their own destiny. They can rein in costs, eliminate chip shortages and eventually sell access to their chips to businesses that use their cloud services.
Pierre Ferragu, an analyst at New Street Research, said that while Nvidia sold 2.5 million chips last year, Google spent $2 billion to $3 billion on building one million AI chips of its own. He estimated that Amazon spent $200 million on 100,000 chips last year. Microsoft said it has started testing its first AI chip.
But the job is a balancing act between working closely with the chipmaker and its increasingly powerful Chief Executive Jensen Huang while competing with Nvidia.
Mr. Huang’s company accounts for more than 70 percent of AI chip sales, according to research firm Omdia. This provides an even larger percentage of the systems used in building generative AI. Nvidia’s sales have grown 206 percent in the past year, and the company has added nearly a trillion dollars in market value.
Nvidia’s revenue is a cost for the tech giant. Gil Luria, an analyst at investment bank DA Davidson, said orders from Microsoft and Meta have accounted for about a quarter of Nvidia’s sales over the last two full quarters.
According to Mr. Ferragu, Nvidia sells each of its chips for about $15,000, while Google spends an average of $2,000 to $3,000 on each of its chips.
“When he encountered a salesman who held him over a barrel, he reacted very strongly,” Mr. Luria said.
Companies continue to praise Mr Huang for being at the front of the line for their chips. He regularly appears on the event’s stage with their CEOs, and the companies are quick to say they are committed to their partnership with Nvidia. They all plan to keep offering its chips as well as their own.
While big tech companies are moving into Nvidia’s business, it is moving into theirs. Last year, Nvidia launched its own cloud service, where businesses can use its chips, and it’s funneling the chips into a new wave of cloud providers like CoreView, which is part of the big three of Amazon, Google, and Microsoft. Compete together.
“The tensions here are thousands of times greater than the usual skirmishes between customers and suppliers,” said technology consultant and investor Charles Fitzgerald.
Nvidia declined to comment.
According to research firm Gartner, the AI chip market is projected to more than double by 2027, reaching nearly $140 billion. Reputable chipmakers like AMD and Intel are also building specialized AI chips, as are start-ups like Cerebra and Sambanova. But Amazon and other tech giants can do things smaller competitors can’t.
“In theory, if they can get to scale enough and they can get their costs down, these companies should be able to deliver something that’s even better than Nvidia,” said Naveen Rao, who co-founded the first A.I. One of the chip starts was established. UPS and later sold it to Intel.
Nvidia makes graphics processing units, or GPUs, which were originally designed to help render images for video games. But a decade ago, academic researchers realized that these chips were also really good at building systems, called neural networks, that now run generative AI.
As this technology took off, Mr. Huang immediately began modifying Nvidia’s chips and related software for AI, and they became the de facto standard. Most of the software systems used to train AI technologies were designed to work with Nvidia’s chips.
“Nvidia has great chips, and more importantly, they have an incredible ecosystem,” said Dave Brown, who runs Amazon’s chip efforts. That makes it “very, very challenging” to get customers to use a new kind of AI chip, he said.
Rewriting software code to use the new chip is so difficult and time-consuming that many companies don’t even try, said Mike Schroepfer, a consultant and former chief technology officer at Meta. “The problem with technological development is that a lot of it is over before it even begins,” he said.
Rani Borkar, who oversees Microsoft’s hardware infrastructure, said Microsoft and its partners need to make it “seamless” for customers to move between chips from different companies.
Amazon, Mr. Brown said, is working to make switching between chips “as simple as possible.”
Some tech giants have had success making their own chips. Apple designs the silicon in iPhones and Macs, and Amazon has deployed more than two million of its own traditional server chips in its cloud computing data centers. But such achievements take years of hardware and software development.
Google has the biggest lead in developing AI chips. In 2017, it introduced its Tensor Processing Unit, or TPU, named after a type of computation important for building artificial intelligence. Google used thousands of TPUs to build AI products including its online chatbot, Google Bard. And other companies have used the chip through Google’s cloud service to build similar technologies, including high-profile start-up Cohere.
Amazon is now on the second generation of Tranium, its chip for building AI systems, and a second chip built for offering AI models to customers. In May, Meta announced plans to work on an AI chip tailored to its needs, although it is not yet in use. In November, Microsoft announced its first AI chip, Maia, which will initially focus on running Microsoft’s own AI products.
“If Microsoft makes its own chips, it makes exactly what it needs for the lowest possible cost,” Mr. Luria said.
Nvidia’s rivals have used their investments in high-profile AI start-ups to promote the use of their chips. Microsoft has pledged $13 billion to OpenAI, the maker of the ChatGPT chatbot, and its Maia chip will serve up OpenAI’s technologies to Microsoft customers. Like Amazon, Google has invested billions in Anthropic, and it is also using Google’s AI chips.
Anthropic, which has used chips from both Nvidia and Google, is one of a handful of companies working to create AI using as many specialized chips as possible. Amazon said that if companies like Anthropic use Amazon’s chips extensively and also help design future chips, doing so could reduce costs and improve the performance of these processors. . Anthropic declined to comment.
But none of these companies will be able to overtake Nvidia any time soon. Its chips may be expensive, but they are some of the fastest on the market. And the company will continue to improve its momentum.
Mr. Rao said his company, Databricks, trained some experimental AI systems using Amazon’s AI chips, but built its largest and most important systems using Nvidia chips because they offered higher performance and software. Plays well with a wide range of.
“We have many years of hard innovation ahead of us,” Amazon’s Mr. Brown said. “Nvidia is not going to stand still.”