Intel passed on its ownership of OpenAI years ago. Now, it’s lagging behind

For US chip giant Intel, the darling of the computer age before it fell on hard times in the AI ​​era, things could have been very different.

About seven years ago, the company had an opportunity to buy shares in OpenAI, then a fledgling nonprofit research organization working in a little-known field called generative artificial intelligence, four people with direct knowledge of those discussions told Reuters.

Over several months in 2017 and 2018, executives at the two companies discussed various options, including Intel buying a total of 15% for $1 billion, three of the people said. They also discussed Intel taking an additional 15% stake in OpenAI if it makes the original hardware at cost, two of the people said.

Intel ultimately decided against the deal, because then-CEO Bob Swan did not think that the AI ​​models he was producing would be marketed in the near future and thus return the chipmaker’s investment, according to three sources, all of whom asked not to be identified to discuss confidentiality. news.

OpenAI was interested in the investment from Intel because it would reduce their reliance on Nvidia chips and allow the startup to build its own infrastructure, two of the people said. The deal also fell through because Intel’s data center division didn’t want to make products at cost, the people added.

An Intel spokesman did not respond to questions about the potential deal. Swan did not respond to a request for comment and OpenAI declined to comment.

Intel’s decision not to invest in OpenAI, which went on to launch the disruptive ChatGPT in 2022 and is now reportedly valued at around $80 billion, was never made public.

It is among a series of strategic mishaps that have seen the company, which was on the cutting edge of computer chips in the 1990s and 2000s, stumble in the AI ​​era, according to Reuters interviews with nine people familiar with the matter including former Intel. managers and industry professionals.

Last week, Intel’s second-quarter earnings caused the stock to drop more than a quarter of its value in its worst trading day since 1974.

For the first time in 30 years, a tech company is worth less than $100 billion. The former king of the market – whose marketing slogan “Intel Inside” has long represented the gold standard – is still struggling to get a blockbuster AI chip product to market.

Intel is now dwarfed by $2.6 trillion rival Nvidia, which has moved from video game graphics to the AI ​​chips needed to build, train and run large-scale production AI systems like OpenAI’s GPT4 and Meta Platforms’ Llama models. Intel also fell behind AMD’s 218 billion.

Asked about its AI progress, an Intel spokesperson referred to recent comments by CEO Pat Gelsinger, who said the company’s third-generation Gaudi AI chip, which it plans to launch in the third quarter of this year, will outpace competitors.

Gelsinger said the company has “20-plus” second- and third-generation Gaudi customers and that the next-generation Falcon Shores AI chip will launch in late 2025.

“We are nearing completion of a historic pace of design and development of new technologies, and we are encouraged by the product pipeline we are building to capture a large portion of the AI ​​market going forward,” a spokesperson told Reuters.

The game chips sweep the AI

On the OpenAI front, Microsoft stepped in to invest in 2019, pushing itself into the AI ​​era ushered in by the 2022 release of ChatGPT and a flurry of activity among the world’s biggest companies to use AI.

While in retrospect the deal would have been an opportunity for Intel, the company has been steadily losing the battle for AI supremacy for more than a decade, according to former executives and industry experts interviewed.

“Intel failed in AI because they didn’t present a unified product strategy to their customers,” said Dylan Patel, founder of semiconductor research group SemiAnalysis.

For more than two decades, Intel believed that a CPU, or central processing unit, like the ones that power desktop and laptop computers, could effectively handle the processing tasks needed to build and run AI models, according to four former Intel executives with direct knowledge. of the company’s plans.

Intel engineers viewed the graphics processing unit (GPU) video game chip architecture, which is used by rivals Nvidia and Advanced Micro Devices, in comparison “ugly,” one of the people said.

By the mid-2000s, however, researchers had found that gaming chips were more efficient than CPUs at handling the data collection needed to build and train large AI models. Because GPUs are designed for gaming graphics, they can perform a large number of calculations in parallel.

Nvidia engineers have spent years since then tweaking the architecture of GPUs to push them for AI use, and building the software needed to implement the capabilities.

“When AI hit … Intel just didn’t have the right processor at the right time,” said Lou Miscioscia, an analyst at Japanese investment bank Daiwa.

Nervana and Habana

Since 2010, Intel has made at least four attempts to produce a working AI chip, including two startup acquisitions and at least two major domestic efforts. Neither has made a dent against Nvidia or AMD in the fast-growing and lucrative market, according to three people with direct knowledge of the company’s inner workings.

Intel’s entire data center business is expected to generate sales of $13.89 billion this year – which includes the company’s AI chips but also many other designs – while analysts expect Nvidia to generate data center revenue of $105.9 billion.

In 2016, Intel CEO Brian Krzanich sought to buy a way into the AI ​​business by acquiring Nervana Systems for $408 million. Intel executives were attracted to Nervana’s technology, which is similar to the tensor processing unit (TPU) chip made by Google, according to two former executives.

The TPU – designed specifically for building, or training, large generative AI models – has stripped away the usual GPU features useful for video games and focused exclusively on increasing the AI’s calculations.

Nervana has enjoyed some success with customers installing Meta Platforms for its processor, though not enough to prevent Intel from switching horses and abandoning the project.

In 2019, Intel bought a second chip startup, Habana Labs, for $2 billion before shutting down Nervana’s efforts in 2020.

Krzanich did not respond to a request for comment for this article.

— Max A. Cherney, Reuters

Anna Tong and Arsheeya Singh Bajwa contributed to this report.