한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Key Points

Tencent Technology News, August 8, according to foreign media reports, as a former leader in the computer field, Intel is now facing many challenges. For this American chip giant, if it had made the right choice seven years ago, its fate would have taken a significant turn in the era of artificial intelligence.

$1 billion in cash for 15% equity?

Seven years ago, Intel faced a critical opportunity: to buy a stake in OpenAI, a fledgling nonprofit research organization that was exploring a little-known frontier field called generative artificial intelligence, according to multiple people familiar with the matter.

Three people familiar with the matter said that between 2017 and 2018, executives of both parties discussed several cooperation plans in depth, including Intel spending $1 billion in cash to acquire a 15% stake in OpenAI. In addition, if OpenAI can purchase Intel hardware at cost price, Intel will increase its stake by an additional 15%.

However, the deal ultimately fell through, in part due to Intel's then-CEO Bob Swan, who said:Generative AIIt will take some time for the model to be commercialized and it is unlikely to bring returns to chip manufacturers in the short term.

At the same time, although OpenAI is very interested in cooperating with Intel, hoping to reduce its dependence on Nvidia chips and build its own infrastructure, Intel's data center department is opposed to this, especially the refusal to provide products at cost price, which has become another important reason for the breakdown of the transaction.

An Intel spokesperson declined to comment on the potential deal. Swan also did not respond. OpenAI also declined to comment.

Intel has never disclosed its decision to abandon its investment in OpenAI. Now, OpenAI has launched a chatbot in 2022.ChatGPT, its valuation has soared to $80 billion.

According to nine people familiar with the matter, including former Intel executives and industry experts, this is just one of a series of strategic mistakes made by Intel in the field of computer chips from the 1990s to the early 21st century, and its development in the era of artificial intelligence has been significantly hindered.

Last week, Intel's second-quarter earnings report caused its stock price to plunge more than 25%, marking its worst single-day performance since 1974. The chip giant's market value fell below the $100 billion mark for the first time in more than 30 years, and it now faces the daunting challenge of bringing artificial intelligence chip products to market.

By comparison, Nvidia, with a market cap of $2.6 trillion, has moved away from gaming graphics cards to focus on the AI ​​chips needed to build, train and run large-scale generative AI systems such as OpenAI’s GPT4 and Meta’s Llama, and is well ahead, while Intel lags behind AMD, with a market cap of $218 billion.

When asked about its progress in artificial intelligence, an Intel spokesperson cited the latest views of CEO Pat Gelsinger, who emphasized that the third-generation Gaudi artificial intelligence chip, which is to be released in the third quarter of this year, will surpass market competitors.

Gelsinger also revealed that Intel has won the favor of "more than 20" customers for the second-generation and third-generation Gaudi chips, and announced that the next-generation Falcon Shores artificial intelligence chip will be available at the end of 2025.

“We are accelerating the pace of innovation in design and process technology, and the product pipeline we have built gives us confidence that we will occupy a more prominent position in the future competition in the artificial intelligence market,” the spokesperson said.

Gaming chips sweep AI

In terms of OpenAI, Microsoft made new investments in 2019 to seize the initiative in the era of artificial intelligence. This move is even more important against the backdrop of the release of ChatGPT in 2022 and the active deployment of artificial intelligence by large companies around the world.

Several former executives and industry experts pointed out that for Intel, the aborted potential deal means that it has missed a key opportunity. It has gradually lost ground in the battle for supremacy in artificial intelligence over the past decade.

Dylan Patel, founder of semiconductor research firm SemiAnalysis, pointed out: "Intel's failure in the field of artificial intelligence is rooted in its failure to present a unified and powerful product strategy to the market."

Intel has long believed that CPUs, similar to those in desktops and laptops, are efficient enough to handle the computing demands needed to build and run artificial intelligence models, according to four former executives familiar with its plans.

According to a person familiar with the matter, Intel engineers had reservations about the GPU gaming chip architecture used by competitors Nvidia and AMD, considering it "not elegant enough". However, around 2005, research revealed that gaming chips were far more efficient than CPUs in handling the huge data processing tasks required for building and training artificial intelligence models. This is due to the fact that GPUs are designed specifically for gaming image processing and can handle massive calculations in parallel.

Since then, Nvidia engineers have spent several years improving the GPU architecture to make it more suitable for artificial intelligence applications and developing corresponding software. Lou Miscioscia, an analyst at Japan's Daiwa Bank, pointed out: "Faced with the wave of artificial intelligence, Intel failed to launch the right processor at the right time."

Intel has tried at least four times since 2010 to develop competitive AI chips, including acquiring two startups and at least two large local companies, but its efforts have yet to mount a significant challenge to Nvidia or AMD in the rapidly expanding and lucrative market, according to three people familiar with the matter.

Intel's full-year data center business, which includes its artificial intelligence chips and other designs, is expected to generate $13.89 billion in revenue, compared with analysts' forecasts of $105.9 billion for Nvidia.

In 2016, Intel, under the leadership of then-CEO Brian Krzanich, attempted to acquire Nervana Systems for $408 million in an effort to accelerate its AI business. According to two former executives, Nervana’s technology attracted Intel because of its design, which is similar to Google’s TPU (tensor processing unit) chip, which is tailored for building and training large generative AI models, removing the functions of traditional GPUs that serve video games and focusing on optimizing AI computing.

Although Nervana's processors have achieved some success among customers including Meta, Intel ultimately decided to adjust its strategy and acquired Habana Labs for $2 billion in 2019 and terminated the acquisition of Nervana in 2020. Krzanich did not respond to a request for comment. (Compiled by Jinlu)