Nvidia, Intel to co-develop “multiple generations” of chips as part of $5 billion deal

In a major collaboration that would have been hard to imagine just a few years ago, Nvidia announced today that it was buying a total of $5 billion in Intel stock, giving Intel's competitor ownership of roughly 4 percent of the company. In addition to the investment, the two companies said that they would be co-developing "multiple generations of custom data center and PC products."

"The companies will focus on seamlessly connecting NVIDIA and Intel architectures using NVIDIA NVLink," reads Nvidia's press release, "integrating the strengths of NVIDIA’s AI and accelerated computing with Intel’s leading CPU technologies and x86 ecosystem to deliver cutting-edge solutions for customers."

Rather than combining the two companies' technologies, the data center chips will apparently be custom x86 chips that Intel builds to Nvidia's specifications. Nvidia will "integrate [the CPUs] into its AI infrastructure platforms and offer [them] to the market."

On the consumer side, Intel plans to build x86 SoCs that integrate both Intel CPUs and Nvidia RTX GPU chiplets—Intel's current products use graphics chiplets based on its own Arc products. More tightly integrated chips could make for smaller gaming laptops and could give Nvidia a way to get into handheld gaming PCs like the Steam Deck or ROG Xbox Ally.

On a conference call with Nvidia CEO Jensen Huang and Intel CEO Lip-Bu Tan this afternoon, the CEOs said that the technical collaboration between teams at Nvidia and Intel had been ongoing for nearly a year, though neither company was ready to announce any products or get specific about launch windows. Chip designs can sometimes take years to come together, so it could still be months or years before the first products to come out of this partnership are actually available to buy.

Huang estimated that the partnership would be worth between $25 and $50 billion of "annual opportunity," once the products begin shipping, though it wasn't clear how much of that money would show up on Nvidia's balance sheet and how much would show up on Intel's.

A dramatic reversal

Back in 2005, Intel considered buying Nvidia outright for “as much as $20 billion,” according to The New York Times. At the time, Nvidia was known almost exclusively for its GeForce consumer graphics chips, and Intel was nearing the launch of its Core and Core 2 chips, which would manage to win Apple’s business and set it up for a decade of near-total dominance in consumer PCs and servers.

But in recent years, Nvidia’s income and market capitalization have soared on the strength of its data center chips, which have powered most of the AI features that tech companies have been racing to build into their products for years now. And Intel’s recent struggles are well-documented—it has struggled for years now to improve its chip manufacturing capabilities at the same pace as competitors like TSMC, and a yearslong effort to convince other chip designers to use Intel’s factories to build their chips has yielded one ousted CEO and not much else.

The two companies' announcement comes one day after China banned the sale of Nvidia's AI chips, including products that Nvidia had designed specifically for China to get around US-imposed performance-based export controls. China is pushing domestic chipmakers like Huawei and Cambricon to put out their own AI accelerators to compete with Nvidia's.

Correlation isn't causation, and Intel and Nvidia didn't throw together a $5 billion deal and product collaboration in the space of less than 24 hours. But Nvidia could be looking to prop up US-based chip manufacturing as a counterweight to China's actions.

There are domestic political considerations for Nvidia, too. The Trump administration announced plans to take a 10 percent stake in Intel last month, and Nvidia's Huang has worked to curry favor with the Trump administration by making appearances at $1 million-per-plate dinners at Trump's Mar-a-Lago golf course and promising to invest billions in US-based data centers.

Although the US government's investment in Intel hasn't gotten it seats on the company's board, the investment comes with possible significant downsides for Intel, including disruptions to the company's business outside the US and limiting its eligibility for future government grants. Trump and his administration could also decide to alter the deal for any or no reason—Trump was calling for Tan's resignation for alleged Chinese Communist Party ties just days before deciding to invest in the company instead. Investing in a sometime-competitor may be a small price for Nvidia and Huang to pay if it means avoiding the administration's ire.

Huang said on the joint conference call that the Trump administration had been told about the deal but that it wasn't involved in talks between the two companies.

"The Trump administration had no involvement in this partnership at all," said Huang. "They would have been very supportive, of course. Today I had the opportunity to tell Secretary [of Commerce Howard] Lutnick, and he was excited, very supportive of seeing American technology companies working together."

Outstanding questions abound

Combining Intel CPUs and Nvidia GPUs makes a lot of sense for certain kinds of products—the two companies' chips already coexist in millions of gaming desktops and laptops. Being able to make custom SoCs that combine Intel's and Nvidia's technology could result in smaller and more power-efficient gaming PCs. It could also provide a counterbalance to AMD, whose willingness to build semi-custom x86-based SoCs has earned the company most of the emerging market for Steam Deck-esque handheld gaming PCs, plus multiple generations of PlayStation and Xbox console hardware.

But there are more than a few places where Intel's and Nvidia's products compete, and at this early date, it's unclear what will happen to the areas of overlap.

For example, Intel has been developing its own graphics products for decades—historically, these have mostly been lower-performance integrated GPUs whose only job is to connect to a couple of monitors and encode and decode video. But more recent Arc-branded dedicated graphics cards and integrated GPUs have been more of a direct challenge to some of Nvidia's lower-end products.

Intel told Ars that the company "will continue to have GPU product offerings," which means that it will likely continue developing Arc and its underlying Intel Xe GPU architecture. But that could mean that Intel will focus on low-end, low-power GPUs and leave higher-end products to Nvidia. Intel has been happy to discard money-losing side projects in recent years, and dedicated Arc GPUs have struggled to make much of a dent in the GPU market.

On the software side, Intel has been pushing its own oneAPI graphics compute stack as an alternative to Nvidia's CUDA and AMD's ROCm and has provided code to help migrate CUDA projects to oneAPI. And there's a whole range of plausible outcomes here: Nvidia allowing Intel GPUs to run CUDA code, either directly or through some kind of translation layer; Nvidia contributing to oneAPI, which is an open source platform; or oneAPI fading away entirely.

On Nvidia's side, we've already mentioned that the company offers some Arm-based CPUs—these are available in the Project DIGITS AI computer, Nvidia's automotive products, or the Nintendo Switch and Switch 2. Nvidia is also said to be working on some Arm products that haven't been announced yet, including Arm-based chips for Windows PCs that it's said to be co-developing with MediaTek.

Huang said that the partnership with Intel would have no bearing on its Arm products and that development on those products would continue.

"We're fully committed to the Arm roadmap," Huang said, mentioning a number of upcoming products, including future generations of its Vera CPU architecture.

Finally, there's the question of where these chips will be built. Nvidia's current chips are manufactured mostly at TSMC, though it has used Samsung's factories as recently as the RTX 3000 series. Intel also uses TSMC to build some chips, including its current top-end laptop and desktop processors, but it uses its own factories to build its server chips and plans to bring its next-generation consumer chips back in-house.

Will Nvidia start to manufacture some of its chips on Intel's 18A manufacturing process or another process on Intel's roadmap? A vote of confidence from Nvidia would be a big shot in the arm for Intel's foundry, which has reportedly struggled to find major customers—but it's hard to see Nvidia doing it if Intel's manufacturing processes can't compete with TSMC's on performance or power consumption, or if Intel can't manufacture chips in the volumes that Nvidia would need.

Huang didn't close the door to working with Intel on manufacturing but responded to multiple questions about it by heaping praise on TSMC, suggesting that the manufacturer was a known quantity that Nvidia was in no rush to stop working with.

"We've always evaluated Intel's foundry technology, and we're going to continue to do that," said Huang. "I think Lip-Bu and I would both say that TSMC is a world-class foundry, and in fact, we're both very successful customers of TSMC's. The capabilities of TSMC, from process technology, their rhythm of execution, the scale of their capacity and infrastructure, the agility of their business operations... just all of the magic that comes together for being a world-class foundry, supporting customers of such diverse needs. I just can't overstate the magic that is TSMC."

This article was updated to add details from a conference call with Nvidia CEO Jensen Huang and Intel CEO Lip-Bu Tan.