Analysis: After Big Nvidia Win, Will Intel Ever Escape Its Rival’s Shadow?

While the Intel-Nvidia deal will help drive CPU sales for Intel, the agreement will nevertheless keep the CPU in the back seat as the GPU drives greater interest and benefit for Nvidia—a point that Nvidia CEO Jensen Huang emphasized in remarks last Thursday.

Nvidia CEO Jensen Huang has come a long way from viewing Intel as an existential threat to his company and calling for his employees to “go kill Intel” back in 1997.

As recalled in last year’s book “The Nvidia Way” by reporter Tae Kim, Huang was responding then to Intel, armed with significantly more resources at the time, announcing the i740 graphics chip with double the frame buffer over the Riva 128 processor it put out months before. Nvidia, only four years old then, had been on very fragile financial footing.

Fast forward roughly 28 years later to last week, and the two companies are in radically different positions: Nvidia’s annual revenue and profitability have far surpassed Intel’s, giving it a big enough war chest to invest $5 billion in Intel common stock. This growth has been fueled by its full-stack computing platform that has been widely adopted for AI workloads, giving it the cachet to land a joint development deal with Intel—now struggling in part because of its failure to execute a successful AI strategy—that keeps Nvidia’s GPUs at the center.

Nvidia and Intel have called the investment and co-development deal, announced last Thursday, a “historic” agreement that will, in Huang’s words, expand the market for Intel “very significantly” in addition to expanding Nvidia’s opportunities.

The news sent Intel’s struggling stock price sky high that day, with the company’s shares growing by 22.8 percent. Meanwhile, Nvidia—whose market capitalization was roughly 27 times larger than its rival’s at $4.3 trillion as of Thursday afternoon this week—saw its stock price grow by 3.5 percent last Thursday.

While the deal is a meaningful development for Intel, particularly as CEO Lip-Bu Tan seeks to stabilize Intel’s waning CPU market share—particularly on the server side—it validated how foundational Nvidia has become to the modern technology world, where the concept of accelerated computing has gripped developers big and small, particularly for AI.

Deal Creates x86 Option For Nvidia Rack-Scale Platforms

In the lead up to the agreement, Intel had fought hard to achieve the status of host CPU provider for recent Nvidia platforms such as the DGX B300, for which it designed a custom Xeon chip. This is noteworthy considering the fierce competition Intel has faced from AMD, which won the socket for an earlier DGX design before ceding the space to its rival.

Now, as part of the deal, Intel is designing a custom CPU to serve as the host for an x86 version of Nvidia’s power-hungry rack-scale computing platforms, which have traditionally run on custom Arm-based CPUs. These platforms provide the fastest possible computing performance in a server rack by using Nvidia’s NVLink interconnect technology to form high-speed connections between many GPUs—a total of 72 for the latest offerings.

What Intel and Nvidia are essentially doing here is giving customers more options for the kind of Nvidia platforms they want to power AI workloads. Currently, if customers wanted the most powerful Nvidia platform right now, they would have to choose the GB300 NVL72, whose host CPU is Nvidia’s Arm-based Grace chip. In the future, customers will have the option for an Nvidia platform with what will likely provide a similar performance level but instead use a host CPU from Intel that is based on the x86 instruction set architecture.

Nvidia is doing this because many enterprise customers continue to prefer x86, a legacy architecture that has long served as the underpinnings of their computing environments, including data center infrastructure.

“The vast majority of the world’s enterprises [are] still x86-based. They now have state-of-the-art AI infrastructure,” Huang said last Thursday.

Intel Opens Big, New PC Market For Nvidia

The second part of the Intel-Nvidia deal covers the other area where enterprises have great preference for x86 over Arm: PCs. As part of the agreement, Intel will design a system-on-chip for laptops that will fuse an x86-based Intel CPU chiplet with an Nvidia RTX GPU chiplet, bringing the two companies on the same silicon package for the first time.

Huang said that this jointly developed product will create a “new class of integrated graphics laptops,” exposing Nvidia to a major segment of the laptop “that has been largely unaddressed” by his company and sees roughly 150 million shipments every year.

This will mark a major expansion for Nvidia in the laptop market because the company has traditionally focused on providing discrete GPUs for gaming laptops and mobile workstations, leaving integrated graphics mostly to Intel and AMD.

Like the data center side of the deal, the PC side will put Nvidia at the center of appeal for future computers that use these jointly developed products. That’s because of the significant reputation Nvidia has built for itself as the computing platform of choice for developers creating AI and machine learning applications.

While this has largely benefited Nvidia in the data center market, the company has been investing in software and hardware to unleash a wave of new AI and machine learning capabilities with its RTX GPUs for PCs. The company made clear last year that it sees this as a big opportunity by labeling RTX-powered computers as “RTX AI PCs.”

With the new deal, Nvidia is getting the opportunity to significantly expand the number of people who can take advantage of its RTX GPUs for PCs to accelerate a growing number of AI-powered applications—or develop their own applications using Nvidia’s NIM microservices, which works with several AI developer tools.

How The Deal Could Feed Into Nvidia’s ‘Virtuous Cycle’

This could all feed into what Huang has called the “virtuous cycle” of Nvidia’s CUDA parallel computing platform, which is the foundational programming layer of many applications that take advantage of Nvidia GPUs. This cycle has been critical to Nvidia’s success, allowing it to attract over 6 million developers and more than 27,700 startups to design GPU-bound programs.

The cycle starts with Nvidia’s install base, which counts the total number of devices—including PCs, servers and edge devices—powered by Nvidia GPUs. The idea put forward by Huang is that by selling more Nvidia-powered devices, the company can increase its investments in research and development. This speeds up the performance of applications and, in turn, reduces the costs of computing. As a result, developers can unleash new capabilities, further increasing demand for Nvidia-powered devices.

“Install base is growing, computing costs [are] coming down, which causes more developers to come up with more ideas, which drives more demand, and now we’re in the beginning of something very, very important,” Huang said at last year’s Computex event.

This means that by enabling new x86 options for Nvidia’s rack-scale platforms and a new wave of mainstream laptops with Nvidia integrated graphics, Intel is playing a bigger role than before to help Nvidia grow the rival’s install base and aid with the tech industry’s ongoing transition from CPU- to GPU-bound workloads that Huang sees as inevitable.

While the deal will help drive CPU sales for Intel, the agreement will nevertheless keep the CPU in the back seat as the GPU drives greater interest and benefit for Nvidia—a point that Huang emphasized in his opening remarks discussing the partnership last Thursday.

“General-purpose computing has reached its limits. To keep advancing, we invented a new way forward. Nvidia pioneered GPU-accelerated computing, increasing performance by orders of magnitude—tens, hundreds, thousands of times faster—while dramatically improving energy and cost efficiency,” he said.

Deal Keeps Intel’s Rival Strategy Out Of The Spotlight

The deal comes with knock-on effect for Intel: In reinforcing Nvidia’s dominance, it keeps the company’s competing accelerated computing strategy out of the spotlight, putting some extra hurdles between Intel and any virtuous cycle that could take hold should its accelerator chip products ever catch on like they have for Nvidia.

The company has struggled significantly over the past several years in defining and executing its accelerated computing strategy on the data center side. After failing to meet its modest revenue goal of $500 million for its line of Gaudi accelerator chips last year, the company scrapped a successor product originally due this year to focus on a next-generation GPU code-named “Jaguar Shores” designed for rack-scale platforms further in the future.

Outside of Jaguar Shores, the company has an additional, unannounced GPU design with a lower power requirement for servers on its road map that could arrive next year at some point, according to a source familiar with the company’s plans.

Intel has experienced some success in the PC market with its Xe GPU architecture. While most of this success has come from Xe’s integrated graphics prowess, Intel recently eked out a win on the discrete GPU side with the Intel Arc Pro B50 reportedly becoming the best-selling workstation GPU at online retailer Newegg earlier this month.

(There is an exception to these concerns about Nvidia taking the accelerated computing spotlight from Intel in this deal. It’s not yet known if Intel’s custom system-on-chip with Nvidia will feature its neural processing unit (NPU), for which the company has directed substantial resources to attract developers as part of its AI PC strategy. If the NPU is involved, Intel could benefit from these broader investments and grow developer mind share in this area)

An Intel spokesperson told CRN in a statement last Friday that the company remains “committed” to its GPU road map, adding that “everything we discussed aligns with and complements Intel’s existing strategy.”

“We’ll be collaborating with Nvidia to serve specific market segments, but we’re also continuing to execute on our own path,” the representative said, adding that the company does not comment on rumors regarding unannounced products on its road map.

While Intel will likely continue down this path in Nvidia’s shadow for the years to come, the company’s ability to win custom chip design projects with a company as significant as Nvidia shows the opportunity it faces down an adjacent path, where Tan is making moves for Intel to become a substantial player in custom chip design services.

Between this and the wait-and-see status of its push to gain a major customer for the Intel Foundry contract chip manufacturing business, the semiconductor giant could eventually find itself in a different position than Nvidia and cast a shadow over others with a new shape. But future success isn’t guaranteed, so down these paths Intel will continue.