Neither Intel nor Nvidia have said exactly when the first fruits of its co-designed integrated CPUs will ship. But the thinking right now seems to be that it might a take a few years.
Nvidia announced a $5 billion investment into Intel last week, where Intel will supply CPU cores to Nvidia for potential use in the data center. In the PC, both Intel and Nvidia will collaborate on presumably mobile processors, where Nvidia will supply RTX chiplets for Intel to integrate, potentially upending the direction of GPUs.
Nvidia chief executive Jensen Huang told journalists last week that the partnership dates back a year, as per PCWorld’s reporting. Intel also told PCWorld that the partnership wouldn’t change its own roadmap, essentially adding on premium options to a number of product categories. But even with a head start the development work may take some time, the thinking goes, and the two sides won’t be prepared to talk about their efforts for some time.
Sources at competitors to Intel and Nvidia said that they expect the first products from the collaboration are more than two years away, and that they too are leaving their roadmaps unchanged as a result. One said that their company has doubts that Intel could work together with Nvidia to deliver the sort of complex, highly-integrated products both sides described.
“We’re creating an SOC [system-on-chip] that fuses two processors,” Huang said on a conference call with reporters last week. “It fuses the CPU and Nvidia GPU, RTX GPU, using NVLink and it fuses these two dies into one essentially virtual giant SOC, and that would become essentially a new class of integrated graphics laptops that the world’s never seen before.”
One issue is NVLink, Nvidia’s high-speed interface that can be used to combine the power of two Nvidia graphics card, at speeds higher than the PC’s backbone, PCI Express, achieves. A source at one competitor said that it has doubts that Intel has the engineering capabilities to make an integrated CPU-GPU with NVLink system-on-chip actually work, given Intel’s past history of engineering missteps dating from Arrow Lake’s poor desktop performance or the recent bugs that caused some processors to crash. They also wondered if Nvidia really cares to enable such a chip when its discrete GPUs already serve as a viable alternative.
Another source referenced a note from BofA Global Research, which worried about what role Softbank’s $2 billion investment into Intel might have on the development, as well as input from the U.S. government which has secured its own investment.
Such thinking is typically referred to as FUD, or “fear, uncertainty, and doubt,” a now fairly traditional means of criticizing one’s competitors in the technology industry. Still, this is a time where Intel’s dominance is seen as especially vulnerable, and AMD’s ongoing resurgence in desktop market share is evidence of that. Intel’s competitors would be especially eager to cut into Intel’s share in laptops, where Intel stubbornly holds on to about 80 percent of the market.
This week, rival Qualcomm is expected to unveil new Snapdragon mobile processors for laptops, hoping to cut into Intel. Still, shipments of Copilot+ PCs (which include Qualcomm’s Snapdragon processors) were just 2.3 percent of all Windows PCs sold during the first quarter of 2025, market researcher IDC reported earlier this year. Mercury Research reported that “growth in ARM in Copilot+ PCs also appeared to be at a standstill,” based on the firm’s estimates of the PC CPU market for the second quarter of 2025.
Dean McCarron, principal of Mercury Research, pointed out that Intel had worked together with AMD to develop the “Kaby Lake G” chip, announced in November 2017. In January 2018, Intel announced the “8th-gen Intel Core with Radeon RX Vega M graphics,” shipping the Core i7-8705G chip based on the partnership in June 2018. In PCWorld’s review of the processor, we noted that chip wasn’t tremendously exceptional, compared to the existing CPU + discrete GPU landscape, but that future iterations could have more impact. But only a handful of PC vendors built systems around the chip, and those future iterations never happened, possibly because it wasn’t quite clear which company was support the Kaby Lake G chip and its successors.
“I don’t think the challenge of adapting a GPU to a chiplet is a significant one, particularly for lower power graphics (which is what would likely be used for mobile designs),” McCarron said in an email.
Adapting Nvidia’s “Blackwell” architecture, the basis of its GeForce 5000 series of GPUs, wouldn’t be too long if Nvidia used a standard bus structure like PCI Express — two years maximum, with most of the time associated with CPU integration, packaging, and test. McCarron projected spring 2027 as a guess for when that could happen, though that assumed using a standard PCIe bus, not NVLink.
Intel is thinking of the new collaborative CPUs as a premium offering, and McCarron said he agreed with that.
“I would agree it’s probably a premium play, but probably not at the very top end,” McCarron wrote. “I could see this fitting well with the upper end of Core Ultra 5 and lower end of Core Ultra 7 [using Intel’s new naming scheme] especially in the thin and light segment of notebook.”
“Higher-end [Core Ultra 7] and [Core Ultra 9] would still go with separate GPUs for performance reasons,” McCarron added. “It would be reasonable to assume it’s going to be a normal Intel core with an extra chiplet rather than some new custom core just to support graphics integration, which points to re-using Panther Lake or Nova Lake,” the two Intel CPUs due in late 2025 and late 2026, respectively.
