Intel made some major announcements Monday that I believe are positive indicators for its turnaround efforts under CEO Pat Gelsinger—assuming it can execute flawlessly on them. All kinds of rumors have been flying, and last week I gave my best assessment of what’s happening with the company. Now there’s been a crucial board of directors meeting backing Gelsinger’s strategic moves, along with announcements about deals with AWS and the U.S. Department of Defense, plus internal reorganization steps.
All of these developments have put more wind into Intel’s sails. I’ve had a chance to digest these announcements—and to speak with Gelsinger directly before they were made public—so I want to share my findings here.
Note that Intel is a client of my firm, Moor Insights & Strategy, as are most of its competitors and major customers. This article conveys my independent analysis of its strategy and prospects.
Deal No. 1: AWS Gives Intel A Strategic Proof Point And Long-Term Reference Customer
The number-one cloud provider has signed a multiyear, multibillion-dollar “framework” deal with Intel for wafers and custom Xeon 6 chips. The agreement extends across multiple Intel production nodes including the forthcoming 18A, 18AP and 14A. In fact, it ties AWS to upwards of five years on 14A, which isn’t even finished yet—an important commitment.
First out will be a custom Xeon 6 processor on the Intel 3 node. In my conversation with Gelsinger, he couldn’t tell me exactly what made it custom, but he did say it was beyond what it does today, with custom voltages and frequencies. My guess is that it integrates some of AWS’s networking and AI IP blocks. Later on, an 18A-based “AI fabric” chip might support AWS’s own Trainium and Inferentia AI chips and possibly a heterogeneous AWS accelerator as well as Nvidia and AMD GPUs. That would be very consistent with what AWS has done with its Elastic Fabric Adapter.
This agreement makes sense to me on a couple of levels. First, AWS and the other cloud providers all want chip customization to meet their very specific datacenter needs. Intel can help with this, and the two partners could do all kinds of creative things through a co-development approach that takes advantage of Intel’s chiplet architecture. By adding a custom Xeon 6, AWS ostensibly has custom chips on Arm (Graviton) and now x86.
Second, AWS represents a major customer—and a great reference customer—for Intel Foundry. This of course means money coming in the door, but beyond that AWS will be Exhibit A showing that Intel Foundry can support some of the biggest companies around. Gelsinger also cites the deal as a proof point for the strategic expansion of x86 technology in the datacenter.
Overall, this can only be looked at as a positive—a strategic, long-term deal that Intel didn’t have before. If you can get the number-one cloud provider to do not just one but two highly custom chips with you, one in the foundry and one a custom server chip, that portends a very good future. This also adds AWS to Microsoft as committed to Intel’s leading-edge 18A process. That’s another significant positive.
Naysayers will ask, “What about Inferentia, Trainium and Graviton? Why isn’t Intel Foundry making those?” (One part of the answer: AWS already has long-term production arrangements with TSMC for those chips.) Optimists will say, “You have to walk before you can run, and this is the onramp for the AI ASICs and Graviton.”
In this instance, I side with the optimists. Xeon 6 with custom networking and AI blocks (my guess for what’s coming) is extremely strategic. In fact, it’s what I would have expected AMD to propose with a semi-custom offering—yet here’s Intel stepping up to do it.
Deal No. 2: The Department of Defense Signs On For ‘Secure Enclave’
This classified project, funded by the CHIPS and Science Act, is worth up to $3 billion to Intel over the coming years. That’s a tidy sum, sure, and it will roll in sooner than later. But more importantly, this is another win for Intel Foundry, and even more so for the highly secure supply chain that Gelsinger has been intent on building for Intel over the past few years.
This extends on work done through SHIP, RAMP-C and other secret programs, I’m guessing with backing from DARPA, CIA and NSA. (Let me repeat: only my educated guess.) SHIP gave DoD access to advanced packaging tech, and RAMP-C is focused on custom chip wafers for DoD systems and important defense suppliers including Boeing, Northrop Grumman, Microsoft and IBM.
Because the specifics of this deal are classified, there are no details about what Intel is getting paid to do, and Gelsinger could not comment on it to me, even under NDA. Just thinking out loud, this could be wafers for DoD, NSA and DARPA designs; it could be packaging; it could be making sure the supply chains of other companies (Nvidia?) are secure. We’re not supposed to know. I’m thinking that this is separate from Intel’s $8.5 billion CHIPS Act deal announced earlier this year.
Intel Foundry Will Do A ‘Spin-In’
My Forbes article last week focused on the irrationality of spinning Intel Foundry out now—with a focus on the “now.” This week we found out that Intel plans to spin it in. The company is setting up Intel Foundry as an independent subsidiary of Intel. It will have its own operating board and independent directors, although the current leadership team remains in place and still reports to Gelsinger.
The CEO believes that this structure will bring a few significant benefits. First, it will provide greater clarity, transparency and accountability for external foundry customers. While Gelsinger believes that the foundry and products businesses are “better together,” he also told me he wants external foundry customers to “have absolute confidence that this is their factory” thanks to independent oversight. His goal is for all foundry customers to “look at [Intel Foundry] as their balanced supply chain for the future.”
Second, the subsidiary arrangement will give Intel flexibility down the road to pursue independent sources of funding for Intel Products and Intel Foundry as they are increasingly treated as separate businesses. For understandable reasons, Gelsinger and the Intel board are paying close attention to capital structure and what he calls an “efficient CapEx environment.”
I pressed him on this point: If spinning in Foundry is a good idea, why wait until now? Why not do it six months or a year ago?
He made the point that when he started as CEO a few years ago, the two businesses were “highly commingled.” It’s been a slow process of “teasing them apart,” for example by putting Foundry and Products on separate ERP systems. He summarized the challenge by saying, “This is a 55-year-old, integrated, singular company. . . . If it could have been a day faster, I would have loved it to be a day faster, but I’m not sure, even at the pace we’re on, that we could accelerate it any faster than we are now.”
As for how external foundry customers will regard this separation, here’s what I think: If by installing an independent board, operating structure and bylaws Intel can make Foundry appear truly independent to the contrarians, that will likely make a big difference to potential customers including Qualcomm, Broadcom, Apple and maybe even AMD. The independent structure could be more inviting to lead customers like these that are wary of Intel competing with their designs. And Intel’s positioning will be even more important if the U.S. government starts requiring chip design and manufacturing to be carried out by U.S. companies for critical tech infrastructure, and not just defense. The degree of success of the Foundry spin-in in the eyes of AMD, Apple, Broadcom and Qualcomm will rely first on the tech, but then on the perception of strict governance determined by the new corporate bylaws, including voting, operating agreements and the independence of the directors. We will see…
Slowing Expansion Projects In Germany, Poland And Malaysia
This one is more straightforward. To better control its capital expenditure and match its manufacturing investments to current demand, Intel is pausing its fab expansion projects in Germany and Poland by two years, relying in the meantime on the recent expansion of capacity in its Ireland fab for European production. Meanwhile, it will slow-roll construction on its advanced packaging facility in Malaysia to align its completion with market conditions. Investments in new U.S. manufacturing capacity in Arizona, New Mexico, Ohio and Oregon are continuing unchanged.
This rationalization of capital investments is pragmatic—and should have an immediate impact on the company’s finances in terms of debt, cash and operating expenses. Intel is responding to the realities of Euro government funding of technology, global economic uncertainties and its own need to deliver the $10 billion of cost savings it promised investors when it announced its restructuring efforts last month.
Realignment Of Intel Business Units
In the spirit of “simplicity,” Intel is rearranging some of its business units. Its edge computing and automotive businesses are being rolled into the Client Computing Group, which continues to be managed by Michelle Johnston Holthaus. The silicon photonics unit is being moved into the Data Center and AI Group led by Justin Hotard. Its Network and Edge business unit, often referred to as NEX, continues under Sachin Katti with a greater focus on networking and telecom. And the software and incubation businesses—I assume including quantum and neuromorphic R&D—are being integrated in these core business units to streamline things.
Honestly, I’m not sure how much these moves simplify things other than for Gelsinger and the board. I do agree that edge and automotive fit better under CCG than DCAI, if they must be moved. Putting SiPho in DCAI makes a ton of sense as Intel is “Customer Zero.” And the focus of NEX seems good to me.
Could the new additions distract CCG and DCAI leadership? I don’t know. Maybe moving complementary businesses together will indeed make for “a simpler, nimbler organization,” as Gelsinger put it. Time will tell.
A Clear Affirmation Of Intel’s Path Forward
At the start of our call, Gelsinger put his cards on the table when he said, “Obviously, we’ve had a noisy, challenging environment” since Intel announced its big restructuring plan on August 1. From my perspective, what he talked about on our call and what Intel then announced publicly was exactly what I wanted to hear. I suspect it will also dampen some of the noise.
That said, everyone at Intel, from Gelsinger on down, knows they have their work cut out for them. I talk regularly with all of Intel’s big customers, and the general attitude—even for Intel’s supporters—is that the company is out of excuses and simply must execute like hell from here on out.
Based on these announcements and the legitimately strong work Intel has done implementing its “five nodes in four years” plan, the groundwork has been laid. Clearly the company is taking steps to be more cost-efficient and competitive in the shorter and longer term. That said, it also needs to get through a successful reduction in force of 15,000-plus people, and that’s not easy, because it means Intel has to change the way it works. It’s not just selling off subsidiaries or exiting a business, but taking layers out. And I believe that could increase execution risks.
Quick yet successful large-scale production of processors for Copilot+ PCs and Intel’s new datacenter lineup could help customer confidence. Signing up AWS and more DoD on top of Microsoft for 18A should be a confidence booster, too, albeit delivery won’t come until 2025 and 2026. Intel has to achieve a lot more than that, including active production on the 18A and 14A nodes and the introduction of datacenter low-precision AI GPUs (again in 2025 or 2026) that can take market share from Nvidia and AMD in the datacenter. I could go on, but the underlying message is clear: Intel needs to build on the moves announced this week to rack up wins and then rack up more wins—and keep on doing it until this period of its history is way in the rearview mirror.
Taking a long-term view, the moves this week have reduced risk and should increase confidence for Intel’s investors and customers. Now we need to see Intel execute flawlessly—remorselessly—like it used to.
