Intel Foundry spent years promising outside customers, and this week it finally got two of them in the headlines. On Wednesday, November 18, 2026, Microsoft and Intel announced a long-term agreement under which Microsoft will use Intel Foundry’s advanced EMIB packaging technology for the chips it designs for its own Azure cloud. On the same day, Intel chief executive Lip-Bu Tan told an audience in Austin, Texas, that the company had just signed its first external customer for 18A, its newest manufacturing process — without naming who it is.
Neither announcement settles Intel’s future on its own. Together, they are the clearest sign yet that a top cloud company is willing to put its name behind Intel’s attempt to become a serious contract chipmaker — securing packaging capacity and processor supply for Microsoft, and giving Intel the reference customer its foundry business has been missing.
What Microsoft and Intel actually announced
The joint announcement, published by Intel on November 18 and by Microsoft a day earlier, expands a collaboration that goes back decades. Its centrepiece is packaging: Microsoft is the first company to announce that it will use Intel Foundry’s EMIB — Embedded Multi-die Interconnect Bridge — advanced packaging technology for Azure first-party silicon, the chips Microsoft designs itself for its own cloud.
The agreement also confirms multi-year supply of Intel Xeon processors for Azure data centres, including instances powered by the next generation of server chips, codenamed Diamond Rapids. The two companies will also work together on CXL memory technology, Intel QuickAssist and confidential computing features for Azure.
Intel Foundry chief Naga Chandrasekaran said the deal validates its packaging capabilities at hyperscale; Microsoft called it part of its plan to build differentiated Azure infrastructure for growing cloud and AI demand.
Be precise about what kind of deal this is. Microsoft is committing to Intel’s packaging technology and to buying Intel processors — not, in this announcement, to having its leading-edge chips manufactured on Intel’s wafers. That distinction matters, because manufacturing, not packaging, is where Intel has struggled most and where TSMC still dominates.
What EMIB packaging is, and why it matters
Modern data centre chips are rarely built as a single piece of silicon. Instead, several smaller dies — the processor, the memory controllers, the input and output blocks — are joined together inside one package. How those dies are connected decides how fast the chip runs and how much it costs to make.
EMIB is Intel’s answer. Instead of laying the dies on top of one large silicon interposer, as TSMC does with its CoWoS packaging, EMIB embeds small silicon bridges inside the package’s organic substrate, exactly where a die-to-die connection is needed and nowhere else.
The advantage is efficiency. Intel says EMIB turns more than 90 percent of a packaging wafer into usable product, compared with roughly 60 percent for the interposer designs behind TSMC’s CoWoS-L packaging, which is fully booked. For a cloud operator buying chips by the thousands, that difference translates directly into cost and availability. Analyst Patrick Moorhead expects the packaged parts are Microsoft’s Cobalt CPU, its Maia AI accelerator, or both.
Intel’s 18A milestone: a first customer and 70 percent yields

The Microsoft deal arrived alongside a second piece of Intel news. Speaking at the TomorrowX Summit in Austin on Wednesday, November 19, chief executive Lip-Bu Tan said Intel had “just signed the first customer” for 18A, its most advanced manufacturing process. He did not name the customer. Intel’s chief technology officer, Pushkar Ranade, later clarified on X that the deal covers 18A-P, an improved variant that entered risk production in June. Because the customer is unnamed, the size and timing of the commitment are unknown.
Tan also gave a number Intel watchers have been waiting for. He said yields on 18A have reached about 70 percent, a sharp improvement from the state of the process when he took over as chief executive in March 2025, when he described the yields as “horrible”. Tom’s Hardware, which reported the remarks, noted that Tan did not say which kind of yield he meant.
Every point of yield improvement shows up in Intel’s margins, and in how seriously prospective foundry customers take the company. Shares in Intel rose in overnight and premarket trading after the two announcements.
Diamond Rapids and the Xeon supply deal

For Azure customers, the most concrete part of the announcement may be the least glamorous: guaranteed supply of Intel Xeon server processors for years to come, including machines built on Diamond Rapids, the next generation of Xeon chips.
Diamond Rapids is still a codename. Intel’s third-quarter 2026 earnings remarks put it at the sampling stage, with production set to ramp in 2027. Neither company disclosed core counts, power envelopes or launch dates, and the release gives no contract value, purchase volumes or exact term for the agreement.
What the companies did disclose is the engineering work that comes with it: collaboration on CXL memory, Intel QuickAssist offload and confidential computing, including Trust Domain Extensions, which encrypt a virtual machine’s memory and processor state as it moves between hosts. Vendor tests claim gains of up to 2.3 times for certain Azure virtual machines — marketing numbers with a “results may vary” caveat, describing existing machines, not the new EMIB chips.
Why each side needs the other
The deal makes sense for both sides because each has something the other is short of. Microsoft, like every hyperscaler, is trying to reduce its dependence on a single supplier for the chips that run its cloud. TSMC’s most advanced packaging capacity is fully booked, and the memory market is tightening fast — Micron said this week that 75 percent of its 2027 production is already sold. In that environment, a second source for advanced packaging is worth having, even before it ships a single chip.
Microsoft also designs its own silicon. At its Ignite 2026 event this week, the company announced Maia 300, its next custom AI accelerator, and made Cobalt 200 virtual machines generally available. Our report on the Microsoft Ignite 2026 announcements has the full list of what was unveiled in San Francisco. Chips that Microsoft designs itself still need someone to package them, and Intel Foundry is now officially in that picture.
For Intel, the value is credibility. Intel Foundry is the contract-manufacturing business Intel has built as the only United States-based leading-edge alternative to TSMC, and it has spent two years arguing that its packaging business could win outside customers before its leading-edge manufacturing nodes did. A public commitment from Microsoft is exactly the reference it has been missing. The same week, Anthropic priced the largest stock market debut in history at $83 a share — our report on the Anthropic IPO price has the details — and Anthropic has committed to nearly $900 billion of computing spending, every dollar of which eventually flows to someone who makes chips.
What the deal does not say
Honest reporting requires saying what is missing, and this announcement is missing a lot. There is no product name for the Azure chips that will use EMIB, no manufacturing process named, no volume, no dollar figure and no date when the first packaged parts will appear in Azure data centres. It is not even clear whether the EMIB-packaged silicon is the same custom data centre chip Microsoft committed to build on Intel’s 18A process back in 2024. Neither company has said.
The 18A customer is unnamed, so there is no way to judge how large that commitment is. The 70 percent yield figure comes from remarks at an event, not from a published Intel result, and the company has not said which kind of yield it means. None of this makes the announcements unimportant, but it is a step, not a finish line.
What it means for Azure customers and the chip industry
For businesses that run on Azure, the practical effects will arrive slowly: more predictable Xeon supply in the near term, and continued investment in CXL memory and confidential computing. The EMIB-packaged custom chips, whenever they arrive, should in theory deliver better performance per watt for the AI workloads behind the Copilot agents and Maia accelerators announced at Ignite.
For the wider chip industry, the deal tests whether Intel’s foundry strategy can work: win packaging customers first, then manufacturing customers, using its position as the only American company with leading-edge fabrication. If Microsoft’s commitment is followed by a second packaging customer and then by a manufacturing commitment on 14A, the node after 18A that reaches production in late 2027, the strategy will look credible. Analysts do not expect meaningful external wafer revenue until 2029 or 2030.
What happens next
Three things will turn this week’s headlines into numbers: the identity of the unnamed 18A customer and whether its volumes are large enough to matter; which Azure chip — Cobalt, Maia or something else — actually ships with EMIB packaging, and when; and whether Microsoft’s commitment extends to 14A wafers, moving the relationship into leading-edge manufacturing itself.
Intel’s next earnings report and Microsoft’s next Azure announcements are the dates to watch. Until then, the week’s news is best read as it is: Intel’s foundry business finally has a famous customer willing to say so in public, and a first manufacturing customer it is not yet willing to name. That is progress — but silicon, not press releases, will decide whether it is enough.
Sources: Intel newsroom, November 18, 2026; TrendForce, November 19, 2026.
