Microsoft picks Intel Foundry's EMIB packaging for its own Azure AI chips
Three weeks after weak guidance cut more than 20% off Intel's shares, the foundry push has its first named customer. The wafers that would justify Intel's capital spending arrive in 2028–29 at the earliest, so holders are still betting on execution.
Richard Tangthree-minute read
A first named customer, with no chip, volume or terms
Microsoft+1.22% — Microsoft, up 1.22 percent today and Intel+0.83% — Intel, up 0.83 percent today announced on 18 November 2026 a long-term agreement under which Microsoft is "the first company to announce use of Intel Foundry's EMIB advanced packaging" for Azure first-party silicon, the chips Microsoft designs for its own cloud 2. Intel Foundry is the contract-manufacturing business Intel has built as the only US-based leading-edge alternative to TSMC+1.99% — TSMC, up 1.99 percent today 5.
The agreement also confirms multi-year Xeon supply capacity, including the next-generation Diamond Rapids server CPU, plus joint work on CXL (Compute Express Link) memory, Intel QuickAssist acceleration and TDX confidential-computing features in Azure 2. Neither company named the chip, the volumes, the terms or the timing 2.
Analyst Patrick Moorhead, who flagged the deal, expects the packaged parts are the Cobalt CPU, the Maia AI accelerator, or both, and stresses the commitment covers EMIB rather than the newer EMIB-T 1.
Packaging capacity has become the AI bottleneck
EMIB embeds small silicon bridges in a package's organic substrate so several chip dies can talk to each other at high bandwidth, with no full silicon interposer underneath 4. EMIB-T adds through-silicon vias, vertical connections running through the bridge, and supports HBM4, the next generation of stacked high-bandwidth memory 34.
The economics matter at AI scale. Intel says EMIB turns more than 90% of a packaging wafer into usable product, against roughly 60% for the interposer designs behind TSMC's CoWoS-L, which is fully booked 4.
Google, Amazon+4.72% — Amazon, up 4.72 percent today and MediaTek are evaluating Intel's packaging for the same reason 4. A customer does not need Intel to fabricate its dies to buy Intel packaging 4, and Moorhead calls packaging "a smaller commitment than wafers" that goes straight at a real bottleneck 1.
The follow-through Tan promised, at the worst moment
Intel chief executive Lip-Bu Tan said on the 29 October earnings call that a couple of foundry customers would be ready to announce within two quarters 5. Three weeks later, here is the first, announced the same day Tan disclosed a first, unnamed external customer for 18A, Intel's newest manufacturing process, now at 70% yield 6.
The timing lands hard. Intel shares fell more than 20% after that call, where the company guided fourth-quarter revenue to $15–16 billion against the $17.1 billion analysts expected 5. Finance chief David Zinsner's rule, that Intel "will not spend $1 of capital" without clear evidence of customer demand, is what makes a named customer matter beyond the headline 5.
Revenue climbed back to $16.5B; Intel's own Q4 guide points down
The first packaging customer lands three weeks after weak guidance cut the shares
- Revenue
- Estimate
Data
| Revenue | |
|---|---|
| Q4 '24 | $14.26B |
| Q1 '25 | $12.67B |
| Q2 '25 | $12.86B |
| Q3 '25 | $13.65B |
| Q4 '25 | $13.67B |
| Q1 '26 | $13.58B |
| Q2 '26 | $16.13B |
| Q3 '26 | $16.51B |
| Q4 '26 (guide) (estimate) | $15.5B ($15–16B) |
One name does not make a foundry
For Microsoft, the Xeon half is the near-term money. Intel's data-center revenue grew 78% to $7.3 billion in the third quarter while Xeon supply remained the constraint 57, and Moorhead's case is that AI agents need CPUs 1. Multi-year capacity including Diamond Rapids is supply insurance for Azure's roadmap 2.
For Intel holders, the open question is wafers. The 14A process, the node after 18A, reaches production in late 2027 with external customers expected in 2028–29 6, and Moorhead puts meaningful external wafer revenue in 2029 or 2030 1.
Capex has run below operating cash in four of the last six quarters
Any foundry payoff has to come out of a smaller capex budget than 2024's
- Capital spending
- Operating cash flow
Data
| Capital spending | Operating cash flow | |
|---|---|---|
| Q4 '23 | $6.7B | $4.62B |
| Q1 '24 | $5.97B | -$1.22B |
| Q2 '24 | $5.68B | $2.29B |
| Q3 '24 | $6.46B | $4.05B |
| Q4 '24 | $5.83B | $3.17B |
| Q1 '25 | $5.18B | $0.81B |
| Q2 '25 | $3.55B | $2.05B |
| Q3 '25 | $2.43B | $2.55B |
| Q4 '25 | $3.49B | $4.29B |
| Q1 '26 | $3.64B | $1.1B |
| Q2 '26 | $2.56B | $7.01B |
| Q3 '26 | $4.02B | $5.08B |
Three things would turn the announcement into a number: which Azure chip, Cobalt or Maia, actually ships with EMIB; whether Microsoft's commitment extends to 14A wafers; and the second packaging customer Tan said would be ready within two quarters 15.
Get Chips by email, free
How AI demand is reshaping chipmakers, memory and the fabs behind them. In our daily email, every weekday.






Reader comments
0 comments