Confirmed 18 Nov 2026 · Compute A one-minute read

Microsoft is the first named customer for Intel Foundry packaging of its own Azure chips.

The 2 companies announced a long-term agreement covering Intel Foundry's EMIB packaging for Microsoft's first-party silicon and multi-year Xeon supply. Caveat: no volumes, dates or dollar figures were disclosed.

Until now

Microsoft designs its own chips, and someone else quietly builds and packages them.

Now

Intel Foundry has its first named hyperscaler for advanced packaging.

Why it matters

Advanced packaging, not the transistor, is the bottleneck in AI silicon, and Microsoft had said nothing about who does that step for its own chips. Naming Intel Foundry gives Intel the hyperscaler reference it has lacked. The release says nothing about volume or money.

Established

  • Microsoft is first to announce Intel Foundry EMIB packaging for Azure silicon.
  • A long-term agreement confirms multi-year Xeon capacity, including Diamond Rapids.
  • Both companies issued the release jointly at 11:47 Eastern Time.

Not established

  • Any volume, wafer or packaging commitment. None is disclosed.
  • Any dollar value, or whether the obligations bind either party.
  • That Microsoft is moving work away from its current packaging supplier.
The five-minute readThe record, the timeline, the weight and 4 sources

The record

The joint release was issued at 11:47 Eastern Time on November 18, 2026, which is 16:47 Universal Time and 20 minutes inside this edition's window, with a dateline of Redmond, Washington and Santa Clara, California. It describes an expansion of a decades-long collaboration under a "long-term agreement", and discloses no contract terms, no financial value, and nothing about whether the obligations bind either party. On packaging, Microsoft is described as the first company to announce using Intel Foundry's EMIB advanced packaging for Azure first-party silicon; no products and no volumes are named. On supply, the agreement "confirms multi-year Intel Xeon supply capacity", including instances built on the next-generation server processors codenamed Diamond Rapids, with no quantities, amounts or delivery dates given. Other named areas of work: CXL memory on Intel Xeon, Intel QuickAssist, and confidential computing using Intel TDX. The performance claims are the vendors' own and footnoted as varying by configuration: up to 2.3 times the performance for Azure M-series v4 virtual machines using CXL memory and Intel Flat Memory Mode on Xeon 6, and up to 2.5 times faster SQL backups on the first QuickAssist-enabled cloud virtual machines. Saurabh Dighe, corporate vice president for Azure System and Architecture at Microsoft, is quoted saying "Our customers need infrastructure built around the workloads that matter to them", and that the companies are connecting silicon and systems engineering to meet growing demand for cloud and AI computing. Naga Chandrasekaran, Intel's chief technology and operations officer and general manager of Intel Foundry, describes AI infrastructure moving toward "increasingly complex systems designed through the co-optimization of silicon, packaging, compute memory and software". The release never uses the words data centre, accelerator or GPU, and names no competitor it is replacing.

How it got here

  1. Oct 12

    Microsoft unveils its own AI chip and says it will build Azure around it from 2027. Edition 049's card.

  2. Nov 6

    Microsoft switches off Oracle AI capacity it had rented and reviews a much larger book. Edition 078's card.

  3. Nov 12

    A withdrawn power connection costs Nscale a $3.22 billion Microsoft contract. Edition 090's card.

  4. Nov 17

    Nvidia reports supply and capacity commitments of $390 billion, up from $279 billion, primarily multi-year memory procurement. Edition 100's capital card.

  5. Nov 18

    Microsoft and Intel issue a joint release at 11:47 Eastern Time naming Intel Foundry's EMIB packaging for Azure first-party silicon and confirming multi-year Xeon capacity including Diamond Rapids.

Who has to react

TSMC, which packages most leading-edge AI silicon today; Intel Foundry's unnamed prospects, who now have a reference to point at; Nvidia and AMD, whose supply lines run through the same packaging capacity; Amazon and Google, who design their own chips too; and Intel shareholders, who have waited years for a named hyperscaler.

What to watch

  • Would strengthen itIntel names a second hyperscaler or a top-10 fabless customer for Intel Foundry packaging or wafers, or reports external foundry revenue above $1 billion in a quarter, by September 30, 2027.
  • Would strengthen itMicrosoft, Amazon or Google discloses a dated volume or dollar commitment for advanced packaging capacity outside Taiwan, by December 31, 2027.
  • Would weaken itMicrosoft ships its first-party AI silicon in 2027 with no Intel Foundry content named, or neither company mentions the agreement in its next annual filing, by June 30, 2028.

Why it weighs this much

Signal strength
CostLow
CommitmentMedium
ReversibilityEasy
PersistenceYears

Cost is low and we score it that way: a joint press release with no disclosed volume, price or penalty costs the parties almost nothing. Commitment is medium, not high. Two public companies naming each other in a long-term agreement, with a named packaging technology and a named processor codename, is harder to unwind than a memorandum and much easier than a filing. Reversibility is easy for the same reason: nothing here obliges Microsoft to buy a single part. Persistence is years, because packaging qualification takes years once begun, and because the Xeon capacity named reaches a processor generation that has not shipped.

The words

EMIB
Embedded Multi-die Interconnect Bridge. Intel's method for wiring separate chips together inside one package at high speed, so a product can be built from several pieces.
Advanced packaging
Assembling finished chips and memory into one component. It now limits how many AI parts the industry can ship, more than making the transistors does.
Foundry
A business that manufactures chips other companies design. Intel ran one only for itself until recently, which is why a named outside customer matters.
First-party silicon
Chips a cloud company designs for its own machines instead of buying. Microsoft unveiled its own AI chip in October 2026.

Sources

Your call

Which explanation do you believe?

A call is final. Other people's calls show after yours.

8 contributions. Thank you.

You have built
a record here.

Keep it: create an account to verify your contributions and put your name to the answers you choose. Your record stays yours.

Keep my record ↗
Optional: help show whose perspectives are missing
Kept with your answers under a random id, never with your name. Used for one thing: showing you, on your record, where you stand next to people like you, once at least 20 of them have answered.

Add your signature

Stand behind your judgment.

Create an account to attach your name to this answer. Your other answers show no name unless you sign them too.

Sign up to add your name ↗

Already have an account?