Two engineers examine advanced chips as glowing cloud-security graphics connect a server room to stored data.
Microsoft and Intel say they are expanding their decades-long partnership under a long-term agreement. The target is "select areas" of Azure infrastructure, and the stated driver is growing cloud and AI demand. The announcement is mostly a strategic supply and engineering deal. The specifics matter, though, because several of them touch SQL Server, Windows Server and confidential VMs.

Two engineers examine advanced chips as glowing cloud-security graphics connect a server room to stored data. What was announced​

Microsoft is the first company to announce use of Intel Foundry's EMIB advanced packaging technology for Azure first-party silicon. The agreement also confirms multi-year Intel Xeon supply capacity, including instances of next-generation server CPUs codenamed Diamond Rapids, and expands collaboration across the Azure roadmap and Intel-based Azure hardware systems.

The companies also named three joint work areas:

  • Xeon with Compute Express Link (CXL), for memory efficiency.
  • Intel QuickAssist Technology (QAT), for compression and encryption offload.
  • Confidential computing.

Microsoft's Saurabh Dighe, CVP of Azure System & Architecture, framed the deal around joining packaging, custom compute, memory and security. Intel Foundry's Naga Chandrasekaran said EMIB lets the two companies combine heterogeneous integration with manufacturing scale. Both are statements of intent, not independent validation.

What is not disclosed​

The announcement leaves out a lot:

  • Contract value.
  • Processor quantities.
  • Ship dates.
  • Azure regions.
  • Customer availability dates.
  • Which Azure first-party chips will use EMIB.
  • Pricing changes.
  • Any new VM SKU beyond the M-series v4 sizes named in the benchmark footnote.

Treat it as a roadmap and supply commitment, not a service launch.

The three headline claims​

Up to 2.3x on M-series v4 with CXL memory​

The claim is up to 2.3x greater performance and memory flexibility on Azure M-series v4 VMs. It uses CXL memory and Intel Flat Memory Mode on Xeon 6 processors. The footnote shows this is not a like-for-like comparison:

  • Standard_M76ds_2_v4: 76 vCPUs and 1,764 GB of memory, made up of 1 TB DRAM plus 1 TB CXL in Flat Memory Mode.
  • Standard_M64lds_v4: 64 vCPUs and 1,007 GB of DRAM only, across two NUMA nodes.

The workload was a "cube-hypervisor microVM" sandbox test. Each sandbox had 2 vCPUs and 24 GB, and ran a persistent Python interpreter with a 12 GB dataset. The cycle was 13 steps using nine tools, with six seconds of think time per step, for a 126-second cycle. Testing finished on November 10, 2026, and Intel says results may vary.

The test looks like a memory-hungry, many-sandbox pattern, the kind of density problem AI agent hosting creates. The larger and differently configured VM means the multiplier shows what the CXL setup enables, not a pure per-core speedup. Don't expect 2.3x on an arbitrary M-series workload.

Up to 2.5x faster SQL backups with QAT​

The companies claim up to 2.5x faster SQL backups through what they call the first QAT-enabled virtual machines in the cloud. Intel describes QAT as built-in acceleration that offloads compression, decompression and encryption work from CPU cores. Backup compression is an obvious fit.

The comparison again uses very different systems:

  • Baseline: a two-socket Xeon Gold 8362 server with 128 CPUs, 1,024 GB of DDR4-2933 and Windows Server 2025.
  • Azure side: an Intel Xeon 6973P-C (Granite Rapids) VM with 384 processors, 2.06 TB of memory and six NUMA nodes. It ran Windows Server 2025 Datacenter and SQL Server 2025.

Testing finished on November 2, 2026, and results may vary. The baseline is an older-generation, much smaller machine, so the figure mixes generational gains with QAT gains. Intel's separate QAT page says Microsoft previously showed QAT accelerating backup compression on SQL Server 2022. That fits the plausibility of the claim, but it doesn't prove your own backup jobs will see 2.5x. If you run SQL Server on Azure VMs, the practical step is to benchmark your own backups, including compression settings, once QAT-enabled sizes become available. The announcement gives no date or size list.

Confidential VMs and Intel TDX​

The announcement says Intel Trust Domain Extensions (TDX) provides encryption of VM memory and CPU register state as confidential VMs move between hosts. That addresses a real concern: protecting state during host maintenance or rebalancing. It does not say which services support this, what the customer configuration requirements are, or what the performance impact is.

There is a planning angle for confidential-computing users. A WindowsForum report from October says Intel will make Diamond Rapids the last Xeon platform with SGX, and that its confidential-computing focus is shifting to TDX. Teams that write enclave software or run confidential workloads should read this announcement alongside that roadmap.

Diamond Rapids and EMIB: a wrinkle​

The announcement ties Intel's EMIB packaging to Azure first-party silicon. It names Diamond Rapids only as the next-generation server CPU in the Xeon supply commitment. Don't conflate the two.

Hot Chips 2026 coverage suggests Diamond Rapids itself is not an EMIB design. Techzine reports that Intel is adopting Foveros Direct 3D and the UCIe-S interconnect, moving away from EMIB. The Register says Intel skipped EMIB in favor of a UCIe-S interconnect carried over the organic package. EMIB would have required edge-to-edge connectivity between compute and I/O dies. So the EMIB piece most likely refers to Azure's own custom silicon, not the Xeon itself. The announcement doesn't say, and that is my inference.

Timing also matters. VideoCardz reports Diamond Rapids is expected to launch in 2027. The Register says the launch was delayed until 2027. Neither Microsoft nor Intel gave an Azure deployment schedule, so don't assume Diamond Rapids instances are imminent.

Why it matters​

  • Supply assurance: a multi-year Xeon commitment from Intel gives Azure capacity planning certainty. It also shows Microsoft keeping several silicon suppliers in play alongside its own chips.
  • Foundry credibility: a hyperscaler named as the first EMIB customer for first-party silicon is a useful reference point for Intel Foundry. Microsoft's chip details remain unannounced.
  • Memory tiering: CXL-expanded VMs hint at cheaper ways to get very large memory footprints for in-memory databases and AI agent workloads.
  • SQL Server and Windows Server: both benchmarks used Windows Server 2025, and the backup test used SQL Server 2025. That is relevant to anyone on that stack.

What admins should do now​

  1. Don't plan around the multipliers. Both are "up to" figures from vendor-run tests with mismatched configurations.
  2. Watch Azure announcements for QAT-enabled VM sizes and CXL-backed M-series v4 sizes, and confirm region support before designing around them.
  3. If you use confidential VMs, review your TDX and SGX dependencies, and ask Microsoft what live-migration protection requires.
  4. Benchmark your own backup and memory-heavy workloads before and after moving to any new SKU.

Bottom line​

This is a strategic agreement with a few interesting technical hints, not a product launch. The concrete takeaways are secured Xeon supply, a first EMIB use for Azure silicon, and three areas of joint work on memory, SQL backups and confidential VMs. Everything about pricing, availability and real-world gains is still to be disclosed.

 

References

  1. Microsoft and Intel deepen collaboration to deliver differentiated Azure infrastructure - Intel Intel Wed, 18 Nov 2026 16:29:27 GMT
  2. Intel unveils Diamond Rapids: 256 cores on its own 18A-P process - Techzine Global techzine.eu
  3. Intel details 256-core Xeon 7 "Diamond Rapids" with 16 CPU tiles and 1.28GB cache - VideoCardz.com videocardz.com