← Back to feed

Intel and Google announced a multiyear infrastructure collaboration on April 9 that expands the role of Xeon CPUs across Google Cloud and accelerates joint development of a custom ASIC called the Infrastructure Processing Unit, or IPU.

The announcement covers two threads. First: Google will continue deploying Intel Xeon processors at scale for AI inference and general-purpose cloud workloads, including Intel’s latest Xeon 6 chips in C4 and N4 Google Cloud instances. Second: the two companies will expand co-development of custom IPUs — programmable ASIC accelerators designed to offload networking, storage, and security functions from the CPU, reducing overhead costs at the system level. Intel and Google have worked on IPU architecture together since 2022; this agreement deepens that work across future generations.

No financial terms were disclosed. No deployment timeline was provided for next-generation IPU hardware.

The Strategic Angle

The AI infrastructure conversation has been dominated by GPUs since 2022 — first NVIDIA H100, then Blackwell, now Vera Rubin. Intel’s Gaudi line has gained limited traction. This deal is a different bet: rather than compete on accelerators, Intel is arguing for the irreducibility of CPUs in heterogeneous AI systems.

The argument has merit. Large-scale AI deployment involves significant non-GPU work: tokenization, KV-cache management, load balancing, memory orchestration, batching logic. These workloads run on CPUs. As inference scales, CPU efficiency at the orchestration layer becomes a real cost factor. Intel’s internal framing for this announcement — “AI doesn’t run on accelerators alone — it runs on systems” — is positioning, but it’s not wrong.

IPUs serve a related function: networking and storage acceleration that GPUs don’t want to do. As data center interconnect complexity grows with multi-node inference (think Vera Rubin NVL288 clusters), offloading those functions to dedicated ASIC fabric has real TCO implications.

What It Does Not Tell Us

The announcement is light on specifics that would let anyone evaluate magnitude. No revenue commitment from Google. No Xeon 6 deployment numbers. No IPU roadmap milestones. No performance benchmarks comparing IPU-augmented vs baseline inference stacks.

This reads as a relationship signal — Intel and Google reaffirming collaboration in a market moment where Intel’s relevance to the AI stack is under genuine question. Intel’s stock has declined significantly relative to NVIDIA over the past three years. A press release deepening an existing partnership is not the same as a capability breakthrough.

The IPU angle is the more interesting technical thread. Custom silicon for infrastructure offload is an underreported part of hyperscaler AI economics. If Google’s IPU can measurably reduce per-inference system cost at the C4/N4 layer, that will show up in Gemini API pricing over time.

Key Facts

• Announced: April 9, 2026
• Parties: Intel Corporation, Google (Alphabet)
• Deal type: Multiyear collaboration agreement
• Xeon: Continued deployment in Google Cloud C4 and N4 instances (Xeon 6)
• IPU: Expanded co-development of custom ASIC-based infrastructure processing units
• IPU history: Intel–Google collaboration on IPU began 2022
• Financial terms: Not disclosed
• Timeline: Not specified