On April 9, Google and Intel announced a multi-year, multi-generation collaboration to advance AI and cloud infrastructure together. The deal covers two areas: continued deployment of Intel Xeon 6 CPUs across Google Cloud instances, and expanded co-development of custom ASIC-based Infrastructure Processing Units (IPUs), programmable accelerators that offload networking, storage, and security tasks from host CPUs.

The announcement adds an important chapter to the AI infrastructure story that has been dominated by Nvidia for the past three years.

What IPUs Actually Do

The IPU concept is less flashy than a GPU but arguably more consequential at scale. When a data center runs thousands of AI inference jobs simultaneously, significant CPU overhead goes to handling networking packets, storage I/O, and security operations — work that has nothing to do with the AI computation itself. IPUs offload those functions onto dedicated chips, freeing the main CPU to focus on coordination and compute tasks that actually benefit from general-purpose processing.

The result is better utilization, more predictable performance, and lower effective cost per AI workload. Google has been quietly deploying Intel IPUs for years. This deal formalizes a multi-generation roadmap for that co-development, aligning Intel’s chip design roadmap with Google’s infrastructure needs.

Intel Xeon 6 processors are already powering Google Cloud’s C4 and N4 instances, handling workload types ranging from large-scale AI training coordination to latency-sensitive inference. The multi-year alignment means future Xeon generations will be developed with Google’s workload requirements explicitly in the design brief.

Why This Matters Beyond Chips

The GPU market for AI is still heavily Nvidia-dominated. Nvidia holds roughly 80 percent of AI accelerator revenue, and its CUDA software platform has created a deep ecosystem that is difficult to compete with directly. Intel has largely stopped trying to compete on that front.

Instead, Intel’s strategy is to own the rest of the data center: the CPUs, the network chips, the infrastructure layer that sits underneath the GPU clusters. The Google deal is the clearest signal yet that this strategy is working. Google is one of the world’s largest cloud providers and has its own in-house AI chip program (TPUs). The fact that Google is also deepening its Intel relationship reflects a pragmatic recognition that heterogeneous infrastructure, where multiple chip types work together, outperforms any single-chip approach at scale.

This is the “balanced compute” thesis in action. AI infrastructure is not just accelerators. It is accelerators plus CPUs plus IPUs plus networking plus storage, all optimized to work together. The companies getting this right will run AI workloads at lower cost and higher efficiency than those stacking homogeneous GPU clusters.

What This Means for Business

For most businesses, the details of IPU co-development are infrastructure plumbing. The business-level implications are more practical.

First, it signals continued investment in Google Cloud’s AI infrastructure competitiveness. Businesses building AI workloads on Google Cloud are building on a platform with long-term infrastructure commitments from major chip makers, not just access to rented Nvidia capacity.

Second, and more broadly, the expansion of serious non-Nvidia AI infrastructure investment is a sign that compute competition is increasing. More competition in AI chips tends to drive efficiency improvements and, eventually, lower costs per workload. The current economics of enterprise AI — still expensive for many smaller organisations — will improve as infrastructure competition intensifies.

Third, it reinforces a pattern worth understanding: the AI infrastructure layer is becoming as strategically important as the model layer. For data and technology leaders, understanding the compute stack — not just which LLM to use — is increasingly part of making sound vendor decisions.

The companies that will deploy AI most effectively over the next three years will be those that understand their compute dependencies, not just their model subscriptions. The infrastructure bets being made now by Google, Intel, Nvidia, Microsoft, and others will determine the cost and capability curves for enterprise AI through the end of the decade.

If you’re deciding where to start with agents, start here. The free Working With Claude field guide walks through the ecosystem, Claude Code, and a real rollout plan. Get your copy.