Google leans on Intel for more SmartNICs as IPU demand ripples through the cloud

What happened

It has been reported that Google will keep buying SmartNICs — or infrastructure processing units (IPUs) — from Intel rather than designing its own ASICs the way AWS did with Nitro. Intel’s Mount Evans IPU, which debuted with Google C3 instances in 2022, is the baseline. Now, the two companies have allegedly expanded that collaboration to build next‑generation IPUs. No specs were given, but with AI clusters gobbling bandwidth, faster networking hardware is a safe bet.

Why it matters

It has been reported that Intel’s custom ASIC business grew more than 50 percent in 2025 and reached an annualized run rate above $1 billion at the end of Q4 — a tidy bit of breathing room for a company under constant pressure. For Intel, this deal is more than product placement; it’s validation that hyperscalers still want the x86 ecosystem and specialized NIC offload silicon. For Google, it keeps options open: the Chocolate Factory also runs its Arm Axion chips and GPUs and TPUs, but many customers still prefer x86 for compatibility and performance. Who wins? Customers, if competition stays fierce.

What’s next

Expect performance bets rather than architecture wars. Intel didn’t spell out the next IPU’s capabilities; it has been reported that details remain under wraps. The bigger story is trend, not a single product: hyperscalers are mixing offload silicon, in‑house CPUs, and GPUs in ever more creative combos. In short — Google isn’t going all in on bespoke NIC ASICs, and Intel just bought itself another headline. That matters in a market where the best CPU for an agentic AI might simply be whatever’s idle at the moment.

Sources: The Register