Intel's coverage in our analysis has been almost entirely about the AI accelerator side of the business — and the signal is consistently bearish on that line.
Gaudi catalog status. Per the May 18 daily brief: GAUDI2 shows is_active=false on AWS (launched May 2022 — abandoned at 3.5 years). GAUDI3 is also is_active=false. This is meaningfully faster catalog rationalisation than the equivalent Nvidia products (H100 still active at 2.5 years and growing). Our interpretation: Intel's AI ASIC strategy is not gaining traction with AWS's own engineers — the people best positioned to evaluate whether Gaudi can substitute for Nvidia at meaningful workload scale. If Intel's biggest hyperscaler customer can't maintain the catalog economically, the broader market opportunity is materially smaller than Intel's product roadmap implies.
Implication for the broader ASIC thesis. Intel's failure pattern is a useful contrast to the AWS Trainium and Google TPU stories. Trainium spot prints (+332% in Ohio over 7 days) suggest real AWS-internal demand even where AWS hasn't opened the silicon to external customers. The Gaudi pattern suggests that "build your own AI accelerator" is harder than is widely believed — and that even hyperscaler partners struggle to maintain a position once a product launches. This is part of why the Broadcom picks-and-shovels thesis (covered separately) holds: the chip-design IP and packaging matter more than the brand on the die.
The CPU side. Less heavily covered in our recent briefs, but AMD's EPYC share gains in hyperscaler procurement disclosures (referenced in the AMD page) imply ongoing Intel share loss in server CPU. The GPU-to-CPU ratio shift (8:1 → 1:1) would in theory expand the CPU market — but only as a CPU-favourable thesis if Intel can hold share.
The composite picture: a name with structural headwinds across both AI ASIC ambitions and server CPU. Coverage in our briefs has been almost entirely about the failure modes of the Gaudi line rather than positive catalysts.