Intel Earnings, AMD Earnings, Intel and AMD's Divergence (Stratechery Update, 2026-11-11)
Why this is in the vault
Thompson explains why Intel's Q3 guide cratered while AMD's data center revenue nearly doubled, tracing it to chiplet modularity versus monolithic IDM design, and floats that Intel's fix may be to double down on being an IDM rather than copy AMD or TSMC. It is the next data point in the vault's Intel and agentic-CPU-demand thread.
The core argument
Intel: Q4 guide of $15-16B versus ~$17B expected, the worst stock rout in four decades. Causes: a PC slump driven by memory shortages and component price spikes, Intel 3 yield and substrate (EMIB packaging) constraints, and an Israel fab shutdown. The market read it as a warning for AMD and the AI trade: where is the agentic CPU demand?
AMD: Q4 guide of $17B versus ~$16.3B expected; data center revenue up 97% to $8.5B. Lisa Su credits personal and enterprise agents (Meta Muse, OpenAI Dots) as the emerging CPU killer app. AMD expects double-digit PC unit declines in 2027 but "aligns sell-in with sell-through," while Intel is at the mercy of customer inventory.
The divergence: AMD, forced into modularity by outsourcing manufacturing (GlobalFoundries spin-out 2008, TSMC from 2018), reuses largely the same compute chiplets across consumer and server, so it can shift supply from the crashing PC market to booming servers. Intel's consumer and server compute dies differ, so reallocating means retooling wafer lines (CFO Zinsner on Intel 3 and 18A). EMIB packaging adds a similar problem versus AMD's simpler conventional packaging. Thompson's closing turn: the foundry strategy may now be a distraction, and Intel's best path may be a performance-obsessed IDM, not "AMD or TSMC." Interpretation is Thompson's; figures are from his quoted Bloomberg and earnings-call excerpts.
Mapping against Ray Data Co
Most concrete tie: the AMD side supplies direct evidence for the agentic-demand-for-CPUs claim already tracked in [[2026-01-27-stratechery-intel-earnings-agentic]], and it matters to the capital-cycle work (project_investing_markov_capital_cycle): server CPU demand is surging on agent workloads while consumer silicon is hit by the same memory shortage logged in the 11-03 Apple note. It also supports the "agents in production" credibility lens (project_credibility_for_phdata_sales): hyperscaler EPYC capacity is being added specifically for agent workloads. No direct RDCO action; this is thesis evidence, not an operational change. Mapping strength: medium.
⚠️ Sponsorship
Stripe holds the "Presented by" slot (top banner plus a mid-issue block on Link, Stripe's agent wallet, citing 38x agentic-commerce growth in September, with a utm_source=stratechery link). Standing sponsor since the 2026-10-21 announcement. Bias implication: the Link pitch is unrelated to the Intel/AMD analysis, but its "38x" claim is a vendor figure and unverified here.
Related
- [[2026-01-27-stratechery-intel-earnings-agentic]] - prior Intel earnings framing the agentic CPU opportunity
- [[2026-04-29-stratechery-intel-earnings-terafab]] - Intel differentiation and foundry (Terafab) thread this note challenges
- [[2026-11-03-stratechery-apple-earnings-hardware-regression]] - same memory-shortage thread seen from the Apple side
- [[2026-10-21-stratechery-presented-by-stripe-sponsorship]] - Stripe sponsorship origin