Independent, line-by-line verifiable research from primary SEC filings · no jargon · willing to differ from consensus

This is research produced by the DeedAlpha engine, published 2026-09-23; content reflects data as of that date.

AMD Deep-Dive: The Trillion-Dollar Party and Its Reefs — What Are the Odds for AI's Second Pole?

Thesis verdict: holds, with conditions — the data-center ramp has genuinely lifted gross margin, but a lasting rise in EPS depends on whether hyperscaler orders can absorb $30.28B of rigid commitments and offset dilution from 320 million warrant shares.

AMD is, at its core, a fabless chip designer. It builds no fabs of its own; wafer manufacturing is entrusted entirely to TSMC, and AMD earns its profit by designing high-value, high-performance chips and system platforms. Its money comes from four customer camps: cloud providers and hyperscalers such as Microsoft, Meta and Oracle buying AI accelerators and server CPUs (Data Center); PC brands buying its processors (Client); console makers such as Sony and Microsoft buying custom SoCs (Gaming); and industrial and automotive customers buying its adaptive and FPGA chips (Embedded).

The financial snapshot for Q2 2026 (period ended 2026-06-27) shows a business structure in the middle of a seismic shift. Company-wide net revenue was $11.536B for the quarter, with operating income of $1.990B. By segment:

Dissected by value-stack layer (the value stack measures how high a company sits in the industry's value allocation; the higher the layer, the more control), AMD's data-center AI business is climbing laboriously from L2 (capital-intensive manufacturing and physical conversion) toward L3 (proprietary assets and system wiring). AMD used to sell bare accelerator chips, easily commoditized and exposed to customers' in-house ASICs flattening its margins; now it is trying to lock in customer systems through its own Helios rack-scale system and the open-source ROCm software stack (AMD's own open deep-learning development platform). Yet on a workflow-depth test one fact stands: AMD's EPYC server CPUs are deeply embedded in the cloud providers' underlying infrastructure, with high migration costs; but in the eyes of most hyperscalers its AI accelerators are still only a "second source" beside Nvidia's expensive hardware — shallow substitute compute that has not truly captured control of the customers' upper development ecosystem.

That exposes a wide gap between the current market narrative and contractual substance: the bulls, deep in the trillion-dollar party, have granted AMD a 157.6x P/E premium; what the market has not fully seen through is that, to win hyperscalers' favor, AMD not only has to shoulder the $30.28B supply-chain burden first, but has also handed those big customers a massive number of warrants at very low strike prices.

The above is the opening. The full research has 6 more sections, available with purchase.

Register to read one full report free, and to buy this one Request AMD research

This research was generated by the DeedAlpha engine for research reference only. It does not constitute an offer to buy or sell any security, nor investment advice. Data is drawn from public SEC filings and public news and may be delayed or contain errors. Investing involves risk.

← All stock research