The Data Center Just Became AMD's Main Character

AMD's data center revenue just hit $5.8 billion in a single quarter. That's 57% of $AMD's total company revenue — the first time DC has crossed 50% in company history. And the MI400 hasn't even shipped yet. The chip built to battle NVIDIA's Blackwell Ultra is still on the runway, and AMD is already printing numbers that sounded like fan fiction three years ago. Stock's up 137% YTD, around $531 with an $866 billion market cap. Lisa Su dragged this company into an AI empire while everyone watched Jensen Huang's leather jackets.

The Receipts: AMD by the Numbers

No fluff. Just the tape: Q1 2026 Revenue: $10.25B (+38% YoY). Data Center Revenue: $5.8B (+57% YoY). DC as % of Total: 57% — first time north of 50%. Free Cash Flow: $2.6B. Q2 Guidance: $11.2B (+46% YoY). TTM Revenue: $37.45B. TTM Net Income (GAAP): $5.01B. Trailing P/E: ~173x GAAP. Forward P/E: ~40.5x. YTD Return: +137.69%. Market Cap: $866B.

These aren't "maybe someday" numbers. The thesis already printed — and the MI400 isn't even in the game yet.

The Silicon Coming for the Crown

The MI300X was the breakout: 192 GB HBM3, 153 billion transistors, CDNA 3. Proved AMD could build AI accelerators hyperscalers actually buy. But the MI300X was the warm-up. The MI400 is the headliner — TSMC 2nm, 432 GB HBM4, CDNA 5. AMD's already locking anchor tenants: the OpenAI 6-gigawatt multi-year deal names AMD a "core AI-factory partner." That's industrial-scale commitment from the most important AI company alive. Meta dual-sourcing. Microsoft dual-sourcing. Every hyperscaler wants a second silicon supplier, and $AMD is the only credible one standing.

Second Place Is a Multi-Billion-Dollar Throne

Nobody's saying AMD overtakes NVIDIA. Jensen's empire owns 80-81% of AI GPUs. AMD's at 6-8%. Sounds like an L until you run the math. AI chip TAM is headed toward $500 billion+. AMD doesn't need half. Capture 10-15% and that's $50-75B in annual DC revenue — more than AMD's entire company today. The real thesis: inference. By 2028-2030, inference will be 70-80% of all AI compute. Inference loves memory bandwidth, hates the NVIDIA tax. AMD packs more memory per dollar. When inference becomes the main event, that equation gets dangerous — for NVIDIA.

The CUDA Moat Is Real — Moats Don't Last Forever

NVIDIA's CUDA ecosystem is 15+ years of developer lock-in. ROCm is catching up but still feels like a knife in a gunfight. Nobody gets fired for buying NVIDIA. Yet. But the AI stack is consolidating around PyTorch, JAX, Triton — frameworks that abstract the hardware. When inference runs behind an API, you don't care whose silicon is underneath. You care about throughput per dollar. On that metric, $AMD's argument gets stronger every generation. Execution risk is real — 2nm supply chain, ROCm maturity, enterprise inertia. But AMD only needs to be good enough for hyperscalers writing nine-figure purchase orders. With DC at 57% and climbing, it is.

Bottom Line

AMD's AI story isn't about one quarter — it's a structural shift. Hyperscalers want dual-sourcing. Governments want sovereign AI. Enterprises want cheaper inference. Every vector points to a credible #2, and AMD is the only name on the board. Data center is now the majority of the company. MI400 on 2nm inbound. OpenAI deal signed. Stock up 137% YTD with a forward P/E that looks reasonable against the growth rate. The setup isn't complicated. Just stop staring at NVIDIA long enough to see it.

— The Signal