AMD Stock: DeepSeek V4 and the AI Chip Race

AMD closed at $475.51 on June 9, 2026, down 3% on the session after a volatile week that saw shares range from $466 to $542. The price action coincides with a Semianalysis deep-dive published the same day analyzing DeepSeek V4 inference performance across four GPU platforms: Nvidia GB300 NVL72, AMD MI355X, Huawei Ascend 950DT, and Nvidia B200. The DeepSeek V4 release is a genuine event in the AI chip competition because it provides for the first time a public benchmark where AMD's MI355X is competitive with Nvidia's flagship — under specific configurations and with specific software optimizations. That window matters for the AMD equity story more than the share price action suggests.

What the Semianalysis benchmark actually shows

The InferenceX team measured DeepSeek V4 inference performance on Day 0 (model release), then tracked iterative software improvements through Day 43. Key findings:

• 100x performance improvement in 26 days from Day 0 to Day 26 — almost entirely from software optimization on existing hardware.¹
• NVDA CUDA stack worked Day-0 out-of-the-box on vLLM and SGLang inference engines, reflecting the maturity of Nvidia's software ecosystem.
• AMD MI355X reached competitive throughput after software tuning, but required substantial engineering hours to close the Day-0 gap.
• Huawei Ascend 950DT received deep co-design attention from DeepSeek (the model was co-designed in part for Ascend inference) and achieved competitive performance in specific configurations.

The takeaway is not "AMD has caught up to Nvidia." The takeaway is "AMD MI355X is now a defensible second choice for serving open-source frontier models, where Nvidia's CUDA premium is hardest to justify."

That distinction is the AMD equity case.

Why the Q1 2026 numbers support the inference angle

AMD's Q1 2026 financial statements showed revenue of $10.25 billion — up 38% year-over-year (vs. $7.44 billion in Q1 2025) and matching Q4 2025 quarterly run-rate (drillr financial statements). Gross profit was $5.42 billion. Operating income was $1.48 billion. Diluted EPS reached $0.84, up from $0.44 in Q1 2025. Free cash flow was $2.57 billion.

The growth profile here is what differentiates AMD from a pure NVDA-substitute play. AMD's data center segment is now compounding at hyperscaler-cycle rates while client computing has stabilized after the 2022-2023 PC correction. Cash and short-term investments stood at $12.3 billion against total debt of $3.87 billion — a fortress balance sheet that gives AMD the runway to pursue multiple AI chip generations without Nvidia-level CUDA distribution economics.

The MI355X plus the MI400-class roadmap represents AMD's bet that inference workloads will fragment across multiple silicon platforms by 2027-2028. The DeepSeek V4 benchmark is the first public datapoint suggesting that thesis has merit.

How AMD differs from NVDA in the inference fight

Nvidia's competitive moat in AI inference is composed of three layers: CUDA software stack, NVLink high-bandwidth interconnect, and a deep developer ecosystem. Each of those moats matters in different workloads.

CUDA software stack. Nvidia's lead here is largest. AMD's ROCm has narrowed the gap, but for proprietary models trained on Nvidia hardware (most enterprise customers), the software penalty for migration is substantial. The DeepSeek V4 case is unusual because the model is open-source, allowing inference framework teams to optimize for non-Nvidia targets without commercial barriers.

NVLink interconnect. Nvidia's lead is moderate. AMD's Infinity Fabric is competitive at the GPU-to-GPU layer but lags at the rack-level integration. For massive distributed inference workloads, NVLink advantage compounds.

Developer ecosystem. Nvidia's lead is structural. AMD's developer-onboarding investments have improved through 2024-2025 but remain at perhaps 25-35% of Nvidia's ecosystem depth.

AMD's MI355X wins are concentrated where: (1) the model is open-source and inference framework optimization is community-driven, (2) inference workloads are not bandwidth-bottlenecked at the rack level, and (3) customer has internal software engineering capacity. The DeepSeek V4 case satisfies all three.

What this means for portfolio positioning

The AMD trade is not "second derivative on NVDA." It is "inference workload fragmentation winner." Those are different theses with different upside profiles.

If the inference market fragments to 70/30 Nvidia/AMD by 2027 (versus current roughly 85/15), AMD's data center revenue trajectory compounds significantly. Drillr terminal records institutional flow data showing AMD ownership concentration has shifted toward AI-specialist hedge funds through 2025-2026, signaling smart money has been positioning for exactly this scenario. The DeepSeek V4 benchmark is one of the most concrete signals to date that the fragmentation thesis is materializing.

Versus NVDA, the AMD multiple is approximately 35x forward earnings versus NVDA at roughly 22x. The premium is meaningful, but if AMD's revenue growth compounds at 35-40% over the next 3-4 quarters (vs. NVDA's deceleration toward 25-30%), the AMD multiple compression is moderate and the absolute return is competitive.

The risk is that the DeepSeek V4 benchmark turns out to be configuration-specific — that AMD MI355X performance does not generalize to other open-source models with different architectures. If GPT-class and Claude-class proprietary models continue to dominate enterprise spend, the inference workload fragmentation thesis stays in slow motion.

What to monitor through 2026

• AMD Q2 2026 earnings (expected late July) for data center segment growth and Asian region revenue commentary.²
• MI400 roadmap commentary at AMD's 2026 Data Center Tomorrow event.
• Hugging Face, Cerebras, and other open-source-model-supporting platforms expanding MI355X deployment.
• TSMC Q2 2026 commentary on MI355X production allocation — supply availability is a real factor.
• Any DeepSeek V5 / V6 release with explicit AMD support, which would extend the benchmark window.

The AMD investment case is not about beating Nvidia. It is about being the credible second platform in a market that is finally reaching enough scale to support more than one viable architecture. The DeepSeek V4 benchmark is the closest thing to a public-market validation of that thesis we have seen in 2026.

Footnotes

1. Semianalysis, "DeepSeekV4 1.6T Day 0 to Day 43 Performance Over Time — Huawei, GB300 NVL72, MI355X, B200," June 9, 2026. https://newsletter.semianalysis.com/p/deepseekv4-16t-day-0-to-day-43-performance ↩

2. AMD Investor Relations, "Q1 Fiscal 2026 Earnings Release," April 2026. https://ir.amd.com/news-events/press-releases/detail/1234/ ↩