Nvidia’s AMD Paradox is Exactly Why I Keep Buying

I keep buying NVIDIA (NASDAQ:NVDA | NVDA Price Prediction) every time the AMD narrative gets louder, and the louder it gets, the more shares I add. That paradox is why my cost basis keeps climbing and my conviction keeps deepening.

The story I keep hearing is that AMD (NASDAQ:AMD) is finally taking hyperscaler share. It is true. AMD’s Data Center segment printed $5.78B last quarter, up 57% YoY, with a 6 GW Meta partnership anchoring the MI450 ramp. Good for them. NVIDIA’s Data Center revenue in the same quarter was $75.246 billion, up 92% YoY. NVIDIA’s Data Center Networking line alone, at $14.8 billion, is larger than AMD’s entire company revenue of $10.25B. The pie is expanding faster than any competitor can eat.

The Three Reasons I Keep Adding

First, the margin structure. NVIDIA posted a 75.0% non-GAAP gross margin and a 60.4% operating margin against AMD’s 55% non-GAAP gross margin and 10.66% operating margin. When AMD absorbs price-sensitive inference workloads, NVIDIA’s mix shifts toward high-margin enterprise systems, sovereign AI, and CUDA-monetized software. AMD’s growth secures the unit economics of the entire AI capex ecosystem, which protects the terminal value of my position.

Second, cash returns. NVIDIA raised its dividend from $0.01 to $0.25 per share, layered on a fresh $80.0 billion buyback on top of $38.5 billion remaining, and returned roughly $20.0 billion to shareholders in a single quarter. FY26 free cash flow hit $96.58B. AMD pays no dividend and generated $5.52B in FY25 free cash flow. That cash-return profile compounds meaningfully over time.

Third, valuation. NVIDIA trades at a P/E of 41 with ROE of 101.5% and ROIC of 92.2%. AMD trades at a P/E of 189 with ROE of 7.19%. I am paying less than a quarter of the earnings multiple for roughly fourteen times the return on equity. AMD’s stock has run 220.77% over one year while NVDA is up 18.06%. That gap is where my opportunity lives.

The Risk I Will Not Wave Away

China export controls cost NVIDIA $4.6 billion in year-ago H20 shipments, and Q2 FY27 guidance excludes any Data Center compute revenue from China. Hyperscalers are roughly 50% of Data Center revenue, and total supply commitments have swelled to $119.0 billion. That is real concentration and real demand-risk exposure. The reason it has not shaken my thesis: Q2 FY27 guidance still calls for $91.0 billion in revenue at a 75.0% gross margin without a dollar from China. The customer base is diversifying into sovereign AI, industrial enterprise, DRIVE Hyperion automakers, and T-Mobile/Nokia AI-RAN. Concentration is loosening while the total addressable market keeps widening.

Why the Buy Button Stays Active

Jensen Huang called it “the largest infrastructure expansion in human history“, and analysts still peg the consensus target at $302.31 with 58 Buy ratings against one Sell. Vera Rubin is announced, Blackwell 300 is ramping, and Dynamo 1.0 delivers a 7x inference boost on existing silicon. Every time AMD wins a hyperscaler slot, NVIDIA’s mix gets richer and its regulatory tail-risk gets shorter. That is the paradox, and it is why I keep buying the one stock everyone assumes I should be worried about.

Contact [email protected] for any questions or corrections.