NVDA Inference Share Forecast at 50% by 2028 as AMD and TPUs Gain

The narrative around NVIDIA's competitive moatA sustainable competitive advantage that protects long-term returns on capital. has begun to crack under scrutiny. While NVIDIA remains the overwhelming leader in AI training chips, the real battlefield is shifting to inference, where margins are lower but volumes are vastly larger. AMD, Google's internally developed TPUs, and Amazon's Trainium chips are all making material gains, and a growing cohort of market observers believes NVIDIA's inference market share could fall to 50% by 2028, down from current levels above 80%.

This structural shift has several dimensions. First, hyperscalers have grown increasingly sophisticated at building proprietary silicon for their own workloads. Google, Amazon, and Meta have the engineering talent, the financial resources, and the incentive to reduce their dependence on NVIDIA. Second, AMD's MI300 and successor products have closed the performance gap in many workloads, and third-party benchmarks increasingly show feature parity at lower cost. Third, the software ecosystem around AMD (ROCm), Google (TPU), and others is maturing, lowering switching costs for new deployments.

For NVIDIA, this matters because gross margins in inference are 10-15 percentage points lower than in training, and if NVIDIA loses half the inference market, it directly pressures the blended gross margin that the street has baked into valuations. Several sell-side houses have not yet adjusted their margin forecasts to reflect this risk, creating a subtle but material disconnect between consensus estimates and the actual competitive landscape. The sell-side remains bullish on NVIDIA's FY2026-2027 earnings, but that view rests on assumptions about sustained market share that are increasingly contested.

Retail traders and momentumThe empirical fact that winners keep winning over the medium term. players have extrapolated from the Q1 earnings beat and the cup-and-handle breakout into a conviction that NVIDIA remains a one-way trade. But institutional flows suggest more caution. Money is flowing into broadsheet semiconductor and AI exposure (via SPY and QQQ) rather than concentrating in NVDA, a potential warning sign that smart money is diversifying away from single-stock concentration risk.