The 1 Number: 2,957,362 H100-equivalent GPUs. That is what NVIDIA (NASDAQ:NVDA | NVDA Price Prediction) shipped in a single quarter, and it explains why the AI compute leader still has a long runway.
The Number
NVIDIA shipped 2,957,362 H100-equivalent units in Q4 2025, according to Epoch AI, a figure that represents nearly two-thirds of all measured AI compute capacity in the period. The metric comes from a standardized industry ranking that converts different chips into “H100 equivalents” so that processors from different vendors and architectures can be measured on a common AI training yardstick. This is a unit-volume figure, not a dollar figure, and it is reported as an estimate for the calendar quarter.
What It Means
The next three competitors combined sold roughly half of what NVIDIA shipped: Alphabet (NASDAQ:GOOG) at 976,313, Advanced Micro Devices (NASDAQ:AMD) at 226,485, and Amazon (NASDAQ:AMZN) at 221,354. NVIDIA out-shipped that trio combined by a 2-to-1 margin.
That manufacturing scale translates straight into the income statement. Q4 FY2026 Data Center revenue reached $62.31 billion, up 75% year over year, with Data Center Networking alone at $10.98 billion, up 263% year over year on NVLink fabric demand for GB200 and GB300 systems. Total Q4 revenue came in at $68.127 billion (+73.21% YoY), non-GAAP EPS landed at $1.62, beating the $1.52 consensus by 6.58%, and free cash flow hit $34.902 billion, up 124.42% year over year. Non-GAAP gross margin held at 75.2% as Blackwell mix improved.
The volume lead also feeds a software moat. There are 1.5 million AI models on Hugging Face running on NVIDIA CUDA, and CEO Jensen Huang told analysts “In this new world of AI, compute is revenue.”
Market Reaction
NVDA closed $208.27 on April 24, 2026, up 95.73% over the prior year (from $106.41 on April 24, 2025), 16.56% over the prior month, and 11.68% year to date. By comparison, AMD is up 268.17% over one year and Alphabet is up 112.73% over one year, both sharing the AI infrastructure tailwind. NVIDIA’s market capitalization stands at $5.11 trillion.
Bull Case
The volume lead is compounding rather than narrowing. NVIDIA reported that nearly 9 gigawatts of Blackwell infrastructure have been deployed across hyperscalers and enterprises, with Hopper and Ampere products sold out in the cloud and capacity secured into calendar 2027. Sovereign AI revenue more than tripled year over year to over $30 billion in FY2026.
Customer commitments back the demand thesis. NVIDIA disclosed a multi-year partnership with Meta Platforms (NASDAQ:META) covering millions of Blackwell and Rubin GPUs, an Anthropic technology partnership tied to a $10 billion investment, and a CoreWeave collaboration on 5+ gigawatts of AI factories by 2030. Top-five hyperscaler 2026 capex visibility is approaching $700 billion.
Forward guidance reinforces the trajectory. Management projects Q1 FY2027 revenue of approximately $78.0 billion (plus or minus 2%), explicitly excluding any Data Center compute revenue from China. Note that this is forward guidance, not a reported result. The Vera Rubin platform, with up to 10x reduction in inference token cost vs. Blackwell, is scheduled for production shipments in the second half of calendar 2026.
For long-term holders, capital return adds support. NVIDIA returned $41.1 billion to shareholders in FY2026 and has $58.5 billion remaining under its share repurchase authorization.
Bottom Line
The roughly 3 million H100-equivalent shipment number matters because it quantifies the gap between NVIDIA and the field at the unit level, before pricing, software, or systems integration enter the picture. CUDA lock-in, networking attach, and locked-in supply through 2027 sit on top of that base. The forward catalyst for retirement-focused holders to watch is the next reporting cycle: Q1 FY2027 earnings, paired with Vera Rubin deployments at AWS, Google Cloud, Microsoft Azure, and Oracle Cloud.
Top Gaining Stocks
Top Losing Stocks