NVIDIA does not report a total number of GPUs sold. Its filings disclose revenue by business, not unit sales, and its fast-growing data-center business includes much more than GPU chips. The clearest public estimate is narrower: industry figures imply NVIDIA shipped about 41.6 million to 42.1 million desktop add-in-board GPUs worldwide in calendar 2025. That is not a count of all NVIDIA GPUs.
The distinction matters for anyone trying to interpret NVIDIA’s revenue or the AI boom: a graphics card, a server GPU module, and a rack-scale system are different units. As of August 16, 2026, no reliable public figure combines them into an all-in total.
What counts as a GPU sold?
“GPU sales” can refer to several different things, and switching between them can make an estimate look more certain than it is.
- GPU die: the silicon made by a foundry.
- GPU package or module: a packaged processor installed in a board or system.
- Add-in board: a desktop graphics card containing an NVIDIA GPU.
- Laptop GPU: a mobile processor incorporated into an original equipment manufacturer’s laptop.
- Server GPU: a data-center accelerator module, such as an H100, H200, or B200.
- Server or rack-scale platform: a system that may combine multiple GPUs with CPUs, memory, cooling, networking, and other components.
- Installed GPU: hardware deployed at a customer’s site or in a cloud provider’s fleet. It may be counted at a different time from NVIDIA’s shipment or revenue recognition.
A sale of a complete system is not the same unit as a GPU package. Nor does a customer renting cloud GPU capacity necessarily buy or control a physical GPU.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
#1 Best Overall
- AI Performance: 767 AI TOPS
- OC mode: 2632 MHz (OC mode)/ 2602 MHz (Default mode)
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Axial-tech fan design features a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure
- A 2.5-slot design maximizes compatibility and cooling efficiency for superior performance in small chassis
What NVIDIA reports—and what it does not
NVIDIA reports revenue by business area, including Data Center, Gaming, Professional Visualization, and Automotive. Its SEC filings do not provide a consolidated total for GPU dies, boards, modules, systems, or deployed GPUs. The company also sells through add-in-board manufacturers, distributors, original design manufacturers, original equipment manufacturers, cloud providers, hyperscalers, and system integrators, so its recognized revenue does not map neatly to retail purchases. NVIDIA’s FY2026 filing describes the revenue categories, while an earlier filing explains the range of direct customer types.
| Period | Reported revenue | What it tells us about units |
|---|---|---|
| Fiscal 2025 | $130.5 billion total; $115.2 billion Data Center; about $11.4 billion Gaming; $1.7 billion Automotive | Revenue by market, not GPU quantities. |
| Fiscal 2026, ended January 25, 2026 | $215.9 billion total; $16.0 billion Gaming; $3.2 billion Professional Visualization; approximately $193.7 billion Data Center based on the disclosed annual market categories | The Data Center figure is not GPU-only revenue; the company does not publish an equivalent unit count. |
| Q1 fiscal 2027, ended April 26, 2026 | $81.6 billion total; $75.2 billion Data Center, including $60.4 billion compute and $14.8 billion networking | Even within Data Center, a material portion is networking rather than compute accelerators. |
The fiscal 2026 and Q1 fiscal 2027 figures come from NVIDIA’s fiscal 2026 results and first-quarter fiscal 2027 results. NVIDIA had scheduled its Q2 fiscal 2027 results for August 26, 2026, so they were not available by the August 16 cutoff.
The most defensible unit estimate: desktop graphics cards
For desktop discrete graphics cards, external industry reporting offers a basis for an approximate estimate. Global desktop graphics-card shipments were reported at about 44.28 million units in calendar 2025, and NVIDIA’s share of that market was reported at roughly 94% to 95%. Applying those figures gives:
Rank #2
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5070 Ti
- Integrated with 16GB GDDR7 256bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
44.28 million desktop graphics cards × 94%–95% NVIDIA share = about 41.6 million–42.1 million NVIDIA-based boards.
This is an industry shipment estimate, not a number NVIDIA disclosed. The underlying figures were reported in Tom’s Hardware coverage of 2025 desktop graphics-card shipments. It primarily estimates desktop add-in boards and may reflect shipments into the channel rather than cards purchased by consumers.
It excludes laptop GPUs, workstation products, data-center accelerators, console processors, and integrated graphics. It also comes from a year with an RTX 40-to-RTX 50 product transition, which can shift shipment timing. It is useful as a desktop-board estimate, not as a proxy for NVIDIA’s total GPU production.
Rank #3
- Powered by the NVIDIA Blackwell architecture and DLSS 4. System Requirements: Minimum 850W PSU with 16-pin 12V-2x6 (12VHPWR) connector required. Verify before purchasing.
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability. Compatibility: 348mm (13.7") length, 3.6 slots, 4.3 lbs. Confirm case clearance and slot spacing. GPU bracket included.
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.6-slot design with massive fin array optimized for airflow from three Axial-tech fans
- Phase-change GPU thermal pad helps ensure optimal thermal performance and longevity, outlasting traditional thermal paste for graphics cards under heavy loads
Why Gaming revenue cannot produce an exact card count
NVIDIA reported $16.0 billion in Gaming revenue for fiscal 2026, up 41% year over year. Dividing that revenue by one card’s retail price would not reveal how many cards NVIDIA sold. The Gaming category spans different price tiers and includes desktop and laptop products; revenue also reflects sales through OEMs and channel partners rather than only consumer checkout transactions. Product mix, regional pricing, promotions, rebates, and the timing of a new architecture affect the amount recognized for each unit.
One $299-class board and one $1,999-class board each count as one unit but contribute very different revenue. Some high-end gaming-class GPUs may also be used for AI or professional workloads, so the business label does not necessarily describe the end use. Console processors and related products further complicate any attempt to treat Gaming revenue as desktop-card revenue.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Why data-center GPU sales are much harder to count
NVIDIA’s Data Center business is now a platform business. It includes compute products such as Hopper and Blackwell accelerators, CPUs such as Grace, NVLink, InfiniBand and Ethernet networking, and complete servers and rack-scale systems. NVIDIA’s filings describe growth from networking alongside compute, and fiscal 2026 results discuss the ramp of Blackwell systems and related infrastructure. A Data Center revenue total therefore cannot be relabeled as “AI GPU revenue.”
Rank #4
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5060
- Integrated with 8GB GDDR7 128bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
System configurations also vary. An HGX board may contain multiple GPUs; an eight-GPU server contains more than one accelerator; and a rack-scale platform can combine many compute GPUs with CPUs and networking. A cloud customer may then rent access to that hardware without buying it. NVIDIA has described AI installations ranging from hundreds to tens of thousands of GPUs, with some reaching approximately 100,000, but those customer-scale examples describe installations or plans—not NVIDIA’s shipments in a reporting period. NVIDIA’s investor presentation provides that deployment context.
What a revenue-to-unit model can—and cannot—show
A scenario calculation can illustrate why a precise data-center unit answer is elusive. NVIDIA reported $60.4 billion of Data Center compute revenue in Q1 fiscal 2027. If that amount were divided by various assumed realized values per GPU-equivalent, the arithmetic would look like this:
| Assumed realized value per GPU-equivalent | Units represented by $60.4 billion of Q1 FY2027 compute revenue |
|---|---|
| $20,000 | 3.02 million |
| $30,000 | 2.01 million |
| $40,000 | 1.51 million |
| $50,000 | 1.21 million |
| $75,000 | 805,000 |
These are illustrative calculations, not estimates of actual shipments. Compute revenue can include CPUs and complete systems; systems contain multiple GPUs; product prices and NVIDIA’s realized revenue vary; and revenue may be recognized at different stages of delivery. The assumed value is not a published NVIDIA average selling price. Changing that assumption alone produces a wide range, before resolving what exactly counts as a unit.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchBest Value
- Powered by the NVIDIA Blackwell architecture and DLSS 4 OC mode: 2640MHz/Default mode: 2610MHz (Boost Clock)
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.125-slot design with massive fin array optimized for airflow from three Axial-tech fans
- Phase-change GPU thermal pad helps ensure optimal thermal performance and longevity, outlasting traditional thermal paste for graphics cards under heavy loads
Demand, orders, shipments, and deployments are different measures
Statements about strong demand or sold-out cloud capacity do not establish a shipment total. There are several stages between interest and a GPU actively serving workloads:
- Demand: customers want capacity or signal interest.
- Orders and backlog: customers commit or wait for supply; these do not necessarily become immediate shipments.
- Production and shipment: NVIDIA, manufacturing partners, or system builders produce and deliver components or systems.
- Revenue recognition: NVIDIA records revenue when accounting conditions are met; this need not coincide with end deployment.
- Installation and utilization: equipment is deployed and may be active, reserved, virtualized, or underused.
NVIDIA has cited supply constraints, product transitions, and shipment timing effects; for example, its fiscal 2025 commentary discussed constrained supply during the Blackwell ramp. The company’s Q4 fiscal 2025 CFO commentary provides that context. Later, NVIDIA management described Blackwell sales as exceptionally strong and cloud GPUs as sold out, but that statement is an indicator of demand pressure, not a reported unit count. NVIDIA’s Q3 fiscal 2026 release attributes the claim to CEO Jensen Huang.
How much can we say with confidence?
| Question | Best answer | Confidence |
|---|---|---|
| Does NVIDIA disclose total GPU units? | No consolidated GPU-unit total is reported in its public filings. | High |
| How many NVIDIA desktop add-in boards shipped in 2025? | Approximately 41.6 million–42.1 million, derived from external market estimates. | Moderate |
| How many consumer GPUs, including laptops, did NVIDIA sell? | No comparable public total establishes that figure. | Low to moderate |
| How many data-center GPUs did NVIDIA sell? | Not publicly disclosed; estimates require product- and system-specific assumptions. | Low for any single unit estimate |
| How many GPUs did NVIDIA sell across every business? | No defensible single public number is available. | High |
The practical answer is that NVIDIA ships tens of millions of desktop GPU boards in a year, while its data-center accelerator volume cannot be reliably inferred from revenue alone. Its extraordinary revenue reflects not just more processors, but increasingly the systems, networking, and infrastructure built around them.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
