NVIDIA held 75.97% of PC video card usage by manufacturer in Valve’s September 2026 Steam Hardware & Software Survey, ahead of AMD at 16.67% and Intel at 7.01%. PassMark Software’s count of video cards in use over the last 30 days puts NVIDIA at 55.8%, AMD at 20.1% and Intel at 18.5%. GPU market share therefore depends on what is counted: survey answers or benchmark runs.
In data centers, NVIDIA reported Data Center revenue of $89.0 billion for the quarter ended July 26, 2026, up 117% from a year ago. AMD’s Data Center segment, driven by EPYC processors and Instinct GPUs, reported $6.7 billion in the three months ended June 27, 2026, up 107% year over year. On the rental side, Lambda lists the NVIDIA H100 SXM at $3.99 per GPU per hour.
Key Takeaways
- NVIDIA’s share is 20.17 points higher among Steam survey participants in September 2026 (75.97%) than in PassMark’s 30-day count of cards in use (55.8%).
- Intel holds 18.5% of video cards in use in PassMark’s count of the last 30 days, against 7.01% among Steam survey participants in September 2026.
- Data Center revenue made up 92.5% of NVIDIA’s $96.2 billion total in the quarter ended July 26, 2026.
- AMD’s Data Center business represented 58% of company revenue in the second quarter of 2026, when total revenue rose 50% year over year to $11.5 billion.
- Jon Peddie Research (JPR) counted 12.5 million PC graphics add-in boards (AIBs) shipped in Q2 2026, up 10.0% from the previous quarter and 6.6% year over year.
- 169 of the 276 accelerated systems on the June 2026 TOP500 list use NVIDIA Hopper or Ampere chips.
- CoreWeave’s spot price for an 8-GPU HGX H100 node is about 40% of its $49.24 on-demand hourly rate.
Editor’s Choice
- NVIDIA fiscal 2026 Data Center revenue: $193.7 billion, up 68% for the full year.
- NVIDIA third-quarter fiscal 2027 outlook: $108.0 billion in revenue, plus or minus 2%, with no Data Center compute revenue from China assumed.
- PC GPUs shipped in Q2 2026: 75.5 million, integrated and discrete, up 10.4% quarter-to-quarter.
- GeForce RTX 5070 on Steam, September 2026: 6.15% of all video cards in the survey.
- Lambda B200 SXM6 rental: $6.69 per GPU per hour.
- Amazon Web Services (AWS) P5 (H100) price cut: 44% on On-Demand from the May 31, 2025 baseline.
- CoreWeave HGX B200 node: $68.80 per hour on-demand.
GPU Market Share by Vendor
- NVIDIA accounted for 75.97% of PC video card usage among Steam Hardware & Software Survey participants in September 2026.
- PassMark’s 55.8% for NVIDIA counts the baselines submitted during the last 30 days, which the company says represents video cards in use rather than purchased.
- NVIDIA leads AMD by 59.3 points in the Steam survey, against 35.7 points in PassMark’s count.
- Jon Peddie Research presents AIB market share in a chart and states the quarter-over-quarter changes in its release text.
- Neither Steam nor PassMark counts sales.
- PassMark rounds its shares to one decimal, so its four vendor shares add to 100.1%.
| Vendor | Steam survey, September 2026 (%) | PassMark, 30 days to October 13, 2026 (%) |
|---|---|---|
| NVIDIA | 75.97 | 55.8 |
| AMD | 16.67 | 20.1 |
| Intel | 7.01 | 18.5 |
| Other | 0.35 | 5.7 |
Source: Valve (Steam Hardware & Software Survey, September 2026), PassMark Software (October 2026)
About This Data
- Compiled from 14 sources: 3 company earnings releases filed with the SEC, 4 Jon Peddie Research releases, 2 Steam survey pages, 1 PassMark benchmark page, the TOP500 list highlights, 2 cloud price lists and 1 AWS price announcement.
- Sources date from June 2025 to October 13, 2026.
- Only company filings and price lists, analyst releases, a user survey, a benchmark database, and the TOP500 list qualified; paid market-research reports were excluded.
- Figures are reviewed on a rolling basis and updated when sources publish new editions.
NVIDIA Data Center GPU Revenue
- In its latest quarterly results, NVIDIA reported revenue of $96.2 billion for the quarter ended July 26, 2026, up 18% from the previous quarter and 106% from a year ago.
- Revenue in the second quarter of fiscal 2027 was $96,221 million ($96.2 billion), against $81,615 million ($81.6 billion) a quarter earlier.
- Data Center revenue was $89.0 billion, or 92.5% of the $96.2 billion total, up 18% from the previous quarter and 117% from a year ago.
- NVIDIA’s Data Center figure is about 13.3 times AMD’s $6.7 billion Data Center segment, though AMD’s quarter ended a month earlier and its segment includes EPYC CPUs.
- NVIDIA expects third-quarter fiscal 2027 revenue of $108.0 billion, plus or minus 2%, and assumes no Data Center compute revenue from China in that outlook.
- In the fourth quarter of fiscal 2026, ended January 25, 2026, Data Center revenue was a record $62.3 billion, up 75% from a year ago.
- For the full fiscal year 2026, Data Center revenue rose 68% to a record $193.7 billion.
- Edge Computing revenue was $7.2 billion in the second quarter, up 27% from a year ago, which, with Data Center’s $89.0 billion, matches the $96.2 billion total.
- NVIDIA said its record fourth-quarter Data Center revenue was driven by the major platform shifts to accelerated computing and AI.
View the data
Recent Developments
- October 13, 2026: PassMark’s 30-day count put NVIDIA at 55.8% of video cards in use.
- September 2026: The GeForce RTX 5070 rose 2.38 points to 6.15% of all video cards in Steam’s survey.
- September 9, 2026: JPR reported that AIB shipments were up 10% from the previous quarter to 12.5 million units, with the attach rate in desktop PCs at 89%.
- August 26, 2026: NVIDIA said its Vera Rubin platform is ramping into full production, with racks running at partners including CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius.
- August 4, 2026: AMD reported second-quarter Data Center segment revenue of $6.7 billion, up 107% year over year.
- June 2026: The TOP500 list counted 276 systems using accelerator or co-processor technology, up from 255 six months earlier.
AMD Data Center Segment Revenue
- AMD’s Data Center segment revenue was $6.7 billion ($6,718 million) in the three months ended June 27, 2026, up 107% year over year on demand for EPYC processors and Instinct GPUs.
- Segment revenue for the six months ended June 27, 2026 was up 80.7% from the six months ended June 28, 2025.
- AMD’s Data Center business represented 58% of company revenue in the quarter, when total revenue rose 50% year over year to a record $11.5 billion.
- AMD’s year-ago quarter included $800 million in inventory and related charges due to the US government’s export control on Instinct MI308 data center GPU products.
- Data Center revenue exceeded AMD’s Client and Gaming segment, which reported $3.8 billion: Client business revenue was $3.1 billion, up 23% year over year, while Gaming business revenue fell 31% year over year to $779 million on lower semi-custom revenue.
View the data
NVIDIA vs AMD Data Center Revenue
- NVIDIA accounted for about 93% of the two companies’ combined Data Center revenue in their latest reported quarters.
- NVIDIA’s Data Center revenue grew 18% from the previous quarter, while AMD’s Data Center segment grew 16.3% from its previous quarter.
- NVIDIA’s Data Center revenue rose 42.9% between the quarter ended January 25, 2026 and the quarter ended July 26, 2026.
- The cloud operators buying these chips report their server and data center budgets, compiled in hyperscaler AI capex figures.
| Company | Segment | Quarter ended | Revenue ($ billions) | Change from a year ago (%) |
|---|---|---|---|---|
| NVIDIA | Data Center | July 26, 2026 | 89.0 | 117 |
| NVIDIA | Data Center | January 25, 2026 | 62.3 | 75 |
| AMD | Data Center (EPYC CPUs and Instinct GPUs) | June 27, 2026 | 6.7 | 107 |
Source: NVIDIA and AMD earnings releases, 2026
Key finding: NVIDIA’s filing puts quarterly Data Center revenue at $89.0 billion, up 117% from a year ago, while AMD’s Data Center segment, which also sells EPYC processors, reported $6.7 billion. Neither company states a GPU market share, so filed revenue is the open yardstick, read with different quarter ends and segment scopes.
What is AMD’s share of the data center GPU market?
Neither company’s earnings release states one. AMD’s $6.7 billion Data Center segment equals about 7% of the two companies’ combined Data Center revenue in their latest quarters. That split includes AMD’s EPYC CPUs and NVIDIA’s whole Data Center platform, so it is a revenue share, not a GPU unit share.
Discrete GPU (AIB) Shipments by Quarter
- JPR counted 12.5 million AIBs shipped worldwide in Q2 2026, up 10.0% from the previous quarter and 6.6% year over year.
- JPR’s Q1 2026 release put AIB shipments at 11.82 million, down 0.6% from a revised fourth quarter. Its Q4 2025 release had first reported 11.48 million, so the chart below, which shows first-reported figures, rises from Q4 to Q1 even though JPR reported a decline.
- Fourth-quarter 2025 AIB shipments were up 36.0% from a year earlier.
- JPR said that quarter’s shipments were down from the previous quarter, which it thinks was due to rising prices attributed to memory costs and tariffs.
- NVIDIA’s Gaming revenue rose 41% to $16.0 billion for fiscal 2026, with $3.7 billion of it in the fourth quarter.
- JPR forecasts a compound annual growth rate of -2.5% for AIBs from 2025 to 2030, with an installed base of 193 million units at the end of the period.
- Its Q1 2026 release forecast -3.3% a year from 2024 to 2029, and its Q4 2025 release -5.9% a year from 2024 to 2028.
- Dr. Jon Peddie, president of JPR, said that, “Defying common wisdom, high-end AIB sales spiked as prices increased.”
View the data
Desktop GPU Attach Rate
- The AIB attach rate in desktop PCs rose to 89% in Q2 2026, up 23% from the previous quarter, after a 33.2% rise in Q1 2026 and a 12.3% fall in Q4 2025.
- The desktop PC CPU market decreased 33% year to year and 10.5% quarter to quarter.
- Desktop PC CPU shipments decreased to 14 million units in Q2 2026.
- JPR counted 75.5 million PC GPUs in Q2 2026, up 10.4% quarter to quarter and 1.1% year over year, for a total attach rate of 120% across integrated and discrete GPUs. A rate above 100% means more GPUs shipped than PCs, since one PC can carry both kinds.
- Desktop graphics decreased 4% quarter to quarter while notebooks increased 16.8%.
- Discrete GPUs increased 12.2% in the quarter, according to Dr. Jon Peddie, against 10.0% for AIBs, because JPR’s GPU count covers desktops, notebooks, and workstations while AIBs are add-in boards only.
- JPR expects the penetration of AIBs in desktop PCs to be 130% over the next five years.
- JPR revises earlier quarters, so the chart below shows each release’s first-reported attach rate, and the quarter-over-quarter changes above do not match the chart’s figures.
View the data
NVIDIA, AMD and Intel Discrete GPU Share Changes
- In Q4 2025, NVIDIA’s AIB market share increased by 1.6% from the previous quarter, while AMD’s decreased by 1.6%.
- In Q2 2026, NVIDIA’s AIB share decreased by 0.1% and Intel’s increased by 0.3%.
- Across all PC GPUs, including integrated, AMD’s share increased by 0.6% in Q2 2026 and Intel’s decreased by 1.0%.
| Quarter and measure | AMD (change) | Intel (change) | NVIDIA (change) |
|---|---|---|---|
| Q4 2025, AIB | -1.6% | flat | +1.6% |
| Q1 2026, AIB | -0.04% | +0.4% | -0.4% |
| Q2 2026, AIB | -0.16% | +0.3% | -0.1% |
| Q2 2026, all PC GPUs | +0.6% | -1.0% | +0.46% |
Source: Jon Peddie Research AIB reports, Q4 2025 to Q2 2026, and Jon Peddie Research Market Watch report, Q2 2026
Worth noting: Jon Peddie Research’s Q2 2026 release gives AIB share as quarter-over-quarter changes, such as NVIDIA’s 0.1% decrease, and points to a chart for the levels. A change of a fraction of a point says which way share moved, not how large each vendor’s share is, so these figures cannot be turned into a ranking by size.
Most Used GPUs on Steam
- The NVIDIA GeForce RTX 5070 accounted for 6.15% of all video cards in Steam’s September 2026 survey, up 2.38 points from August 2026.
- The RTX 5060 Ti rose 1.51 points to 4.06%, and the RTX 5060 rose 1.24 points to 4.40%.
- The RTX 3060 slipped 0.26 points to 3.66%, and the RTX 4060 Laptop GPU 0.34 points to 3.50%.
- The RTX 5070, RTX 5060 and RTX 5060 Ti together held 14.61% of Steam survey cards in September 2026.
- The RTX 5070’s share rose 3.06 points from 3.09% in May 2026.
View the data
Which GPU has the highest market share?
The NVIDIA GeForce RTX 5070 topped Steam’s September 2026 card table at 6.15% of survey participants’ video cards. By maker, NVIDIA held 75.97% of Steam video card usage and 55.8% in PassMark’s 30-day count. Both are installed-base samples, not sales, so the leader by units shipped can differ.
What are the top 5 GPUs?
In Steam’s September 2026 survey, the first five cards in the all-video-cards table, ordered by share, were the GeForce RTX 5070 (6.15%), RTX 5060 (4.40%), RTX 5060 Ti (4.06%), RTX 4060 (3.89%), and RTX 3060 (3.66%). All five are NVIDIA cards, and the ranking reflects Steam users who answered the survey, not cards sold.
Supercomputer GPU Share (TOP500)
- 276 systems on the June 2026 TOP500 list use accelerator or co-processor technology, up from 255 six months earlier.
- That is 55.2% of the list’s 500 systems.
- NVIDIA Hopper chips power 107 of the accelerated systems, about 3.3 for every system using AMD Instinct.
- NVIDIA Ampere chips power another 62, so 169 of the 276 accelerated systems use NVIDIA Hopper or Ampere chips.
- The NVIDIA H200 SXM5 141 GB and the NVIDIA H100 SXM5 80 GB each appear in 26 systems, 5.2% of the list.
- AMD’s Instinct MI250X and Instinct MI300A each appear in 13 systems.
- Host processors are counted separately from accelerators: Intel provides the processors for the largest share of TOP500 systems, 53.20%, down from 57.00% six months ago, while 192 systems (38.40%) use AMD processors.
View the data
The model table below lists the top ten accelerator models.
| Accelerator model | Systems | Share of list (%) |
|---|---|---|
| NVIDIA H200 SXM5 141 GB | 26 | 5.2 |
| NVIDIA H100 SXM5 80 GB | 26 | 5.2 |
| NVIDIA A100 | 23 | 4.6 |
| NVIDIA GH200 Superchip | 21 | 4.2 |
| NVIDIA H100 | 17 | 3.4 |
| NVIDIA A100 SXM4 40 GB | 14 | 2.8 |
| NVIDIA H100 SXM5 94 GB | 14 | 2.8 |
| NVIDIA Tesla V100 | 13 | 2.6 |
| AMD Instinct MI250X | 13 | 2.6 |
| AMD Instinct MI300A | 13 | 2.6 |
Source: TOP500, June 2026 list statistics
H100 Price per Hour
- Lambda lists the NVIDIA H100 SXM with 80 GB of memory at $3.99 per GPU per hour on its 8x self-serve instances.
- On the same price list, the NVIDIA A100 SXM with 80 GB costs $2.79 and the Tesla V100 $0.79 per GPU per hour.
- Lambda prices H100 clusters of 16 GPUs at $6.16 per GPU per hour on terms of 2 weeks to 1 year, with lower rates at 64 and 256 GPUs.
- CoreWeave’s $49.24 hourly on-demand price for an 8-GPU HGX H100 node works out to about $6.16 per GPU-hour, one of the hardware costs behind AI inference prices per token.
- Lambda’s 16-GPU cluster rate is about 1.54 times its self-serve instance rate.
View the data
H200 and B200 Rental Prices
- Cloud computing provider CoreWeave lists an 8-GPU NVIDIA HGX B200 node at $68.80 per hour on-demand and $34.11 per hour on spot.
- That is about $8.60 per B200 GPU-hour on-demand, against Lambda’s $6.69 per GPU per hour for the B200 SXM6.
- CoreWeave’s HGX H200 node works out to about $6.31 per GPU-hour on-demand.
- An H200 node costs $1.20 an hour more than an H100 node on-demand.
- B200 spot capacity costs 49.6% of the on-demand rate, against about 40% for the H100.
View the data
By the numbers: CoreWeave charges $49.24 an hour for an 8-GPU HGX H100 node, $50.44 for an HGX H200 node and $68.80 for an HGX B200 node on-demand. The B200 node costs about 1.4 times the H100 node, so the newer chip’s premium shows in list prices before any performance comparison.
Cloud GPU Price Cuts
- AWS cut On-Demand prices for P5 instances with NVIDIA H100 GPUs by 44% from May 31, 2025 baseline prices.
- For P5, reductions under Savings Plans reached 45%.
- The reductions applied to On-Demand purchases beginning June 1, 2025, and to Savings Plan purchases effective after June 4, 2025.
- AWS’s own standing among cloud providers is measured in cloud provider market share.
View the data
What is NVIDIA’s GPU market share?
NVIDIA held 75.97% of video card usage among Steam survey participants in September 2026 and 55.8% of cards in use in PassMark’s 30-day count. In data centers, it reported $89.0 billion of Data Center revenue for the quarter ended July 26, 2026. Jon Peddie Research’s release text gives NVIDIA’s add-in board share as quarterly changes.
How much does an H100 cost per hour?
Lambda lists the NVIDIA H100 SXM at $3.99 per GPU per hour on self-serve instances, and its H100 clusters run $5.54 to $6.16 per GPU per hour. CoreWeave charges $49.24 an hour on-demand for an 8-GPU HGX H100 node, or $19.71 on spot. These are list rates as of October 13, 2026.
Conclusion
NVIDIA’s 75.97% of Steam survey video cards in September 2026 and its $89.0 billion of Data Center revenue in the quarter ended July 26, 2026 put it ahead on every GPU market share measure here that reports a level, though PassMark’s lower count shows how much the answer depends on the sample. NVIDIA expects third-quarter fiscal 2027 revenue of $108.0 billion, plus or minus 2%, and Jon Peddie Research forecasts a compound annual growth rate of -2.5% for AIBs from 2025 to 2030.
