71 GPU Shortage Statistics for 2026

ByShane Tran Updated on

The last GPU shortage was about gamers versus crypto miners. This one is different. AI data centers now pull on the same memory supply and the same factory capacity as every other chip buyer, and the strain reaches from cloud contracts to the price tag on a graphics card. Here are 71 GPU shortage statistics that show how deep it runs in 2026.

Top GPU Shortage Statistics

  • Nvidia’s supply commitments jumped from $119 billion to $279 billion in a single quarter as of July 2026, primarily to procure memory (Nvidia Q2 FY2027 CFO Commentary).
  • The GeForce RTX 5090 was selling for about $4,900 in the US in late August 2026 — 145% above its $1,999 MSRP (TechSpot).
  • GPU retail prices climbed 15% in a single month between July and August 2026 (TechSpot).
  • By late October 2025, SK hynix had already sold out its entire 2026 DRAM and NAND production to customer demand (SK hynix, 3Q25 results).
  • SK hynix CEO Kwak Noh-jung forecast in July 2026 that 2027 will be the worst year in the memory industry’s history from a supply perspective (Reuters).
  • Nvidia’s data center segment brought in a record $89.0 billion in the quarter ended July 2026, up 117% from a year earlier (Nvidia Q2 FY2027 CFO Commentary).
  • PC DRAM contract prices were projected to jump more than 100% quarter over quarter in 1Q26, a record quarterly surge (TrendForce, Feb 2026).

How Bad Is the GPU Shortage in 2026?

1. Graphics cards were selling for roughly 28% above launch MSRP on average across 10 countries as of late August 2026, excluding the RTX 5090 (TechSpot).

2. GPU retail prices climbed 15% in a single month between July and August 2026, with GeForce cards up 19% and Radeon cards up 11% (TechSpot).

GPU Prices vs MSRP in 2026

The gap between sticker price and street price varies by model. US and European trackers point the same way.

3. The GeForce RTX 5090 was selling for about $4,900 in the US in late August 2026 — 145% above its $1,999 MSRP, and 136% above MSRP averaged globally (TechSpot).

4. The RTX 5060 Ti 16GB sold for 70% above MSRP in the US (about $730) and 53% above MSRP globally as of late August 2026 (TechSpot).

5. AMD’s RX 9070 XT was selling for about $750 in the US in late August 2026, 25% above its MSRP (TechSpot).

Source: TechSpot — lowest in-stock US prices per model, collected late August 2026.

6. The lowest in-stock US price for the RTX 5090 was $5,199 as of September 3, 2026, versus a $1,999 MSRP (Tom’s Hardware GPU price tracker).

7. The cheapest RTX 5080 listed at $1,579 versus a $999 MSRP, and the RTX 5070 Ti at $1,149 versus $749, in early September 2026 (Tom’s Hardware GPU price tracker).

8. The RTX 5070’s lowest US price stood at $799 on September 3, 2026, versus its $549 MSRP (Tom’s Hardware GPU price tracker).

Source: Tom’s Hardware GPU price tracker — lowest in-stock US price per model, September 3, 2026.

European retail tells the same story.

9. Graphics card prices in German retail rose 13.3% in a single month from July to August 2026, the largest increase of any product group in 3DCenter’s memory-crisis price index, and stood 32% above their September 2025 baseline (3DCenter, August 2026 price index).

10. In German retail, the RTX 5090’s best price hit €4,549 in August 2026, 96% above its September 2025 level of €2,319 (3DCenter, August 2026 price index).

11. Within 3DCenter’s German retail index, Nvidia card prices rose 15.6% month over month in August 2026 (to 37% above baseline) versus 9.1% for AMD cards (22% above baseline) (3DCenter, August 2026 price index).

Supply-side pricing is moving too.

12. AMD reportedly notified its add-in board partners of a roughly 10% price increase on GPU-and-GDDR bundle kits effective July 2026 (TrendForce News, July 2026, via Board Channels).

AI Data Center GPU Demand

The demand side of the shortage starts in the data center. “Our packaging capacity is so tight that now it limits my customers’ growth,” TSMC CEO C.C. Wei told analysts in July 2026 (TSMC Q2 2026 earnings call).

13. Nvidia’s total revenue hit $96.2 billion in the quarter ended July 2026, up 106% year over year (Nvidia Q2 FY2027 press release).

14. Nvidia’s data center segment brought in a record $89.0 billion in the quarter ended July 2026, up 117% from a year earlier (Nvidia Q2 FY2027 CFO Commentary).

15. Nvidia guided revenue for its next quarter to $108.0 billion, plus or minus 2% (Nvidia Q2 FY2027 press release).

16. Nvidia said hyperscale revenue more than doubled year over year in the July 2026 quarter on the strength of Blackwell Ultra (Nvidia Q2 FY2027 CFO Commentary).

17. Nvidia’s supply commitments jumped from $119 billion to $279 billion in a single quarter as of July 2026, primarily to procure memory (Nvidia Q2 FY2027 CFO Commentary).

That’s one company booking components years ahead.

18. TSMC raised its full-year 2026 revenue growth outlook to slightly above 40% in US dollar terms (TSMC Q2 2026 earnings call).

19. TSMC raised its 2026 capital budget to $60–64 billion (TSMC Q2 2026 earnings call).

20. TSMC’s high-performance computing platform grew 20% quarter over quarter to account for 66% of its revenue in Q2 2026 (TSMC Q2 2026 earnings call).

21. TrendForce raised its 2026 global AI server shipment growth forecast to nearly 31% year over year, up from 28% (TrendForce, Aug 2026).

Hyperscaler Spending on AI Infrastructure

The world’s largest cloud providers publish their capital spending every quarter.

22. Microsoft spent $35.8 billion on property and equipment in the quarter ended June 2026 (Microsoft FY26 Q4 earnings release).

23. Microsoft’s Azure and other cloud services revenue grew 43% year over year in the June 2026 quarter (Microsoft FY26 Q4 earnings release).

24. Alphabet’s capital expenditures reached $44.9 billion in Q2 2026 (Alphabet Q2 2026 earnings release).

25. Google Cloud revenue accelerated 82% year over year to $24.8 billion in Q2 2026 (Alphabet Q2 2026 earnings release).

26. Meta’s capital expenditures came to $31.08 billion in Q2 2026 (Meta Q2 2026 earnings release).

27. Meta expects full-year 2026 capital expenditures of $130–145 billion, narrowed upward from its prior outlook (Meta Q2 2026 earnings release).

28. Amazon’s purchases of property and equipment totaled $54.2 billion in Q2 2026 alone (Amazon Q2 2026 earnings release).

29. Amazon CEO Andy Jassy said AWS grew 36.7% year over year in Q2 2026, its fastest growth in 18 quarters, with its AI and chips businesses each passing $25 billion run rates (Amazon Q2 2026 earnings release).

The trajectory was already set in late 2025.

30. Alphabet raised its 2025 capital expenditure guidance to $91–93 billion, up from a previous estimate of $85 billion (Alphabet Q3 2025 earnings call).

Industry forecasts put numbers on the combined picture.

31. Combined capex of the world’s nine largest cloud service providers is expected to jump about 90% year over year in 2026 on AI infrastructure investment (TrendForce, Aug 2026).

32. TrendForce projects combined capex of the top nine cloud providers will reach about $1.3 trillion in 2027, growing nearly 50% year over year (TrendForce, Aug 2026).

33. Combined capex of ByteDance, Tencent, Alibaba, and Baidu is forecast to rise more than 80% year over year in 2026 (TrendForce, Aug 2026).

Cloud GPU Capacity Constraints

Cloud providers say demand keeps running ahead of what they can build. Microsoft CFO Amy Hood told analysts in October 2025 that the company expected to be “capacity constrained through at least the end of our fiscal year” (Microsoft FY26 Q1 earnings call), and Alphabet’s CFO said Google Cloud expected to remain in a “tight demand-supply environment” through Q4 2025 and 2026 (Alphabet Q3 2025 earnings call).

34. Microsoft planned to increase its total AI capacity by over 80% in fiscal 2026, per CEO Satya Nadella in October 2025 (Microsoft FY26 Q1 earnings call).

35. Roughly half of Microsoft’s capital spend in the September 2025 quarter went to short-lived assets, primarily GPUs and CPUs for Azure demand (Microsoft FY26 Q1 earnings call).

36. AWS added more than 3.8 gigawatts of power capacity in the 12 months to October 2025, more than any other cloud provider, per CEO Andy Jassy (Amazon Q3 2025 earnings call).

37. As of late 2025, AWS had double the power capacity it had in 2022 and said it was on track to double again by 2027 (Amazon Q3 2025 earnings call).

38. Meta planned to end 2025 with more than 1.3 million GPUs and to invest $60–65 billion in capital expenditure that year, per Mark Zuckerberg in January 2025 (Mark Zuckerberg on Facebook).

39. Meta is building a 2-gigawatt-plus AI data center that Zuckerberg said would cover a significant part of Manhattan (Meta Q4 2024 earnings call).

40. Announcing GPT-4.5 in February 2025, OpenAI CEO Sam Altman said the company was “out of GPUs” and would add tens of thousands of GPUs the following week (Sam Altman on X, Feb 2025).

41. Sam Altman said in July 2025 that OpenAI would cross well over 1 million GPUs brought online by the end of 2025 (Sam Altman on X, July 2025).

42. xAI says it built its Colossus supercomputer in 122 days, then doubled it in another 92 days to 200,000 GPUs (xAI).

Renting a GPU carries its own price tag.

43. On-demand cloud rental of an Nvidia H100 SXM cost $3.99 per GPU-hour on Lambda as of September 8, 2026 (Lambda GPU Cloud pricing).

44. Lambda listed on-demand Nvidia B200 capacity at $6.69 per GPU-hour as of September 8, 2026 (Lambda GPU Cloud pricing).

45. The GPU-as-a-Service market is projected to grow from $8.21 billion in 2025 to $26.62 billion by 2030, a 26.5% compound annual growth rate (MarketsandMarkets).

How Many GPUs Are Shipping?

46. Global PC GPU shipments grew to 75.5 million units in Q2 2026, up 10.4% quarter over quarter and 1.1% year over year (Jon Peddie Research, Q2 2026).

47. Discrete GPU shipments rose 12.2% quarter over quarter in Q2 2026 even as a global memory crisis sent component prices soaring, according to JPR president Jon Peddie (Jon Peddie Research, Q2 2026).

48. Desktop GPU shipments fell 4% quarter over quarter in Q2 2026 while notebook GPU shipments rose 16.8% (Jon Peddie Research, Q2 2026).

49. Global graphics add-in board shipments reached 11.8 million units in Q1 2026, down 0.6% from the previous quarter (Jon Peddie Research, Q1 2026 AIB report).

50. AIB shipments in Q1 2026 were up 8.3% year over year, with JPR attributing the sequential dip to “higher prices and limited supply due to limited memory and disruptions in the supply chain” (Jon Peddie Research, Q1 2026 AIB report).

51. JPR forecasts the add-in board market to shrink at a 3.3% compound annual rate from 2024 to 2029, ending with an installed base of 183 million units (Jon Peddie Research, Q1 2026 AIB report).

52. Nvidia held 94% of the desktop add-in-board GPU market in Q4 2025, versus 5% for AMD and 1% for Intel (Jon Peddie Research, via VideoCardz).

Steam’s monthly hardware survey shows what gamers actually run.

53. The five-year-old GeForce RTX 3060 was still the most common graphics card among Steam users in August 2026, in 3.92% of surveyed systems (Steam Hardware Survey, August 2026).

54. The RTX 5070 was the most widely adopted RTX 50-series card on Steam in August 2026, at 3.77% of surveyed systems (Steam Hardware Survey, August 2026).

55. The flagship RTX 5090 appeared in just 0.43% of Steam-surveyed systems in August 2026 (Steam Hardware Survey, August 2026).

The Memory Shortage Behind the GPU Shortage

GPU makers and board partners keep pointing at the same input: memory. “Both AI and traditional server demand are constrained by lack of adequate DRAM and NAND supply,” Micron CEO Sanjay Mehrotra said in March 2026 (Micron Q2 FY2026 prepared remarks).

56. Conventional DRAM contract prices were forecast to rise 8–13% quarter over quarter in 4Q25, widening to 13–18% once HBM is included (TrendForce, Sept 2025).

57. TrendForce revised its 1Q26 forecast for conventional DRAM contract prices from 55–60% up to 90–95% quarter-over-quarter growth (TrendForce, Feb 2026).

58. PC DRAM contract prices were projected to jump more than 100% quarter over quarter in 1Q26, a record quarterly surge (TrendForce, Feb 2026).

59. Server DRAM prices were forecast to rise around 90% quarter over quarter in 1Q26, the largest quarterly increase on record (TrendForce, Feb 2026).

60. Desktop DDR5 memory in German retail cost 4.9 times its pre-crisis level by August 2026, up 386% since July 2025 (3DCenter, August 2026 price index).

61. By late October 2025, SK hynix had already sold out its entire 2026 DRAM and NAND production to customer demand (SK hynix, 3Q25 results).

62. Riding AI-driven memory demand, SK hynix posted 24.45 trillion won in revenue and 11.38 trillion won in operating profit in Q3 2025, another record quarter (SK hynix, 3Q25 results).

63. Micron’s revenue hit $23.9 billion in its fiscal Q2 2026, up 196% year over year (Micron Q2 FY2026 prepared remarks).

When Will the GPU Shortage End?

64. SK hynix CEO Kwak Noh-jung forecast in July 2026 that 2027 will be the worst year in the memory industry’s history from a supply perspective (Reuters).

65. The same forecast sees memory demand outstripping SK hynix’s supply capacity beyond 2030 (Reuters).

Those forecasts cover memory supply, not graphics cards directly. Memory is the input GPU makers point to for the current card shortage.

The 2020-2022 GPU Shortage: How It Compares

The previous shortage came with a different kind of buyer: crypto miners. In August 2021, Nvidia CEO Jensen Huang told analysts: “I would expect that we will see a supply-constrained environment for the vast majority of next year” (Nvidia Q2 FY2022 earnings call, via MarketBeat).

66. Jon Peddie Research estimated that about 25% of graphics cards shipped in Q1 2021 — roughly 700,000 high-end and midrange boards — went to crypto miners and speculators (Jon Peddie Research, 2021).

67. Those mining-bound graphics cards carried a market value of about $500 million in Q1 2021 alone (Jon Peddie Research, 2021).

68. In April 2021, the RTX 3060 Ti averaged $1,375 on eBay, 3.4 times its launch price, and the RTX 3080 averaged $2,199, 3.1 times its $699 MSRP (Tom’s Hardware eBay price index).

69. During the crypto-era shortage, German retail GPU prices peaked at 194% above normal in May 2021 and took until July 2022, a bit more than a year, to return to normal (3DCenter, March 2026 price index).

70. Nvidia’s dedicated Cryptocurrency Mining Processor line brought in $550 million in fiscal 2022, but just $24 million of that came in the final quarter (Nvidia Q4 FY2022 CFO Commentary).

71. By May 2022 the crypto-era shortage was unwinding: RTX 30-series street prices in European retail averaged 14% over MSRP and Radeon RX 6000 cards just 6% over (3DCenter, via Wccftech).

Conclusion

Two shortages are running at once: one in the data center, one on store shelves. Suppliers on both sides say the same input is short: memory. Capacity is being added, but the contracts being signed assume tight supply for years. We’ll update this page as new data comes out.