GPU Shipment Statistics 2026
By Axis Intelligence Research
Co-authors: Sarah Mitchell (AI/data center GPU commentary) & Alex Rivera (consumer GPU commentary) | Last updated: July 7, 2026 | License: CC BY 4.0
The global GPU market shipped 44.28 million discrete desktop graphics cards in 2025, a second-decade high, while NVIDIA simultaneously generated $193.74 billion in data center revenue in fiscal year 2026 — a figure that exceeds the entire semiconductor industry’s annual revenue from as recently as 2017. The market has bifurcated into two planets: a consumer AIB segment under pressure from memory shortages, tariffs, and integrated graphics, and an AI-accelerator segment that cannot be built fast enough.
Quick Answer
According to Axis Intelligence Research’s analysis of Jon Peddie Research quarterly data, SEC EDGAR filings, and primary market research:
- 44.28 million discrete desktop GPUs (AIBs) shipped in calendar 2025 — the second-highest annual total this decade
- NVIDIA holds 90% of the discrete AIB market as of Q1 2026, down from a record 94% in Q4 2025
- 70.3 million total PC GPUs (including integrated graphics) shipped in Q1 2026, down 7.5% sequentially but up 2% year-over-year
- Data center GPU market: $125 billion in 2025, projected at $138.88 billion in 2026, growing at 20.7% CAGR through 2034
- Jon Peddie Research projects desktop AIB shipments to fall ~10% in 2026 versus 2025, driven by GDDR7 memory shortages and US tariffs
Key Findings
- NVIDIA’s discrete GPU dominance reached a historical peak of 94% in Q4 2025, up 10 percentage points year-over-year, as AMD’s Radeon RX 9000-series launch failed to recover market share — AMD exited 2025 at a record-low 5% of AIB shipments (Jon Peddie Research, Q4 2025).
- According to Axis Intelligence Research’s GPU Market Divergence Index™ (GMDI™) — a composite of discrete AIB shipment trajectory versus data center GPU revenue growth — the consumer and AI GPU markets are diverging at the widest rate ever recorded: discrete AIB units are forecast to contract ~10% in 2026 while data center GPU revenue grows 11% in the same period.
- NVIDIA’s data center revenue reached $193.74 billion in fiscal year 2026 (ended January 25, 2026), per SEC EDGAR filings — meaning the data center GPU business alone is larger than the entire semiconductor revenues of Intel, AMD, and Qualcomm combined.
- Desktop GPU shipments outperformed the broader PC market decisively in Q1 2026: AIBs fell only 0.6% sequentially versus a 25% year-over-year collapse in total desktop PC shipments, with the GPU-to-desktop attach rate rising to 76% — a 33-percentage-point quarterly jump, per Jon Peddie Research.
- According to Axis Intelligence Research’s analysis of AMD’s 10-K filing (December 2025) and Q1 2026 10-Q, AMD’s data center segment revenue grew 57% year-over-year to $5.8 billion in Q1 2026 driven by Instinct MI350 GPU accelerator demand — but its consumer discrete GPU share simultaneously collapsed to historic lows, illustrating a vendor-level divergence mirroring the market-level split.
How Many GPUs Are Shipped Per Year?
The honest answer depends entirely on which GPUs you count. Jon Peddie Research tracks three overlapping markets: total PC GPUs (integrated + discrete, all platforms), add-in boards (discrete desktop cards only, the AIB market), and broader GPU shipments including notebooks, workstations, and embedded. These numbers are not interchangeable, and conflating them produces the most common GPU statistic errors online.
Total PC GPU Shipments (All Types)
Jon Peddie Research’s Market Watch report, the authoritative primary source for PC GPU shipment data since 1984, tracks total shipments across all PC platforms — desktops, notebooks, workstations — including both integrated and discrete graphics.
| Quarter | Total PC GPUs Shipped | QoQ Change | YoY Change | Source |
|---|---|---|---|---|
| Q3 2025 | 76.6 million units | +2.5% | +4.0% | Jon Peddie Research |
| Q4 2025 | ~75.6 million units | -1.3% | -3.3% | Jon Peddie Research (est.) |
| Q1 2026 | 70.3 million units | -7.5% | +2.0% | Jon Peddie Research |
In Q1 2026, the PC GPU market reached 70.3 million units — down 7.5% from Q4 2025 but still up 2% year-over-year. The sequential decline is steeper than the historical Q1 average because of memory price pressure and supply-chain instability from ongoing tariff fluctuations.
Jon Peddie Research’s five-year forecast revised to a compound annual growth rate of -3% from 2025 to 2029, with the installed base reaching approximately 3 billion units by end of forecast period. Discrete GPU penetration in PCs is projected at 12% over this window — down from earlier forecasts of 25%+, reflecting the faster-than-expected improvement in integrated graphics on both Intel and AMD CPUs.
Desktop Add-in Board (AIB) Shipments — The Discrete GPU Market
The AIB market — discrete graphics cards sold into desktop PCs — is what most people mean when they say “GPU shipments.” This is a smaller, more premium-oriented segment where NVIDIA, AMD, and Intel compete directly.
| Quarter | AIB Units Shipped | QoQ Change | Market Value | Source |
|---|---|---|---|---|
| Q1 2025 | 9.2 million | — | — | Jon Peddie Research |
| Q2 2025 | 11.6 million | +27% | — | Jon Peddie Research |
| Q3 2025 | 12.02 million | +2.8% | $8.8 billion | Jon Peddie Research |
| Q4 2025 | 11.48 million | -4.4% | — | Jon Peddie Research |
| Q1 2026 | 11.82 million | +0.6% | — | Jon Peddie Research |
Calendar year 2025 total: 44.28 million discrete desktop AIBs — up from 34.7 million in 2024, the second-highest annual total this decade, driven primarily by NVIDIA’s GeForce RTX 50-series Blackwell consumer launch.
The Q2 2025 spike (+27% sequentially) is notable and was explicitly attributed by Jon Peddie Research president Dr. Jon Peddie to tariff-front-running — buyers accelerating purchases ahead of US import tariffs: “AIB prices dropped for midrange and entry-level, while high-end AIB prices increased, and most retail suppliers ran out of stock. This is very unusual for the second quarter.” The panic-buy cycle pulled forward demand that suppressed Q3 2025 growth.
Q1 2026’s AIB result — down only 0.6% from Q4 despite a 25% collapse in total desktop PC shipments — is the single most counterintuitive number in recent GPU history. The explanation: the GPU-to-desktop attach rate jumped to 76%, up 33.2% quarter-over-quarter. Buyers are choosing discrete GPU upgrades over new system purchases, which compresses the denominator without collapsing the AIB numerator.
Alex Rivera, Axis Intelligence Research consumer tech correspondent:
I’d been watching this attach-rate number for months, waiting for it to do something interesting. It did. A 76% discrete-GPU attach rate in Q1 2026 — a quarter where desktop PC sales fell off a cliff — tells you something specific about who’s still buying. It’s not families replacing a family computer. It’s enthusiasts replacing a graphics card on a system they plan to keep. The GPU is the upgrade, not the PC. That’s a very different purchase decision, and it’s one the $1,500-plus bracket makes a lot more comfortably than the $700 bracket does. The market is narrowing to the people who care most, which is fine for GPU revenues but does not make for a healthy installed-base story.
GPU Market Share Statistics by Vendor
Discrete AIB Market Share (Desktop Graphics Cards)
NVIDIA’s dominance of the discrete GPU market reached levels that, a few years ago, would have seemed cartoonishly concentrated. In Q4 2024, NVIDIA held approximately 84% of discrete AIBs. By Q4 2025, that figure was 94%.
| Quarter | NVIDIA | AMD | Intel Arc | Source |
|---|---|---|---|---|
| Q1 2025 | 92% | 8% | 0% | Jon Peddie Research |
| Q2 2025 | 94% | 5% | 1% | Jon Peddie Research |
| Q3 2025 | 92% | 7% | 1% | Jon Peddie Research |
| Q4 2025 | 94% | 5% | 1% | Jon Peddie Research |
| Q1 2026 | 90% | 8% | 1% | Jon Peddie Research |
Reading these numbers correctly requires one key disambiguation: the AIB figures above cover discrete add-in boards for desktop PCs only. When integrated graphics are included in the total PC GPU count, Intel leads the overall market (holding approximately 61% in Q3 2025) because Intel ships integrated graphics with the vast majority of CPUs. The AIB share and the total PC GPU share measure different things and should not be compared directly.
AMD’s situation at 5% in Q4 2025 is historically low — the company entered 2025 with 8% AIB share and exited at 5%, despite a new product generation (RDNA 4, Radeon RX 9000 series). The RX 9070 XT received strong reviews — Lisa Su noted it “outsold its predecessors by 10× in its first week” — but strong launches cannot overcome structural inertia when NVIDIA controls the software ecosystem, the streaming and content-creation workflows, and the brand mindshare among GPU buyers.
Q1 2026 shows AMD recovering to 8% while NVIDIA dipped to 90%. This is almost entirely Intel’s gain (+0.4%) and the statistical noise of a single quarter, not a structural AMD recovery. The first AMD discrete GPU appearing in Steam’s May 2026 hardware survey ranked 25th — below 24 NVIDIA models.
Vendor Market Share: Overall PC GPU (Including Integrated Graphics)
| Quarter | NVIDIA | AMD | Intel | Source |
|---|---|---|---|---|
| Q3 2025 | 24% | 15% | 61% | Jon Peddie Research |
| Q1 2026 | ~26% | ~15% | ~59% | Jon Peddie Research (est.) |
Intel’s 61% overall PC GPU share in Q3 2025 reflects the reality that it ships integrated graphics with virtually every PC that doesn’t have a discrete card. NVIDIA’s 24% share in the same quarter looks modest by comparison but represents the premium discrete layer. The gap between NVIDIA’s 92% AIB share and 24% total GPU share is a geometry lesson: there are roughly four times more integrated-GPU PCs than discrete-GPU PCs.
GPU Shipment Statistics by Segment: Consumer vs. AI/Data Center
This is where the story splits. The consumer AIB market and the AI data center accelerator market are nominally the same technology — parallel-compute silicon — but operate on entirely different demand curves, price points, and trajectories.
Consumer GPU Segment: Under Pressure
The consumer discrete GPU market faces simultaneous pressure from three directions:
From below: AMD and Intel continue improving integrated graphics in CPUs, with the latest Ryzen and Core Ultra APUs handling 1080p gaming adequately for casual players. Jon Peddie Research’s revised forecast — discrete GPU penetration at 12% over the next five years versus earlier projections of 25%+ — reflects how quickly integrated graphics has closed the casual gaming gap.
From above: Memory prices (GDDR7 for high-end AIBs, GDDR6 for mid-range) have risen sharply through 2025-2026 as HBM demand from AI accelerator production competes with GDDR supply on the same NAND/DRAM fabrication capacity. Jon Peddie Research attributes this directly: the AIB market is being “squeezed by higher memory prices driven by the AI boom.”
From policy: US tariffs on semiconductor imports — frequently adjusted through 2025-2026 — have created pricing uncertainty that alternates between front-running spikes (Q2 2025) and demand hangover (Q3 2025). Jon Peddie Research’s full-year 2026 forecast calls for approximately a 10% unit decline in desktop AIBs.
The Steam Hardware Survey — Valve’s opt-in census of approximately 130 million active PC gaming accounts — shows the installed-base effect of this pressure. As of April 2026, the RTX 3060 is still the single most common GPU on Steam at 3.99%, four years after its launch. The NVIDIA RTX 4060 holds 4.36% as of January 2026. AMD’s RX 9070 XT, the RDNA 4 flagship, appeared in the May 2026 survey at 1.33% — 25th overall.
AI/Data Center GPU Segment: Demand Outstrips Supply
The data center GPU market is a different animal entirely.
| Metric | Value | Period | Source |
|---|---|---|---|
| Global data center GPU market size | $125.02 billion | 2025 | Fortune Business Insights |
| Global data center GPU market projection | $138.88 billion | 2026 | Fortune Business Insights |
| Data center GPU market CAGR | 20.7% | 2026–2034 | Fortune Business Insights |
| Projected data center GPU market (2034) | $624.17 billion | 2034 | Fortune Business Insights |
| NVIDIA data center revenue, FY2026 | $193.74 billion | Jan 2025–Jan 2026 | NVIDIA SEC EDGAR |
| NVIDIA data center revenue, Q4 FY2026 | $62.3 billion | Q4 FY2026 | NVIDIA SEC EDGAR 8-K |
| NVIDIA data center revenue, Q1 FY2027 | $75.2 billion | May 2026 quarter | NVIDIA SEC EDGAR 8-K |
| AMD data center segment revenue, FY2025 | $16.6 billion | Full year 2025 | AMD 10-K, SEC EDGAR |
| AMD data center segment revenue, Q1 2026 | $5.8 billion | Q1 2026 | AMD 10-Q, SEC EDGAR |
NVIDIA’s Q1 FY2027 data center revenue — $75.2 billion in a single quarter — annualizes to approximately $300 billion. That number was the entire annual revenue of the global semiconductor industry in 2015.
Data center GPU shipments grew 19% in Q1 2026 per Jon Peddie Research’s Q1 2026 Market Watch, even as consumer AIBs declined. The AI hardware market is not slowing; it is accelerating. NVIDIA’s Blackwell 300 series products ramped during Q1 FY2027 as the primary revenue driver, with hyperscaler revenue (Microsoft, Google, Amazon, Meta) representing approximately 50% of data center revenue — down from earlier quarters where it exceeded 50%, as non-hyperscaler customers (sovereign AI programs, enterprise, AI cloud providers) expanded their share.
AMD’s data center trajectory is meaningful: $3.7 billion in Q1 2025 to $5.8 billion in Q1 2026, a 57% year-over-year increase driven by Instinct MI350 GPUs. Per AMD’s Q1 2026 10-Q filing, the revenue gain was “primarily driven by strong demand for our 5th generation AMD EPYC processors and AMD Instinct MI350 Series GPUs.” AMD is growing its AI GPU business — it’s just doing so from a base roughly 13× smaller than NVIDIA’s equivalent data center revenue.
Sarah Mitchell, Axis Intelligence Research AI correspondent:
The Blackwell 300 ramp in NVIDIA’s most recent quarter is worth reading carefully. Q1 FY2027 data center networking revenue — $14.8 billion, up 199% year-over-year — is the number most coverage buries in the footnotes. That’s not GPU chip revenue; it’s InfiniBand and Spectrum-X Ethernet, the fabric that connects GPU clusters into functional AI supercomputers. NVIDIA isn’t just selling chips. It’s selling the interconnect that makes the chips work together at scale, and it’s now generating $60 billion annualized in networking alone. The question I keep asking: at what point does a hyperscaler decide the networking margin is worth building its own? Amazon already has Trainium and Graviton. Google has TPUs. The custom silicon pressure on NVIDIA’s long-term share is real — it just doesn’t show up yet in the quarterly numbers because training still runs overwhelmingly on NVIDIA hardware.
GPU Market Share by Region
Consumer GPU Market — Regional Breakdown
Regional data for consumer discrete GPUs is less granular than vendor-level shipment data, as Jon Peddie Research does not publicly disclose regional AIB breakdowns by quarter. The Steam Hardware Survey provides an active-user proxy for regional installed-base patterns, with North American and European users typically representing the highest discrete GPU penetration rates.
Data Center GPU Market — Regional Revenue
| Region | Data Center GPU Revenue (2025) | Market Share | CAGR (through 2034) | Source |
|---|---|---|---|---|
| North America | $47.83 billion | 38.25% | 20.9% | Fortune Business Insights |
| Asia-Pacific | $41.46 billion | 33.2% | 22.2% | Fortune Business Insights |
| Europe | $20.92 billion | 16.7% | 30.9% | Fortune Business Insights |
| United States (subset) | $38.1 billion | ~30.5% | — | Fortune Business Insights |
| Japan (subset) | $4.86 billion | ~3.8% | — | Fortune Business Insights |
| Germany (subset) | $3.96 billion | ~3.2% | — | Fortune Business Insights |
North America’s $47.83 billion in 2025 data center GPU revenue represents the largest single-region share, driven by Microsoft, Amazon, Google, and Meta hyperscaler spending — all headquartered or primarily operating their largest clusters in US facilities. The US market alone ($38.1 billion) is larger than the entire European market.
Asia-Pacific at $41.46 billion is the second-largest region and growing fastest among the major markets (22.2% CAGR), driven by China’s domestic AI infrastructure push (Alibaba Cloud, Baidu, Tencent Cloud), South Korea’s HBM memory supply position (Samsung, SK Hynix), and Japan’s hyperscale datacenter build-out. The broader discrete GPU market places Asia-Pacific’s share even higher: Mordor Intelligence data shows the region holding 43.16% of overall GPU market revenue in 2025 — the global leader — because the region combines demand with critical supply chain positions in memory, packaging, and system manufacturing.
Europe’s 30.9% CAGR projection through 2034 is the highest of any region, reflecting both a low 2025 base and the EU’s accelerating sovereign AI and cloud infrastructure investment program.
GPU Price Statistics: Consumer Cards in 2026
Consumer discrete GPU pricing in 2026 is structured by a combination of MSRP from NVIDIA and AMD, and the street premium driven by memory shortages and tariff pass-throughs.
| GPU Model | MSRP | Q2 2026 Street Price | Architecture | Memory |
|---|---|---|---|---|
| NVIDIA GeForce RTX 5090 | $1,999 | $2,200–$2,800 | Blackwell (GB202) | 32GB GDDR7 |
| NVIDIA GeForce RTX 5080 | $999 | $1,050–$1,300 | Blackwell (GB203) | 16GB GDDR7 |
| NVIDIA GeForce RTX 5070 Ti | $749 | $980–$1,200 | Blackwell (GB203) | 16GB GDDR7 |
| AMD Radeon RX 9070 XT | $599 | $649–$750 | RDNA 4 (Navi 48) | 16GB GDDR6 |
| NVIDIA GeForce RTX 5070 | $549 | $600–$750 | Blackwell (GB205) | 12GB GDDR7 |
| NVIDIA GeForce RTX 5060 Ti | $379 | $420–$500 | Blackwell (GB206) | 16GB GDDR7 |
| NVIDIA GeForce RTX 5060 | $299 | $299–$340 | Blackwell (GB206) | 8GB GDDR7 |
Sources: NVIDIA and AMD official pricing; retail tracking May–June 2026. Street prices reflect actual retail availability and vary by region and retailer.
The RTX 5070 Ti’s $980–$1,200 street price versus its $749 MSRP illustrates the supply constraint directly. Memory shortages affecting GDDR7 supply have compressed availability at the mid-to-high tier, pushing prices above MSRP across most of the GeForce RTX 50-series line. The AMD RX 9070 XT has experienced less severe markup — $649–$750 versus $599 MSRP — partly because RDNA 4 uses GDDR6 rather than GDDR7, sidestepping the most acute memory shortage.
AI Accelerator GPU Pricing (Data Center)
The pricing dynamics of data center GPUs are a different order of magnitude.
| GPU Product | Architecture | Approximate Sell Price | Gross Margin | HBM Spec |
|---|---|---|---|---|
| NVIDIA H100 SXM5 | Hopper | ~$28,000–$30,000 at peak; $12,000–$18,000 secondhand April 2026 | ~88.1% | 80GB HBM3 |
| NVIDIA H200 SXM5 | Hopper | $35,000–$40,000 | ~85% | 141GB HBM3e |
| NVIDIA B200 SXM6 | Blackwell | $35,000–$40,000+ | ~84% | 192GB HBM3e |
| AMD Instinct MI350 | CDNA 4 | $20,000–$25,000 (est.) | ~65% | 288GB HBM3 |
| AMD Instinct MI325X | CDNA 3.5 | $15,000–$18,000 (est.) | ~65% | 256GB HBM3e |
Sources: Company SEC filings, analyst estimates. NVIDIA secondhand H100 pricing from April–May 2026 market surveys.
The H100’s secondhand price collapse — from approximately $40,000 at the 2023–2024 peak to $12,000–$18,000 by April 2026 — is the clearest illustration of how quickly GPU generations age when Blackwell supply catches up with Hopper demand. AWS cut H100 cloud pricing approximately 44% in mid-2025.
B200 availability remains the primary constraint in H1 2026. Buyers who did not secure allocations in 2024 or early 2025 face 12–18-month wait times for volume B200 orders, per market reporting. Supply-side: an estimated 3.6 million B200 units were in delivery pipeline as of April 2026.
GPU Shipment Forecast: 2026–2031
Consumer AIB Outlook
Jon Peddie Research projects:
- Desktop AIB shipments down approximately 10% in calendar 2026 versus 2025
- PC GPU (all types) installed base reaching approximately 3 billion units by 2029
- Discrete GPU penetration in PCs: 12% over the next five years (revised down sharply from prior 25% estimate)
- Overall PC GPU CAGR: -3% from 2025 to 2029
The forecast deterioration between JPR’s 2025 and 2026 projections is itself a data point: as recently as the Q3 2025 report, JPR projected a modest +1.5% CAGR. The revision to -3% came after Q4 2025 data showed the memory crisis accelerating, PC shipments collapsing harder than expected (IDC: -11.3% in 2026), and no new GPU generation launches with enough differentiation to trigger mass upgrade cycles in the consumer mid-range.
Data Center / AI GPU Market Forecast
| Year | Data Center GPU Market | Source |
|---|---|---|
| 2025 | $125.02 billion | Fortune Business Insights |
| 2026 | $138.88 billion | Fortune Business Insights |
| 2028 | ~$188 billion (est.) | Axis Intelligence Research |
| 2034 | $624.17 billion | Fortune Business Insights |
The Mordor Intelligence discrete GPU market forecast shows a different scope (including consumer and professional visualization):
- 2025: $83.31 billion (discrete GPU market)
- 2026: $96.06 billion
- 2031: $229.29 billion (CAGR 19.01% from 2026–2031)
Axis Intelligence Research note: the spread between research firm forecasts ($83B–$128B for 2025 discrete GPU market size, $138B–$145B for 2026 total GPU market size) reflects definitional differences — primarily whether AI accelerators sold as full server platforms (B200 NVL72 rack systems) are counted as GPU market revenue or as server market revenue. These are not analytical disagreements; they are scope definitions. Axis Intelligence Research presents the range rather than single-point estimates to reflect this accurately.
Sovereign and Enterprise AI GPU Demand — Emerging Demand Vector
A structural shift in AI GPU purchasing emerged in Q1 FY2027: non-hyperscaler customers (sovereign AI programs, enterprise deployments, AI cloud providers) now represent approximately 50% of NVIDIA’s data center revenue — the first time hyperscalers have not been the majority buyer. South Korea committed to building a national AI compute center anchored by 10,000+ GPUs. The EU’s InvestAI program targets €200 billion in mobilized AI investment including AI gigafactory GPU procurement. Saudi Arabia’s HUMAIN program involves a 200,000-GPU NVIDIA partnership.
This is the emerging demand vector that neither the consumer AIB forecasts nor early data center GPU forecasts captured: state-level GPU procurement programs that operate on long procurement cycles and are relatively insensitive to quarterly memory pricing fluctuations.
Axis Intelligence Research Original Analysis: GPU Market Divergence Index™ (GMDI™)
Methodology: The GPU Market Divergence Index™ (GMDI™) is an original composite metric calculated by Axis Intelligence Research. It measures the spread between (a) the year-over-year growth trajectory of discrete desktop AIB unit shipments (Jon Peddie Research quarterly data) and (b) the year-over-year growth trajectory of AI/data center GPU revenue (NVIDIA + AMD data center segment revenues from SEC EDGAR filings), indexed to a baseline of 0 at Q1 2022.
Calculation: GMDI™ = [(Data Center GPU revenue YoY growth rate) − (Consumer AIB unit shipment YoY growth rate)] × normalization factor. A positive score indicates data center revenue growth outpacing consumer unit growth; a higher score indicates greater divergence. Scores are dimensionless and comparable across periods.
| Period | Consumer AIB YoY Units | Data Center GPU YoY Revenue | GMDI™ Score |
|---|---|---|---|
| Q1 2023 | -38.2% | +14.2% | 52.4 |
| Q1 2024 | +7.3% | +427% | 419.7 |
| Q2 2025 | +27% (tariff front-run) | +154% | 127.0 |
| Q3 2025 | +47.5% (restocking) | +166% | 118.5 |
| Q4 2025 | +36% YoY | +75% | 39.0 |
| Q1 2026 | +28% YoY (AIBs) | +92% (NVIDIA Q1 FY27) | 64.0 |
The Q1 2024 reading of 419.7 represents the peak divergence — the quarter NVIDIA’s H100/H200 data center revenue first reached full ramp while consumer GPU markets were still in post-COVID correction. The current Q1 2026 reading of 64.0 represents a meaningful but more stable divergence as consumer AIB shipments have returned to growth on a year-over-year basis while data center GPU revenue continues expanding at 90%+ rates.
Limitation: GMDI™ compares revenue (data center) to unit volume (consumer AIBs), which introduces price-effect distortion. A more precise comparison would require unit volume data for data center accelerators, which NVIDIA and AMD do not disclose publicly. This limitation is noted explicitly; the metric is directionally valid but should not be used for precise quantitative modeling.
License: GMDI™ is original to Axis Intelligence Research, licensed CC BY 4.0. Citation: Axis Intelligence Research. (2026, July 7). GPU Market Divergence Index™. Axis Intelligence. https://axis-intelligence.com/gpu-shipment-statistics/
GPU Market by Application Segment
Data center accelerators are now the dominant revenue segment of the discrete GPU market by value — a reversal of the historical order.
| Application Segment | Revenue Share (Discrete GPU, 2025) | Growth Driver |
|---|---|---|
| Servers / Data Center AI | 38.42% | AI training and inference; hyperscaler capex |
| Gaming (desktop + laptop) | 37.20% | Consumer GPU refresh cycle; PC gaming market |
| Professional Visualization | ~12% | Engineering, design, medical imaging |
| Automotive / Edge | ~8% | ADAS, autonomous vehicle compute, robotics |
| Other (HPC, scientific) | ~4.38% | Government, research, simulation |
Sources: Mordor Intelligence Discrete GPU Market Report 2026; data center percentages cross-referenced with NVIDIA/AMD segment revenue filings.
The 38.42% server/data center share reflects discrete GPU revenue, not unit volume. In unit terms, consumer GPUs still represent the majority of physical cards shipped — the AIB market at 44.28 million annual units dwarfs AI accelerator unit counts by an order of magnitude. But per-unit prices of $28,000–$40,000 for H200/B200 versus $300–$2,000 for consumer AIBs mean the revenue balance has permanently flipped.
By 2026, over 60% of global GPU computational cycles (as distinct from unit counts or revenue) are estimated to be dedicated to AI workloads, per industry estimates. This reflects the always-on nature of AI cluster operations — hyperscale GPU clusters run at high utilization continuously, while consumer gaming GPUs are idle for the majority of hours in a day.
Steam Hardware Survey: GPU Installed Base Among Active PC Gamers
The Steam Hardware Survey (Valve, monthly, opt-in, approximately 130 million accounts) is the most reliable available proxy for GPU installed-base composition among active PC gamers. It is not a shipment report and should not be compared to JPR AIB shipment data, but it is the closest thing to a census of what cards are actually in use.
Top GPU Models — Steam April 2026
| Rank | GPU | Steam Share | Architecture | Launch Year |
|---|---|---|---|---|
| 1 | NVIDIA GeForce RTX 3060 | 3.99% | Ampere | 2021 |
| 2 | NVIDIA GeForce RTX 4060 | 4.36% (Jan 2026) | Ada Lovelace | 2023 |
| 3–24 | Various NVIDIA models | >1% each | Mixed | 2020–2025 |
| 25 | AMD Radeon RX 9070 XT | 1.33% | RDNA 4 | 2025 |
The most important Steam number is not which card ranks first — it’s the upgrade pace. The RTX 3060’s continued dominance at 3.99% four years after launch indicates most Steam users are not upgrading annually or even biennially. The installed base turns over slowly. NVIDIA’s GeForce RTX 50-series, launched in early 2025, has not meaningfully displaced the 30-series in the installed base through April 2026 — suggesting the premium consumer GPU market is primarily capturing first-time-discrete buyers and dedicated upgraders at the high end, not driving a broad refresh cycle.
1920×1080 (1080p) resolution held 52.21% of Steam hardware survey respondents in April 2026, with 1440p at 21.41% — confirming that the majority of active PC gamers do not need the higher-resolution performance of a flagship GPU.
Frequently Asked Questions
How many GPUs are shipped per year?
In calendar 2025, approximately 44.28 million discrete desktop add-in board (AIB) GPUs shipped globally, per Jon Peddie Research — the second-highest annual total this decade. Total PC GPU shipments (including integrated graphics across desktops, notebooks, and workstations) run roughly 70–77 million units per quarter.
What is NVIDIA’s current GPU market share?
NVIDIA holds 90% of the discrete desktop AIB market as of Q1 2026 (Jon Peddie Research). This is discrete cards only — when integrated graphics are included, Intel leads the total PC GPU market at approximately 59–61% because it ships iGPUs with the majority of PCs. NVIDIA holds approximately 80% of the AI accelerator (data center GPU) market by revenue.
How has AMD’s GPU market share changed in 2025?
AMD entered 2025 with approximately 8% of the discrete AIB market and exited 2025 at 5% — a record low — despite launching the well-reviewed Radeon RX 9000 series. It recovered slightly to 8% in Q1 2026. AMD’s data center GPU revenue grew 57% year-over-year to $5.8 billion in Q1 2026, indicating the company is pivoting its GPU growth ambitions toward AI accelerators rather than consumer graphics.
Why are GPU shipments expected to fall in 2026?
Jon Peddie Research projects desktop AIB shipments to fall approximately 10% in calendar 2026 for three reasons: (1) GDDR7 memory prices have risen sharply as HBM production for AI accelerators competes with GDDR on the same fabrication capacity; (2) US tariffs have increased landed costs for GPU hardware; (3) integrated graphics on modern CPUs has improved enough to handle casual gaming, reducing the necessity of a discrete card for mainstream buyers.
What is the difference between consumer GPU shipments and AI GPU shipments?
Consumer GPUs (gaming AIBs) are counted in unit volume — millions of cards per quarter at $300–$2,000 each. AI data center GPUs are counted in revenue — tens of thousands of units per quarter at $30,000–$40,000+ each. The former is the AIB market; the latter is the data center GPU or AI accelerator market. NVIDIA’s Q1 FY2027 data center revenue of $75.2 billion came from a much smaller unit count than its consumer GPU shipments, but at 100× the price per unit.
Which region buys the most data center GPUs?
North America led the data center GPU market in 2025 at $47.83 billion (38.25% of global revenue), driven by hyperscaler investment from Microsoft, Amazon, Google, and Meta. Asia-Pacific ranked second at $41.46 billion (33.2%), with China’s domestic AI push and South Korea’s HBM supply position as primary growth drivers. Europe held $20.92 billion (16.7%) with the highest projected CAGR of any region (30.9%) through 2034.
What GPU is most commonly used on Steam in 2026?
As of April 2026, the NVIDIA GeForce RTX 4060 held the highest single-model share (4.36% as of January 2026), while the RTX 3060 — launched in 2021 — remains among the most common cards in the active Steam installed base at approximately 3.99%. AMD’s highest-placed discrete GPU, the RX 9070 XT, appeared at 25th place with 1.33% share in May 2026.
What is NVIDIA’s Blackwell GPU and why does it matter for shipment statistics?
Blackwell is NVIDIA’s current data center and consumer GPU architecture, launched in 2025. The B200 (data center) and GeForce RTX 5000-series (consumer) are the flagship products. Blackwell drove NVIDIA’s data center revenue to $193.74 billion in fiscal 2026, and the Blackwell 300 series ramp in early 2026 is the primary driver of Q1 FY2027’s $75.2 billion in data center revenue. In the consumer market, the GeForce RTX 5000-series launch drove the 2025 AIB shipment surge — but at higher memory costs that are now feeding back as 2026 price pressure.
Methodology
Consumer AIB data: Primary source is Jon Peddie Research’s quarterly Market Watch report, the gold standard for PC GPU shipment tracking since 1984. Unit figures cited are from Jon Peddie Research press releases, which are publicly accessible via the firm’s website and widely redistributed by industry publications. JPR’s methodology tracks board shipments from NVIDIA, AMD, and Intel to their add-in board (AIB) partners — companies like ASUS, MSI, EVGA, Sapphire, and XFX — rather than consumer end sales. Shipment figures represent cards entering the distribution channel, not cards purchased by end users; the gap matters during tariff front-running periods like Q2 2025.
NVIDIA and AMD financial data: All revenue figures are sourced directly from SEC EDGAR 8-K (quarterly earnings press releases) and 10-K or 10-Q filings. No estimates or third-party summaries used for company financial data.
Market size estimates: Presented as analyst ranges (Fortune Business Insights, Mordor Intelligence) rather than single-point estimates, because definitional differences between research firms produce material variations. Axis Intelligence Research does not endorse any single forecast as authoritative.
GMDI™ metric: Calculation methodology disclosed in the GMDI™ section above. Data inputs are all primary-source (JPR quarterly unit data, NVIDIA/AMD SEC EDGAR revenue). Limitation acknowledged: revenue-to-unit comparison introduces price distortion.
Steam Hardware Survey: Opt-in survey data from Valve, published monthly. Represents active Steam users who consent to hardware telemetry — a large and reasonably representative sample of active PC gamers in English-speaking markets, not a random sample of all PC users or all GPU owners. Not a substitute for shipment data.
Pricing data: Street prices derived from US retail tracking as of May–June 2026. Regional pricing varies. MSRP figures from NVIDIA and AMD official product pages. AI accelerator pricing from analyst estimates and market reporting; NVIDIA does not publicly disclose per-unit pricing.
Limitations: Data center GPU unit shipments (as opposed to revenue) are not publicly disclosed by NVIDIA or AMD, constraining the GMDI™ precision. Jon Peddie Research’s full quarterly reports require paid subscription; figures cited herein come from publicly released press releases and Q&A sections only.
About This Dataset
This dataset aggregates GPU shipment, market share, revenue, and pricing data from Jon Peddie Research quarterly reports (2025–2026), NVIDIA and AMD SEC EDGAR filings (8-K and 10-K, fiscal years 2025–2026), Fortune Business Insights data center GPU market analysis (January 2026), Mordor Intelligence discrete GPU market analysis (January 2026), and the Valve Steam Hardware Survey (January–May 2026). The GMDI™ composite metric is an original calculation by Axis Intelligence Research.
Update approach: This article is updated when Jon Peddie Research releases new quarterly data, when NVIDIA or AMD file earnings reports that materially change the data center GPU picture, or when the rankings or traffic signal the dataset is becoming stale. There is no fixed calendar schedule.
License: Creative Commons Attribution 4.0 International (CC BY 4.0). You may share and adapt this data with attribution.
Suggested citations:
APA: Axis Intelligence Research. (2026, July 7). GPU shipment statistics 2026: Global data by segment, vendor & region. Axis Intelligence. https://axis-intelligence.com/gpu-shipment-statistics/
MLA: Axis Intelligence Research. “GPU Shipment Statistics 2026: Global Data by Segment, Vendor & Region.” Axis Intelligence, 7 July 2026, axis-intelligence.com/gpu-shipment-statistics/.
Chicago: Axis Intelligence Research. “GPU Shipment Statistics 2026: Global Data by Segment, Vendor & Region.” Axis Intelligence, July 7, 2026. https://axis-intelligence.com/gpu-shipment-statistics/.