Data Center GPUs Market Size, Share & Forecast to 2035

The Data Center GPUs Market covers graphics processing units (GPUs) sold as artificial intelligence (AI) and compute accelerator modules for servers in data centers, including the GPU package with its high-bandwidth memory (HBM), whether shipped as a PCIe card, an 8-GPU baseboard or part of a rack-scale system. DC Market Insights values the global Data Center GPUs Market at USD 153.59 billion in 2025 and forecasts USD 679.88 billion by 2035, a CAGR of 16.04%. The 2025 value is built from deployed capacity: DC Market Insights estimates that about 5.10 GW of accelerated IT capacity went into service in 2025, 78% of it on GPUs, and that each MW of GPU-equipped IT load carried about USD 38.61 million of GPU spend, which gives 3,978 MW × USD 38.61 million = USD 153.59 billion. GPUs with more than 192 gigabytes (GB) of memory are the fastest-growing segment at 37.30% a year because larger models and longer inference contexts need more memory per chip, and the Middle East and Africa is the fastest-growing region at 20.82% a year in the DC Market Insights model, led by Saudi Arabia and the United Arab Emirates growing from a small base.

Executive Summary

The Global Data Center GPUs Market size was valued at USD 4.87 billion in 2020, reached USD 153.59 billion in 2025, and is anticipated to reach USD 679.88 billion by 2035, at a CAGR of 16.04% during the forecast period.

Report attribute	Details
Historical Period	2020-2024
Base Year	2025
Forecast Period	2026-2035
Data Center GPUs Market Size 2025	USD 153.59 Billion
Data Center GPUs Market, CAGR	16.04%
Data Center GPUs Market Size 2035	USD 679.88 Billion

DC Market Insights estimates that the Data Center GPUs Market grew at 99.44% a year between 2020 and 2025, from USD 4.87 billion to USD 153.59 billion, and by 54.4% in 2025 alone. DC Market Insights forecasts 81.1% growth in 2026 to USD 278.10 billion as rack-scale systems ship in volume, then a slower path to 2035: accelerated capacity keeps rising, but the model assumes custom accelerators take a larger share of new halls and GPU spend per MW peaks in 2028 at USD 50.18 million before falling as performance per watt improves.

Market Scope: What Does the Data Center GPUs Market Cover?

DC Market Insights sizes the Data Center GPUs Market at USD 153.59 billion in 2025, counting GPU accelerators installed in data center servers. Values are taken at the GPU vendor’s first sale in current US dollars, with 2025 as the base year and forecasts to 2035.

Included. GPU silicon and its on-package HBM or other memory, sold as PCIe cards, as modules on 8-GPU baseboards, or as the GPU content of rack-scale systems such as NVIDIA Corporation (NVIDIA) Grace Blackwell and Vera Rubin systems and Advanced Micro Devices (AMD) Helios racks. The scope covers AI accelerator chips built on GPU architectures from NVIDIA, AMD and Intel Corporation (Intel) and from Chinese GPU designers such as Moore Threads, for AI training, AI inference, high-performance computing (HPC), data analytics and visualization.

Excluded. Central processing units (CPUs) sold alongside GPUs, including the NVIDIA Vera CPU, networking switches, network interface cards and optics, server chassis, power and cooling hardware, software and cloud rental revenue. Custom application-specific integrated circuits (ASICs) designed by cloud providers, and non-GPU accelerators such as Huawei Ascend chips, sit outside the market; DC Market Insights tracks them as the competing share of accelerated capacity.

How the segments fit together. Every segment split (workload, form factor, memory per GPU, cooling, end-user and data center type) re-cuts the same USD 153.59 billion, so each sums to the market total.

Market Drivers: What Drives the Data Center GPUs Market?

Accelerated capacity deployed each year rises from 5.10 GW in 2025 to 18.70 GW in 2030 in the DC Market Insights model. Three forces drive the Data Center GPUs Market: multi-gigawatt GPU partnerships announced by AI model developers, a rising GPU bill per MW as rack-scale systems replace servers, and sovereign AI programs that add national buyers.

AI Labs Line Up 16 GW Across Two GPU Partnerships

Two partnerships announced two weeks apart in 2025 set the scale of demand. In September 2025, OpenAI and NVIDIA signed a letter of intent to deploy at least 10 gigawatts of NVIDIA systems; NVIDIA intends to invest up to USD 100 billion in OpenAI progressively as each gigawatt is deployed, and the first phase is targeted for the second half of 2026 on the Vera Rubin platform. In October 2025, AMD and OpenAI announced a multi-year agreement under which OpenAI will deploy 6 gigawatts of AMD GPUs, starting with 1 gigawatt of Instinct MI450 Series GPUs in the second half of 2026, and AMD issued OpenAI a warrant for up to 160 million AMD shares that vests as milestones are met. At the DC Market Insights 2026 estimate of USD 45.83 million of GPU spend per MW, 16 GW would be worth more than four times the 2025 market. The building side of these programs is covered in the Data Center Cables Market report.

Rack-Scale Systems Raise GPU Spend per MW to USD 45.83 Million

DC Market Insights estimates GPU spend at USD 38.61 million per MW in 2025 and USD 45.83 million per MW in 2026, because a rack-scale system packs more GPU value into each MW of IT load than a row of servers.

Sovereign AI Buyers Reach USD 76.35 Billion by 2035

Governments are now direct GPU buyers, and NVIDIA’s sovereign AI revenue is covered under Competitive Insights. DC Market Insights values government, sovereign AI and research buyers at USD 15.10 billion of GPU spend in 2025 (9.8% of the market) and forecasts USD 76.35 billion by 2035, growing 17.60% a year.

Market Trends: How Are Data Center GPUs Changing?

Liquid-cooled GPUs reach 81.2% of value by 2035, DC Market Insights forecasts. The Data Center GPUs Market is moving to rack-scale systems, liquid cooling and much larger memory per GPU.

Liquid-Cooled GPUs Rise from 38.5% to 81.2% of Value

DC Market Insights assumes that most rack-scale systems are liquid-cooled, so liquid-cooled GPUs grow at 25.02% a year in the model, from USD 59.19 billion in 2025 to USD 552.20 billion in 2035, while air-cooled GPUs grow only 3.07% a year. Air cooling keeps a place in enterprise and inference servers: Intel’s Crescent Island inference GPU, announced in October 2025 with 160GB of LPDDR5X memory, targets air-cooled enterprise servers. The cooling hardware behind this shift is covered in the Data Center Liquid Cooling Manifolds Market report.

Memory per GPU Passes 192 GB as HBM4 Arrives

DC Market Insights values GPUs with more than 192 GB of memory at USD 21.23 billion in 2025 (13.8%) and forecasts USD 505.22 billion by 2035 (74.3%), growth of 37.30% a year. The HBM4 standard published in April 2025 by the JEDEC Solid State Technology Association (JEDEC) raises total bandwidth to up to 2 TB/s over a 2048-bit interface and allows up to 64GB per stack, so a GPU with several HBM4 stacks can clear 192 GB.

Inference Overtakes Training as the Largest Workload

DC Market Insights estimates that AI training accounted for USD 85.75 billion (55.8%) of 2025 GPU spend and AI inference for USD 56.69 billion (36.9%). The model forecasts inference to reach USD 376.25 billion by 2035 (55.3%), growing at 20.84% a year, against 11.77% for training, because every deployed model and agent generates inference demand for as long as it is in service, while training runs end.

Market Challenges: What Holds the Data Center GPUs Market Back?

Growth slows from 81.1% in 2026 to 16.04% a year over the decade, DC Market Insights forecasts. Export controls, competition from custom accelerators and the power available to new data centers are the three constraints on the Data Center GPUs Market.

Export Controls Cut Access to China, About 7.2% of Value

DC Market Insights values China at USD 11.09 billion in 2025 (7.2% of the market) and models it growing at 16.84% a year, close to the global rate, on the assumption that domestic GPU designers take most of the demand US vendors cannot serve.

Custom Accelerators Take a Rising Share of AI Capacity

In July 2026, Alphabet raised its 2026 capital expenditure guidance to USD 195 billion to USD 205 billion, from USD 180 billion to USD 190 billion. DC Market Insights assumes that a growing part of such cloud budgets goes to custom accelerators rather than GPUs: the model has GPUs equipping 78% of accelerated capacity deployed in 2025, 65.5% in 2030 and 61.7% in 2035. This share loss is the main reason the Data Center GPUs Market grows at 16.04% a year while accelerated capacity deployed grows at 17.60% a year.

Power Limits the Pace of Deployment

DC Market Insights estimates that 1 GW of accelerated capacity carried USD 30.12 billion of GPU spend in 2025 across GPU and non-GPU halls, so each GW of grid connection delayed by a year defers about that amount of GPU demand.

Market Opportunities: Where Are the New Revenue Pools?

Enterprise GPU spend grows at 19.00% a year, the fastest of any buyer group, DC Market Insights forecasts. Enterprises and colocation data centers, where GPU spend grows at 19.29% a year, are the largest new revenue pools in the Data Center GPUs Market.

Enterprise GPU Spend Reaches USD 107.49 Billion by 2035

DC Market Insights estimates that enterprises spent USD 18.88 billion on data center GPUs in 2025 and forecasts USD 107.49 billion by 2035. In the model, enterprise buyers put 52.4% of their 2025 GPU spend into inference, the highest share of any buyer. For an enterprise GPU cluster of 1 MW, GPU spend is about USD 38.61 million at 2025 rates.

Colocation Data Centers Grow to 36.1% of GPU Value

DC Market Insights estimates that colocation data centers hosted USD 42.02 billion of GPUs in 2025 (27.4%) and forecasts USD 245.23 billion by 2035 (36.1%), on the assumption that most GPU cloud and neocloud capacity sits in leased halls. The model has GPU cloud and neocloud providers growing at 18.64% a year, from USD 24.08 billion to USD 133.05 billion. In August 2026, NVIDIA listed CoreWeave and Nebius alongside Google Cloud, Microsoft Azure and Oracle Cloud Infrastructure as partners running Vera Rubin racks.

Pricing: What Is the Cost of Upgrading Data Center GPU Clusters?

Upgrading the GPUs in a 100 MW hall cost about USD 3.86 billion at 2025 prices, DC Market Insights estimates. At 2026 prices the bill is about USD 4.58 billion, based on GPU spend per MW of IT load. These figures cover the GPUs and their memory only, not CPUs, networking, servers, power or cooling.

Data Center GPU spend	2020	2025	2030	2035
GPU spend per MW (USD million)	9.62	38.61	48.57	42.71
Accelerated capacity deployed (GW)	0.55	5.10	18.70	25.80
GPU share of accelerated capacity	92.0%	78.0%	65.5%	61.7%

In the DC Market Insights model, GPU spend per MW quadrupled between 2020 and 2025 as each generation packed more memory and compute into each MW. DC Market Insights expects spend per MW to peak at USD 50.18 million in 2028, then fall about 2.3% a year to 2035 as competition from AMD and custom accelerators lowers price per unit of performance. A premium enterprise workload, such as model training on rack-scale systems, carries the highest spend per MW, while a reliable, cost-optimized data center GPU setup for enterprise inference, chosen for cost efficiency, uses PCIe accelerator cards or GPUs with less memory: DC Market Insights estimates that GPUs with up to 96 GB of memory accounted for USD 34.26 billion of 2025 spend.

Standards and Regulation: Which Rules Shape the Data Center GPUs Market?

China, worth 7.2% of 2025 GPU value in the DC Market Insights model, is where rules weigh most on the Data Center GPUs Market. Three sets of rules shape where and how data center GPUs are sold: US export controls, memory standards from JEDEC and energy reporting rules for data centers.

US export controls. In May 2025, the Bureau of Industry and Security (BIS) of the US Department of Commerce announced the rescission of the AI Diffusion Rule, which had been issued on January 15, 2025 with compliance due on May 15, 2025, and issued guidance on the risks of using PRC advanced computing chips, including specific Huawei Ascend chips (Source: BIS press release). In April 2025, the US government told NVIDIA that exports of its H20 GPU to China (including Hong Kong and Macau) and D:5 countries require a license, and NVIDIA said its first-quarter fiscal 2026 results would include up to about USD 5.5 billion of charges for H20 inventory, purchase commitments and related reserves (Source: NVIDIA Form 8-K). NVIDIA’s outlook for the third quarter of fiscal 2027 assumes no Data Center compute revenue from China. In the DC Market Insights view, licensing for China decides how much of the Chinese market US vendors can serve.

JEDEC JESD270-4 (HBM4). The JEDEC HBM4 standard, released in April 2025, doubles the independent channels per stack from 16 to 32, supports 4-high to 16-high stacks and up to 64GB per stack, and lifts total bandwidth to up to 2 TB/s. Intel, in contrast, announced Crescent Island at the 2025 OCP Global Summit with LPDDR5X memory rather than HBM.

EU data center reporting. Commission Delegated Regulation (EU) 2024/1364, in force since 6 June 2024, requires operators of data centers with IT power demand of at least 500 kW to report 24 indicators on energy, IT and data traffic to a European database, via national reporting systems where they exist, first by 15 September 2024, then by 15 May 2025 and annually after that.

Market Segmentation: Which Segment Leads the Data Center GPUs Market?

AI training leads the Data Center GPUs Market by workload with 55.8% of 2025 value, DC Market Insights estimates. 8-GPU baseboards lead by form factor with 47.6%, and hyperscale cloud providers lead buyers with 56.4%. GPUs with more than 192 GB of memory are the fastest-growing segment at 37.30% a year, ahead of liquid-cooled GPUs at 25.02%, because HBM4 and larger models push memory per GPU up with each generation.

By Workload: AI Training Leads with 55.8%

Workload	2025 (USD billion)	Share 2025	2035 (USD billion)	CAGR 2025-2035
AI Training	85.75	55.8%	260.87	11.77%
AI Inference	56.69	36.9%	376.25	20.84%
HPC and Scientific Computing	7.02	4.6%	26.38	14.16%
Data Analytics and Visualization	4.13	2.7%	16.38	14.77%

Inference overtakes training during the forecast and holds 55.3% of value by 2035.

By Form Factor: 8-GPU Baseboards Lead with 47.6%

Form factor	2025 (USD billion)	Share 2025	2035 (USD billion)	CAGR 2025-2035
Rack-Scale Systems	63.34	41.2%	431.52	21.15%
8-GPU Baseboards	73.08	47.6%	189.07	9.97%
PCIe Cards	17.17	11.2%	59.29	13.19%

Rack-scale systems overtake 8-GPU baseboards during the forecast and reach 63.5% of value by 2035, growing at 21.15% a year.

By Memory per GPU: 97 GB to 192 GB Leads with 63.9%

GPUs with 97 GB to 192 GB are worth USD 98.10 billion in 2025 and grow 4.05% a year, while GPUs with up to 96 GB decline 1.74% a year to USD 28.76 billion.

By Cooling: Air-Cooled GPUs Lead with 61.5% in 2025

Air-cooled GPUs are worth USD 94.40 billion in 2025; by 2035 the order reverses and air-cooled GPUs fall to 18.8% of value.

By End-user: Hyperscale Cloud Providers Lead with 56.4%

End-user	2025 (USD billion)	Share 2025	2035 (USD billion)	CAGR 2025-2035
Hyperscale Cloud Providers	86.64	56.4%	325.80	14.16%
GPU Cloud and Neocloud Providers	24.08	15.7%	133.05	18.64%
Enterprises	18.88	12.3%	107.49	19.00%
Government, Sovereign AI and Research	15.10	9.8%	76.35	17.60%
Telecom and Others	8.89	5.8%	37.19	15.38%

Hyperscale cloud providers lose 8.5 points of share by 2035 in the DC Market Insights model, which assumes they move more of their own workloads to custom accelerators, while enterprises grow fastest among buyers.

By Data Center Type: Hyperscale Data Centers Lead with 61.7%

Hyperscale data centers house USD 94.81 billion of GPUs in 2025 and USD 357.69 billion by 2035, growing 14.20% a year. Edge data centers, worth USD 2.04 billion in 2025, grow fastest at 24.55% a year as inference moves closer to users.

DC Exclusive: The GPU Spend per MW Model and Buyer × Workload Matrix

Hyperscale AI training is the largest cell of the matrix at USD 51.90 billion, 33.8% of 2025 GPU spending. The matrix is a DC Market Insights model built from 92 inputs: 48 cohort inputs (accelerated capacity, GPU share and GPU spend per MW for each year from 2020 to 2035), 24 segment shares, 5 regional splits, 12 country weights and 3 vendor disclosures used as a cross-check.

2025, USD billion	AI Training	AI Inference	HPC	Analytics and Visualization	Total
Hyperscale Cloud Providers	51.90	32.60	0.61	1.53	86.64
GPU Cloud and Neocloud Providers	16.25	6.97	0.47	0.39	24.08
Enterprises	6.78	9.89	0.75	1.46	18.88
Government, Sovereign AI and Research	7.10	2.73	5.04	0.23	15.10
Telecom and Others	3.72	4.50	0.15	0.52	8.89
Total	85.75	56.69	7.02	4.13	153.59

GPU cloud and neocloud providers put 67.5% of their GPU spend into training capacity, the highest share of any buyer, reflecting the model’s assumption that AI labs rent much of their training capacity. Government, sovereign AI and research buyers account for USD 5.04 billion, or 71.8%, of all HPC GPU spend in the model, because national research computing is concentrated with this buyer group.

The GPU spend per MW model behind these figures has three inputs a reader can test against public announcements: accelerated capacity deployed each year (5.10 GW in 2025), the share of it built on GPUs (78% in 2025) and GPU spend per MW (USD 38.61 million in 2025). The model and its assumptions are explained on our research methodology page.

Regional Insights: Which Region Leads the Data Center GPUs Market?

North America leads the Data Center GPUs Market with 63.4% of 2025 value, DC Market Insights estimates. The Middle East and Africa is the fastest-growing region at 20.82% a year because Saudi Arabia and the United Arab Emirates grow at 21.27% and 20.21% a year from a small base.

North America Leads with 63.4% Share

DC Market Insights estimates that North America installed USD 97.35 billion of data center GPUs in 2025 and forecasts USD 359.05 billion by 2035, a CAGR of 13.94%. The model places 94.2% of the region’s 2025 GPU value in the United States and 5.8% in Canada.

Asia Pacific Holds 18.9% Share

DC Market Insights values Asia Pacific at USD 29.09 billion in 2025 and USD 163.58 billion by 2035, growing 18.85% a year. In the model, China is the largest country market in the region, followed by Japan and South Korea, while India and Malaysia grow fastest at 24.86% and 24.02% a year.

Europe Holds 9.6% Share

DC Market Insights values Europe at USD 14.70 billion in 2025 and USD 76.76 billion by 2035, a CAGR of 17.97%. The United Kingdom, Germany and France are the largest European markets in the model, and France grows fastest of the three at 19.06% a year.

Middle East and Africa Holds 6.3% Share

The Middle East and Africa is worth USD 9.66 billion in 2025 and USD 64.04 billion by 2035 in the DC Market Insights model, the fastest growth of any region at 20.82% a year. Saudi Arabia and the United Arab Emirates together account for 81.2% of the region’s 2025 value.

Latin America Holds 1.8% Share

Latin America is worth USD 2.79 billion in 2025 and USD 16.45 billion by 2035 in the DC Market Insights model, growing 19.39% a year, the second-fastest regional rate after the Middle East and Africa.

Country Analysis: Which Countries Buy the Most Data Center GPUs?

The United States leads with USD 91.68 billion, 59.7% of 2025 value, DC Market Insights estimates. The United States is followed by China at USD 11.09 billion (7.2%) and Japan at USD 5.77 billion (3.8%). The 12 largest countries account for 90.2% of 2025 spending.

Rank	Country	Region	2025 (USD billion)	Share 2025	2035 (USD billion)	CAGR 2025-2035
1	United States	North America	91.68	59.7%	333.81	13.79%
2	China	Asia Pacific	11.09	7.2%	52.59	16.84%
3	Japan	Asia Pacific	5.77	3.8%	28.40	17.27%
4	Canada	North America	5.67	3.7%	25.24	16.11%
5	Saudi Arabia	Middle East and Africa	4.07	2.7%	28.03	21.27%
6	United Kingdom	Europe	4.06	2.6%	19.33	16.88%
7	United Arab Emirates	Middle East and Africa	3.77	2.5%	23.79	20.21%
8	Germany	Europe	3.06	2.0%	14.87	17.13%
9	South Korea	Asia Pacific	3.05	2.0%	16.05	18.08%
10	India	Asia Pacific	2.60	1.7%	23.92	24.86%
11	France	Europe	2.10	1.4%	12.01	19.06%
12	Malaysia	Asia Pacific	1.66	1.1%	14.30	24.02%

India is the fastest-growing large country market at 24.86% a year, ahead of Malaysia at 24.02%, and Saudi Arabia grows fastest among the top five at 21.27%. The country figures are a DC Market Insights allocation of the regional totals, weighted by each country’s estimated share of installed and planned accelerated capacity.

Forecast Scenarios: How Could the 2035 Forecast Change?

The base case reaches USD 679.88 billion by 2035, DC Market Insights forecasts. Scenarios give the Data Center GPUs Market a range of USD 471.30 billion to USD 905.40 billion across the low and high scenarios.

Scenario	Assumption from 2027	2035 (USD billion)	CAGR 2025-2035
Low	Accelerated capacity 15% below base; GPU share of capacity falls to 50.3% by 2035	471.30	11.86%
Base	Capacity deployed rises to 25.80 GW a year; GPU share 61.7% in 2035	679.88	16.04%
High	Capacity 20% above base; GPU spend per MW 11% above base	905.40	19.41%

The high case assumes power connections keep pace with the announced multi-gigawatt partnerships and that GPU vendors hold pricing as memory per GPU rises. The low case assumes GPU spend per MW stays at base levels while custom accelerators take a larger share of new halls.

Competitive Insights: Who Leads the Data Center GPUs Market?

• NVIDIA
• AMD
• Intel
• Moore Threads
• SK hynix
• Samsung Electronics
• Micron Technology
• CoreWeave
• Nebius
• Oracle Cloud Infrastructure
• Microsoft Azure
• Google Cloud

NVIDIA leads the Data Center GPUs Market with about 93.4% of 2025 value, and NVIDIA and AMD together hold about 97.6% (DC Market Insights estimate). NVIDIA reported fiscal 2026 Data Center revenue of USD 193.7 billion (Source: NVIDIA results filed with the US Securities and Exchange Commission) and said its networking business exceeded USD 31 billion for the year. On its February 2026 earnings call, NVIDIA said its sovereign AI business more than tripled in fiscal 2026 to over USD 30 billion and that Grace Blackwell systems made up roughly two-thirds of fourth-quarter Data Center revenue. In August 2026, NVIDIA CFO Colette Kress said NVIDIA’s revenue from each gigawatt of data center capacity had risen from about USD 30 billion five years ago to USD 60 billion, according to Data Center Dynamics; that figure covers all NVIDIA revenue per gigawatt, including networking and CPUs. DC Market Insights maps fiscal 2026 to calendar 2025, deducts networking to leave at most about USD 162.70 billion of compute, and attributes 88.2% of compute to GPUs, giving about USD 143.50 billion. AMD’s Data Center segment, which includes EPYC CPUs as well as Instinct GPUs, reached USD 16.6 billion in 2025, after AMD CEO Lisa Su said Instinct revenue passed USD 5 billion in 2024; DC Market Insights estimates 2025 Instinct GPU revenue at USD 6.42 billion (4.2%). DC Market Insights estimates that Intel and Chinese GPU designers share the remaining USD 3.67 billion (2.4%); Intel’s Crescent Island inference GPU is due to sample in the second half of 2026, and Moore Threads raised CNY 8 billion (USD 1.1 billion) in its December 2025 initial public offering on Shanghai’s STAR Market. SK hynix, Samsung and Micron were among the companies named in JEDEC’s HBM4 announcement, and CoreWeave, Nebius, Oracle Cloud Infrastructure, Microsoft Azure and Google Cloud were named by NVIDIA in August 2026 as partners running Vera Rubin racks.

Recent Developments

• In September 2025, OpenAI and NVIDIA signed a letter of intent to deploy at least 10 gigawatts of NVIDIA systems, and NVIDIA said it intends to invest up to USD 100 billion in OpenAI as each gigawatt is deployed (Source: NVIDIA press release).
• In October 2025, AMD and OpenAI announced that OpenAI will deploy 6 gigawatts of AMD GPUs, starting with 1 gigawatt of Instinct MI450 Series GPUs in the second half of 2026 (Source: AMD press release).
• In October 2025, Intel announced Crescent Island, an inference data center GPU with 160GB of LPDDR5X memory, with customer sampling expected in the second half of 2026 (Source: Intel newsroom article).
• In February 2026, NVIDIA reported record fiscal 2026 Data Center revenue of USD 193.7 billion, up 68% (Source: NVIDIA Form 8-K filing).
• In August 2026, AMD reported second-quarter Data Center segment revenue of USD 6.7 billion, up 107%, and announced a partnership with Anthropic to deploy up to 2 gigawatts of MI450 Series GPUs in Helios racks (Source: AMD earnings press release).
• In August 2026, NVIDIA reported second-quarter fiscal 2027 Data Center revenue of USD 89.0 billion, up 117%, and said Vera Rubin was in full production (Source: NVIDIA earnings press release).

We follow these partnerships, launches and earnings as they happen in our data center industry updates.

Methodology: How We Built the Data Center GPUs Market Model

DC Market Insights built the model from 16 annual cohorts, 2020 to 2035. The Data Center GPUs Market is built bottom-up, with each cohort defined by accelerated IT capacity deployed (new halls and GPU replacement), the share of that capacity equipped with GPUs, and GPU spend per MW. The 2025 cohort is 5.10 GW × 78% × USD 38.61 million per MW = USD 153.59 billion. The total was split by 4 workloads, 3 form factors, 3 memory classes, 2 cooling types, 5 end-users, 4 data center types and 5 regions, and every split sums to the market total. The result was checked top-down against 3 vendor disclosures: NVIDIA’s Data Center compute revenue for fiscal 2025 (USD 102.2 billion), NVIDIA’s fiscal 2026 Data Center revenue less its disclosed networking revenue, and AMD’s Data Center segment revenue for 2025. NVIDIA’s fiscal 2021 Data Center revenue of USD 6.70 billion anchors the 2020 value. The analyst named on this page built the model, and the reviewer checked the arithmetic, the sources and the dates before publication.