Facts & Figures
Visual Intelligence
Market Leader
Nvidia makes the graphics processing units (GPUs) that train and run most large artificial intelligence models, and that role produced $215.9 billion in revenue in fiscal 2026, up 65% from $130.5 billion the year before. Data Center, the segment that sells to AI data centers, brought in 89.7% of it. Jensen Huang, who co-founded Nvidia in 1993 and still serves as its CEO, built a company that was once best known for gaming cards; gaming made $16.0 billion, a fraction of what AI infrastructure now earns. The accelerators behind the shift run from the Hopper-generation H100 and H200 through Blackwell to Vera Rubin, which Nvidia said in August 2026 is in full production. Nvidia also sells the CUDA software platform that most AI code is written against, which makes switching to a rival GPU a much bigger decision than swapping one chip for another. Every company in the tiers below is chasing a piece of this buildout: TSMC manufactures Nvidia's chips, SK hynix supplies the high-bandwidth memory (HBM) stacked beside them, AMD competes with GPUs of its own, and Broadcom builds custom silicon for the biggest cloud companies.
Nvidia's fiscal 2026 revenue is built on AI data center chips and networking, not the gaming cards that first made the company famous: Data Center alone brought in $193.7 billion.
215.9B
Nvidia reported $215.9 billion in revenue for fiscal 2026, the year ended 25 January 2026, up from $130.5 billion a year earlier. That is about 30% of the $717.4 billion the ten chipmakers in this ranking earned combined, from a company that sells GPUs, networking and software rather than commodity components.
193.7B
Data Center revenue reached $193.7 billion in fiscal 2026, or 89.7% of the company total. Compute contributed $162.4 billion, while networking, the NVLink, Ethernet and InfiniBand products that link chips into AI clusters, added $31.4 billion, up 142% from the prior year. Gaming added $16.0 billion, and Automotive $2.3 billion.
65%
Revenue grew 65%, from $130.5 billion in fiscal 2025 to $215.9 billion in fiscal 2026, an increase of about $85.4 billion in a single year. That one-year gain is larger than the entire chip revenue of SK hynix ($68.3 billion), Intel ($52.9 billion) or Qualcomm ($38.4 billion), each of which ranks in this top 10.
Above $50 Billion
TSMC's $122.4 billion in 2025 revenue, up 35.9%, comes from a company that designs no chips under its own name. It is a pure-play foundry that manufactures other companies' products: in 2025 it made 12,682 different products for 534 customers. Its CEO has named Apple, Nvidia, AMD, Qualcomm and Broadcom among the US customers whose AI demand is driving its capacity expansion.
That model explains why TSMC ranks second on a list full of rivals. It never competes with its customers, so the companies fighting each other for AI business all depend on it, and Nvidia's annual report names TSMC and Samsung as the foundries that produce its wafers. TSMC's own report credits, most of all, the strong momentum of AI-equipped servers and data centers for growth in its high performance computing platform.
Samsung's chip business, its Device Solutions (DS) division, earned about $91.5 billion in 2025: KRW 130.1 trillion at Samsung's average rate of 1,421.89 won per dollar, up 17% and roughly 39% of the company's KRW 333.6 trillion total. It is built on DRAM and NAND flash memory, alongside a foundry business and a System LSI unit that designs Exynos mobile processors.
Memory is also where Samsung meets SK hynix head-on. Samsung reported expanding HBM sales in 2025 and plans to deliver HBM4 mass products, yet Counterpoint's Q2 2026 data puts its share of the HBM market at 33% against SK hynix's 50%. Samsung is the clear second supplier, and the gap has been narrowing.
SK hynix reported KRW 97.1 trillion in 2025 revenue, up 47% in won terms and about $68.3 billion at the same 1,421.89 won per dollar used for Samsung. The figure comes from its preliminary results of 28 January 2026. SK hynix is the largest supplier of High Bandwidth Memory (HBM), the specialized memory stacked beside AI GPUs to feed them data fast enough for training and inference, and its HBM revenue more than doubled in 2025.
That makes SK hynix one of Nvidia's most important suppliers rather than only a rival in the wider memory market: Nvidia's annual report lists SK hynix, Micron and Samsung as its memory suppliers. SK hynix says it completed preparations for HBM4 mass production in September 2025, the first in the industry, as it works to defend a lead that Samsung is closing.
Intel's $52.9 billion in revenue for the year ended 27 December 2025 was flat on 2024. Most of it came from Intel Products, $49.1 billion, led by Client Computing, the PC chips, at $32.2 billion and Data Center and AI at $16.9 billion, up 5%. Next to Nvidia's $193.7 billion Data Center segment, Intel's whole data center and AI business is under a tenth of the size.
Intel's bigger bet is Intel Foundry, its push to manufacture chips for other companies the way TSMC does. The segment reported $17.8 billion in 2025, up 3%, before intersegment eliminations, and Intel says its 18A process ramped to high-volume manufacturing in Arizona and Oregon. CEO Lip-Bu Tan says Intel's conviction in the essential role of CPUs in the AI era continues to grow.
Below $50 Billion
Qualcomm's chip segment, QCT, earned $38.4 billion in the fiscal year ended 28 September 2025, nearly all of it from chips that never carry the Qualcomm name to consumers. Handsets contributed $27.8 billion, IoT $6.6 billion and automotive $4.0 billion. Counting its licensing business, Qualcomm's total revenue was $44.3 billion, but licensing is left out of this ranking.
The push beyond phones shows in the numbers: automotive revenue grew 36% to $4.0 billion, and Qualcomm's CEO pointed to the availability of its automated driving stack. Even so, handsets still supply roughly 72% of QCT revenue, so Qualcomm remains a phone chip company first.
Micron Technology earned $37.4 billion in the fiscal year ended 28 August 2025, which puts it seventh in this ranking. Its next fiscal year, ended 3 September 2026, tells the AI memory story: revenue reached $133.2 billion, up 256%. Set against the 2025 figures of its rivals, that would rank Micron third, behind only Nvidia and TSMC.
The surge is concentrated in data center demand. Micron's Core Data Center unit made $18.0 billion in the final quarter of fiscal 2026, up from $1.6 billion a year earlier, and its Cloud Memory unit made $16.3 billion, up from $4.5 billion. Memory revenue is notoriously cyclical, and a one-year jump of this size shows how hard it swings with demand.
Broadcom's semiconductor solutions segment earned $36.9 billion in the fiscal year ended 2 November 2025. The company's total was $63.9 billion, but $27.0 billion of that is infrastructure software, including VMware, which this ranking leaves out. The AI side of the chip business is custom accelerators and Ethernet AI switches, the networking that moves data inside AI data centers.
Custom accelerators are chips designed for one customer rather than sold as off-the-shelf GPUs. In April 2026 Broadcom disclosed a long-term agreement to develop and supply custom Tensor Processing Units for future generations of Google's TPUs, and Meta describes its MTIA chips as developed in close partnership with Broadcom. In its fiscal third quarter of 2026, Broadcom's AI semiconductor revenue reached $16.7 billion, up 221% from a year earlier.
AMD's $34.6 billion in 2025 revenue spans PC and gaming chips, which brought in $14.6 billion, and a Data Center segment of $16.6 billion built on Epyc server processors and Instinct AI accelerators, plus $3.5 billion from embedded chips. AMD designs its chips and leaves manufacturing to others: in March 2009 it contributed its Dresden fabs and other manufacturing assets to a joint venture that became GlobalFoundries.
In AI accelerators, AMD launched the Instinct MI350X and MI355X in June 2025. In July 2026 it launched the MI400 series, including the MI455X and MI430X, and the Helios rack, which pairs 72 MI455X GPUs with 18 Epyc Venice CPUs; AMD calls Helios in production, and OpenAI expects to bring it online from the fourth quarter of 2026. Even so, AMD's whole Data Center segment is under a tenth of Nvidia's, and CUDA still gives Nvidia a software head start.
MediaTek earned NT$596.0 billion in 2025, about $19.1 billion at its average 2025 exchange rate of 31.19 NT dollars per US dollar, up 12.3% in Taiwan dollars. The Taiwanese chip designer is best known for its Dimensity smartphone processors, whose flagship Dimensity 9500 ramped strongly late in the year, and it also sells chips for smart devices.
Texas Instruments, which reported $17.7 billion in 2025 revenue, of which $14.0 billion was analog chips, now sits just outside the top 10 at number 11.
Data Table
Every chipmaker from Nvidia's $215.9B down to MediaTek's $19.1B, in one list: company, fiscal year, chip revenue, and the main drivers behind each figure.
| 1 | Nvidia | FY2026 (to 25 Jan 2026) | $215.9 billion | Data center GPUs and AI systems |
| 2 | TSMC | FY2025 (to 31 Dec 2025) | $122.4 billion | Contract chip manufacturing (foundry) |
| 3 | Samsung Electronics (chip division) | FY2025 (to 31 Dec 2025) | $91.5 billion | DRAM and NAND memory, HBM, foundry, Exynos |
| 4 | SK hynix | FY2025 (to 31 Dec 2025) | $68.3 billion | HBM, DRAM and NAND memory |
| 5 | Intel | FY2025 (to 27 Dec 2025) | $52.9 billion | PC and server CPUs, Intel Foundry |
| 6 | Qualcomm (chip segment, QCT) | FY2025 (to 28 Sep 2025) | $38.4 billion | Snapdragon processors, modems, automotive and IoT chips |
| 7 | Micron Technology | FY2025 (to 28 Aug 2025) | $37.4 billion | DRAM, NAND and HBM memory |
| 8 | Broadcom (semiconductor segment) | FY2025 (to 2 Nov 2025) | $36.9 billion | Custom AI accelerators (XPUs), networking chips |
| 9 | AMD | FY2025 (to 27 Dec 2025) | $34.6 billion | EPYC and Ryzen CPUs, Instinct AI GPUs |
| 10 | MediaTek | FY2025 (to 31 Dec 2025) | $19.1 billion | Smartphone and smart-device processors |
Basis: each company's fiscal 2025 chip revenue, with Nvidia's comparable year being fiscal 2026 (27 January 2025 to 25 January 2026). Where a company reports a chip segment, that segment is used: Samsung's Device Solutions division, Qualcomm's QCT and Broadcom's semiconductor solutions. Samsung, SK hynix and MediaTek are converted from won and NT dollars at the companies' own 2025 average rates, so their dollar figures are approximate.
AI Chip Leadership
Nvidia's Hopper-generation H100 and H200, then Blackwell, are the default hardware in most large AI training clusters, and Vera Rubin is now in full production. That lead is not only chip performance: CUDA, which Nvidia says over 7.5 million developers use, makes its hardware the best-supported choice, so rivals with competitive chips still struggle for share. The biggest buyers also build custom silicon, from Google's TPUs and Amazon's Trainium3 to Microsoft's Maia 200 and Meta's MTIA, yet Google Cloud still offers Vera Rubin, and Meta has agreed to deploy millions of Nvidia GPUs.
Nvidia vs Rivals
Nvidia's GPUs remain the safe default for AI work, purely because of CUDA's software support: whether you are training models, running local LLMs or doing heavy machine-learning work, almost every AI framework and tutorial assumes an Nvidia GPU first. AMD is the legitimate alternative, from Radeon cards to the Instinct MI350 and MI400 series, if you are comfortable working around occasional software gaps.
Apple Silicon's unified memory makes it a strong pick for running smaller models locally on a laptop, while Intel and Qualcomm remain better bets for general-purpose computing than for AI-specific acceleration, at least for now. The largest custom chips are not sold as cards at all: Google's TPUs and Amazon's Trainium are rented through their clouds, so most teams end up choosing between Nvidia, AMD and a cloud provider's own silicon.
Quick Answers
Three questions the sections above do not fully cover.
Neural-network layers are computed largely as matrix multiplications, which, as NVIDIA's performance documentation puts it, GPUs handle by executing many threads concurrently, so the chip sets how large and how cheap a model can be. Each generation raises that ceiling: Nvidia says Blackwell Ultra delivers up to 50 times better performance than Hopper on agentic AI, citing SemiAnalysis benchmarks. Memory is the second limit, which is why SK hynix reported that its HBM revenue more than doubled in 2025 and Samsung is preparing HBM4 for mass delivery. The spending shows the dependence: Alphabet's capital expenditure was $44.9 billion in the second quarter of 2026, double the $22.4 billion of a year earlier.
Nvidia GPUs are built to run many operations at once: NVIDIA describes a GPU as hundreds of cores handling thousands of threads, against the few cores of a CPU. Tensor Cores add dedicated matrix-math hardware that works in mixed precision, down to the 4-bit NVFP4 format, trading numerical precision for throughput. High-bandwidth memory sits beside the GPU to keep it fed, with 13.4 TB of HBM3E in a Blackwell GB200 NVL72 rack, and NVLink joins the rack's 72 GPUs so they can work as a single accelerator. The CUDA software platform, used by over 7.5 million developers, ties the stack together.
Cloud providers and AI labs come first: hyperscalers were slightly more than half of Nvidia's Data Center revenue in the fourth quarter of fiscal 2026, and Data Center as a whole brought in $193.7 billion for the year. Their suppliers benefit next, including memory makers such as Micron, whose Cloud Memory unit made $16.3 billion in its latest quarter, up from $4.5 billion a year earlier, and TSMC, which manufactures the chips. Beyond the data center, Nvidia's own filings point to healthcare for drug discovery, financial services for fraud detection and customer support, and automotive for autonomous driving, though Automotive revenue was only $2.3 billion in fiscal 2026 and Gaming $16.0 billion.
Nvidia: Q4 Fiscal 2026 Results Nvidia: Q2 Fiscal 2027 Results Nvidia: Fiscal 2026 Form 10-K TSMC: 4Q25 Management Report Samsung Electronics: 4Q 2025 Earnings Presentation SK hynix: FY25 Results Counterpoint Research: Global HBM Market Share, Q2 2026 Intel: Q4 and Full-Year 2025 Results Qualcomm: Fiscal 2025 Results Micron: Fiscal 2026 Q4 Results Broadcom: Fiscal 2025 Results Broadcom: Form 8-K, 6 April 2026 AMD: Fourth Quarter and Full-Year 2025 Results MediaTek: 2025 Results
The revenue figures in this ranking come from each company's own filings and results releases: Nvidia, TSMC, Samsung Electronics, SK hynix, Intel, Qualcomm, Micron, Broadcom, AMD and MediaTek. Each is fiscal 2025 chip revenue (Nvidia's comparable year is fiscal 2026), using the chip segment where one is reported, with Samsung, SK hynix and MediaTek converted at the companies' own 2025 average rates. Product details (Tensor Cores, NVLink, Google TPU, Amazon Trainium3, Microsoft Maia 200, Meta MTIA, AMD Instinct) come from each company's own announcements, and Texas Instruments' figure from its 2025 results.
FactsFigs reviews, cleans, and cross-checks every source dataset before shaping it into a data story. Each visualization is created and designed in FactsFigs Design Studio - an internal tool developed and owned by FactsFigs - and is the original work of a FactsFigs author, not an AI-generated copy of any existing graphic. Individual assets within a visual may or may not be produced with AI tools, but the design of the visual itself is solely FactsFigs' own.
Figures are as reported by each company, currency conversions are approximate, and nothing here is financial advice.
Last verified: 9 October 2026
Related Publications
Nearly half of Swedish dairy farms milk with robots; the US manages about 4%. Eighteen countries ranked, with the herd-size and budget arithmetic.
07 Oct 2026
Credit cards feature in 39.6% of identity theft reports to the FTC, but state shares run from Wyoming's 26.7% to Florida's 50.1%.
06 Oct 2026
Volunteers have mapped 1,378 US data center buildings in OpenStreetMap. Virginia leads with 291, but count campuses instead and Texas wins.
06 Oct 2026
Three quarters of the world was online in 2025. Of the 2.2 billion who were not, roughly 1.9 billion already live under a mobile broadband signal they do not use.
21 Sept 2026
