Nvidia and AMD shipped roughly 98% of every desktop graphics card sold in the first half of 2026, according to Jon Peddie Research data cited by outlets including KitGuru. That leaves a sliver of the market, and a growing number of shoppers and developers are asking a simple question: who makes GPUs besides Nvidia and AMD? The answer, confirmed across reporting from BGR and other outlets this week, comes down to two real contenders plus a handful of chipmakers that design GPUs nobody can buy in a box.
Intel sells boxed, retail discrete cards under its Arc brand. Moore Threads, a Chinese company barely six years old, builds its own cards and just listed on a stock exchange at a valuation that stunned investors. Beyond those two, Apple, Qualcomm, MediaTek and Arm all design GPU silicon, but none of them sell a standalone graphics card you can drop into a desktop. The distinction matters, and it explains why the duopoly has held for two decades despite plenty of companies wanting in.
Nvidia and AMD’s Grip on the GPU Market
Jon Peddie Research tracks desktop add-in-board shipments every quarter, and the firm’s 2026 numbers show just how narrow the gap has become for everyone else. Nvidia hit 94% of discrete shipments in the fourth quarter of 2025, a new high, before settling back to around 90% in the first half of 2026 as AMD’s RDNA 4 cards gained ground, per coverage from TweakTown. AMD held roughly 8% through the first two quarters of the year. Intel’s Arc lineup sat at around 1% in Q1 and edged up toward 2% by Q2, based on the same JPR tracking. Shattered.io covered the shipment volume side of this story in detail when desktop GPU shipments hit 12.5 million units in a single quarter, the strongest total since early 2022.
Those shipment totals come against a backdrop of rising prices across the board. Memory costs tied to the AI boom, plus US tariff policy, are squeezing every GPU maker’s margins, not just the smaller players. Shattered.io tracked this directly when Nvidia GPU prices jumped 41% in Europe and AMD prices rose 28% over a comparable stretch. When even the market leaders are raising prices, the economics get brutal fast for anyone trying to break in as a third or fourth option.
Intel Arc: The Only Other Boxed Discrete GPU
Intel is the only company outside Nvidia and AMD that still sells a complete, boxed, retail discrete GPU lineup for PCs. The company entered the discrete graphics market with its first Arc cards in 2022, after years of building integrated graphics into its CPUs. BGR’s reporting describes Arc GPUs as offering decent power for the money, while characterizing the division behind them as unstable, a tension that has defined Intel’s graphics push since day one.
Battlemage and the Pricing Play
Intel’s second-generation Battlemage cards, including the B580 and B570, built a small but loyal following by undercutting Nvidia and AMD on price while packing in more VRAM at each tier. Arc Pro workstation cards extended the same approach into professional workloads. None of it has moved Intel’s shipment share much past low single digits, but the cards earned real praise from reviewers for value, which is more than most third-tier GPU attempts manage.
Is Intel Actually Staying In This Fight?
Intel executives have repeatedly pushed back on speculation that the company might quietly exit discrete graphics. “We’re not going anywhere on our discrete business,” said Tom Petersen, an Intel Fellow on the GPU team, addressing doubts about Battlemage’s future. At Computex 2026, Alex Katouzian, Intel’s EVP and GM of Client Computing, told reporters “GPUs are a super important part of our PC range,” according to TweakTown. Katouzian added separately, per HotHardware, “I think we’re just going to continue down this road.”
Intel’s commitment got a strange twist in late 2025 when Nvidia, its biggest rival, bought a stake in the company. Nvidia closed a $5 billion private placement in Intel in December 2025, taking roughly a 4% equity position, according to CNBC and Nvidia’s own investor relations announcement. The deal isn’t charity. Intel is now building custom x86 system-on-chips that integrate Nvidia RTX GPU chiplets for laptops and desktops, blurring the line between competitor and supplier in a way the GPU market hasn’t seen before. Raja Koduri, who ran Intel’s graphics division when Arc launched, framed the stakes years earlier in an interview with Beebom: “If Intel graphics become better, that means, the whole world, because, they’re like 200 million of 300 [million], and it floats all the boats. But to get there you need to create high-performance graphics, that’s what the ARC journey is about.”
Moore Threads: China’s Answer to Nvidia
Moore Threads is the clearest answer to who makes GPUs besides Nvidia and AMD outside the US semiconductor orbit. Zhang Jianzhong, a former Nvidia China executive, founded the company in October 2020. Moore Threads unveiled its first cards less than two years later, in March 2022, built around a proprietary architecture it calls MUSA, short for Moore Threads Unified System Architecture, according to the company’s Wikipedia entry and corroborating coverage.
From Nvidia Veteran to Shanghai’s First GPU Stock
Moore Threads went public on Shanghai’s STAR Market on December 5, 2025, raising nearly RMB 8 billion (about $1.1 billion) and becoming China’s first domestic GPU stock to list, per Law.asia and Jon Peddie Research’s own coverage of the filing. The stock opened wildly above its offer price on debut, a reaction that caught even seasoned China-tech watchers off guard, and the company has since announced plans for a second listing in Hong Kong. Zhang Jianzhong summed up the company’s ambition plainly in comments carried by the South China Morning Post: “Our goal is to become a leading GPU player with international competitiveness.”
What the MTT S60 and S80 Can Actually Do
The MTT S60 was Moore Threads’ debut desktop and workstation card, with the higher-performance MTT S80 following as the flagship of the same 2022 generation. BGR’s hands-on assessment was blunt: the S80 struggles to compete with current-generation Nvidia and AMD cards and runs properly in only a handful of games, a driver and software-compatibility gap that tends to hit newcomers hardest. Moore Threads might close that gap over time, particularly with fresh IPO capital behind it, but as of October 2026 it isn’t there yet. The company’s path mirrors a pattern familiar to anyone who tracks China’s broader push toward domestic AI silicon, where political pressure to de-risk from US chip supply chains is forcing homegrown vendors to scale faster than their software ecosystems can keep up.
The Chipmakers That Design GPUs but Don’t Sell Cards
Apple, Qualcomm, MediaTek and Arm all design their own GPU technology, and all four get left out of “who makes GPUs” conversations for the same reason: none of them sell a discrete graphics card a consumer can buy separately.
Apple builds its GPU cores directly into its M-series and A-series chips, and the results are strong enough that Apple’s M6 GPU has closed in on RTX 4080-class benchmark scores. The silicon is genuinely Apple’s own design, not licensed from Nvidia or AMD, but it ships soldered into a Mac or iPhone rather than as a card you can swap. Qualcomm and MediaTek take the same approach for the GPUs inside their mobile chipsets and, increasingly, their Windows-on-Arm laptop processors. Arm sits a layer further back still, licensing GPU architecture that other companies build into system-on-chip designs rather than manufacturing finished graphics products of its own. All three contribute real GPU engineering to the industry. None of them compete with Nvidia or AMD in the slot where a consumer actually buys a card.
Board Partners Aren’t GPU Makers
A lot of buyer confusion traces back to brand names stamped on the box at checkout. ASRock, Gigabyte and Sparkle all sell graphics cards, and shoppers sometimes assume that makes them GPU manufacturers in their own right. They aren’t. Each of these companies builds the circuit board, cooler and power delivery around a GPU chip they license from Nvidia or AMD. The silicon inside an ASRock card and a Gigabyte card built on the same chip performs almost identically, with the differences coming down to clock speeds, cooling and warranty terms rather than anything architectural.
Why Nobody Has Cracked the Duopoly
The software moat explains more of this story than the hardware does. Nvidia introduced CUDA in 2006, and two decades of developer tooling built on top of it now locks in everything from game engines to AI training pipelines. AMD answered with ROCm in 2016, a full decade behind, and still spends enormous engineering effort chasing feature and driver parity. Intel and Moore Threads face the same climb from an even earlier starting point. A GPU can match Nvidia on raw silicon specs and still lose, because the software stack developers already know how to use sits on the other side.
History offers a blunt reminder of how hard this market is to enter. 3dfx, S3 Graphics and Matrox all sold competitive discrete GPUs in the late 1990s and early 2000s, and all three either exited the business or shrank into niche players once Nvidia and ATI (later absorbed by AMD) pulled ahead on both performance and software support. The current duopoly isn’t an accident of recent market conditions. It’s the outcome of a consolidation that’s been running for more than 20 years, and it’s exactly why Intel’s and Moore Threads’ current struggles look so familiar to anyone who remembers that earlier shakeout.
GPU Makers at a Glance, 2026
| Company | GPU Product | Sold as Discrete Card? | 2026 Status |
|---|---|---|---|
| Nvidia | GeForce RTX, RTX Pro | Yes | ~90% of desktop AIB shipments (JPR) |
| AMD | Radeon RX | Yes | ~8% of desktop AIB shipments (JPR) |
| Intel | Arc (Battlemage B580/B570, Arc Pro) | Yes | ~1-2% of desktop AIB shipments (JPR) |
| Moore Threads | MTT S60, MTT S80 (MUSA architecture) | Yes | Listed on Shanghai STAR Market, Dec. 2025 |
| Apple | Integrated GPU in M-series/A-series | No, soldered into own hardware | Proprietary design, no card sold |
| Qualcomm | Adreno (integrated) | No | Mobile and Windows-on-Arm chips only |
| MediaTek | Integrated GPU | No | Mobile and laptop chipsets only |
| Arm | Mali / Immortalis GPU IP | No | Licensed into third-party SoCs |
Desktop Discrete GPU Shipment Share, Q4 2025 to Q2 2026
| Vendor | Q4 2025 | Q1 2026 | Q2 2026 |
|---|---|---|---|
| Nvidia | ~94% | ~90% | ~90% |
| AMD | ~5-6% | ~8% | ~8% |
| Intel | ~1% | ~1% | ~2% |
Figures are approximate and drawn from Jon Peddie Research tracking as reported by KitGuru and TweakTown. JPR has flagged that high memory prices, driven by AI demand and US tariff policy, are squeezing the entire add-in-board market, a trend that hits smaller vendors with thinner margins first.
Competitive Comparison: Arc vs. MUSA vs. the Duopoly
Intel’s Battlemage cards compete reasonably well on price-to-performance in mainstream gaming, with the B580 regularly cited by reviewers as solid value at its price point. Driver maturity, after years of rocky launches, has also improved noticeably since Arc’s 2022 debut. Where Intel still lags is content creation and professional workloads, where Nvidia’s CUDA ecosystem remains close to unavoidable for many studios and researchers.
Moore Threads sits a tier behind Intel on raw gaming performance and game compatibility, based on BGR’s assessment of the S80, but it holds an advantage Intel doesn’t: a captive domestic market where Beijing is actively encouraging buyers to choose Chinese silicon over US imports. That policy tailwind, not pure technical parity, is what’s driving Moore Threads’ valuation more than its current driver stack. Apple’s integrated GPUs, by contrast, post genuinely strong benchmark numbers inside their own ecosystem but were never built to compete in the open discrete-card market at all, which keeps them out of this comparison entirely despite the silicon being excellent.
Market Impact: What This Means for Buyers and Builders
For anyone building a PC in late 2026, the practical takeaway is narrow but useful. Intel Arc is the only realistic alternative to Nvidia or AMD that you can actually buy at a US retailer today, and it’s worth shortlisting at the budget and mid-range tiers specifically. Moore Threads cards aren’t sold outside China in any meaningful retail channel, so they’re not yet a factor for most shoppers reading this from the US or Europe. The bigger market impact is structural: with Nvidia now holding an equity stake in Intel and the two companies co-developing chips, the industry’s competitive lines are getting blurrier, not clearer. A 4% Nvidia stake in Intel doesn’t erase Arc as a brand, but it does raise real questions about how independently Intel’s graphics roadmap gets to evolve from here.
Component pricing adds urgency to the question. Shoppers already watching flagship Nvidia cards vanish from US shelves at inflated prices have a real financial incentive to take a cheaper third option seriously, even an imperfect one. Intel’s pricing strategy on Battlemage leans directly into that gap.
Historical Context: A Two-Decade Consolidation
Nvidia has held the majority of GPU shipment share continuously since roughly 2006, the same year CUDA launched. That’s not a coincidence. The GPU market of the late 1990s had far more competitors than it does today, including 3dfx, Matrox, S3 Graphics and a since-absorbed ATI. Each lost ground once the market tipped toward a winner-take-most dynamic built on software lock-in rather than chip specs alone. AMD survived that consolidation only by acquiring ATI outright in 2006, buying its way into the race rather than building from scratch, a bet that eventually helped push AMD past a $1 trillion market cap two decades later.
Nvidia’s dominance hasn’t been free of regulatory friction along the way. The SEC fined the company $5.5 million in 2022 after finding that Nvidia failed to disclose, during fiscal 2018, that cryptocurrency mining was driving a meaningful share of what it reported as gaming GPU revenue growth, according to the SEC’s own settlement and reporting from PCWorld. That episode is a reminder that even the dominant player’s growth story has had messy edges, even as its market position has only strengthened since.
Predictions: Where the Alternative GPU Market Goes From Here
- Intel stays in discrete graphics, but the Nvidia tie-up changes the roadmap. Expect Intel to keep shipping Arc-branded cards while simultaneously building Nvidia RTX chiplets into mainstream x86 laptop SoCs, a dual-track strategy that would have looked contradictory a year ago.
- Moore Threads expands its Chinese retail footprint before it exports seriously. With fresh STAR Market capital and a pending Hong Kong listing, the company is more likely to spend 2027 building domestic market share than chasing US or European buyers.
- AMD’s share stays flatter than Nvidia’s. RDNA 4 has helped AMD hold roughly 8% of shipments through the first half of 2026, and nothing in current JPR tracking points to a sharp move in either direction soon.
- Intel’s shipment share stays in low single digits through 2027. Battlemage’s value pricing earns praise from reviewers, but praise hasn’t translated into volume, and that gap is unlikely to close quickly given CUDA’s grip on professional and AI workloads.
- Memory pricing pressure keeps squeezing every vendor, including the leaders. JPR has already flagged AI-driven memory costs and tariffs as a drag on the add-in-board market broadly, and that pressure falls hardest on vendors without Nvidia’s margins to absorb it.
Frequently Asked Questions
Who makes GPUs besides Nvidia and AMD?
Intel sells discrete Arc GPUs at retail, and China’s Moore Threads builds its own MTT-series cards. Apple, Qualcomm, MediaTek and Arm all design GPU silicon too, but they integrate it into their own chips rather than selling standalone graphics cards.
Is Intel Arc actually worth buying in 2026?
Battlemage cards like the B580 and B570 get consistent praise from reviewers for price-to-performance at budget and mid-range tiers. Driver maturity has improved significantly since Arc’s rocky 2022 launch, though content-creation and professional workloads still favor Nvidia’s CUDA ecosystem.
Can I buy a Moore Threads GPU outside China?
Not through any meaningful retail channel as of October 2026. Moore Threads cards remain primarily a domestic Chinese product, sold to a market where government policy actively favors homegrown silicon over US imports.
Why don’t Apple, Qualcomm or MediaTek count as GPU competitors to Nvidia?
They design real GPU silicon, but it ships soldered into their own devices rather than as a card a consumer can buy separately and install in a desktop PC. That distinction, not the quality of the engineering, is why they’re left out of “who makes GPUs” comparisons.
Are ASRock and Gigabyte GPU manufacturers?
No. They’re board partners that build the circuit board, cooling and power delivery around a chip they license from Nvidia or AMD. The underlying GPU silicon on an ASRock card and a Gigabyte card built from the same chip is identical.
Why does Nvidia own a stake in Intel if they’re competitors?
Nvidia closed a $5 billion investment in Intel in December 2025, taking roughly a 4% equity stake as part of a deal where Intel builds custom x86 chips integrating Nvidia RTX GPU chiplets. It’s a supplier relationship layered on top of a competitive one, not a merger.
What percentage of the GPU market does Nvidia control in 2026?
Jon Peddie Research tracked Nvidia at roughly 90% of desktop discrete GPU shipments through the first half of 2026, down slightly from a peak near 94% in late 2025 as AMD’s RDNA 4 cards gained share.
Will Moore Threads or Intel ever catch Nvidia?
Nothing in current shipment data points that direction. Both companies face the same structural problem: Nvidia’s two-decade CUDA software advantage, not raw chip specs, is what keeps the duopoly intact, and neither challenger has closed that gap yet.
