Ask someone to name a GPU maker and they will say Nvidia, then AMD. That answer is correct but incomplete. A small group of chipmakers designs graphics silicon too, they just never put it in a box you can buy at a store. Intel and a Chinese startup called Moore Threads are the only other companies selling a boxed discrete card in 2026. Apple, Qualcomm, MediaTek and Arm all design GPU hardware that ships inside billions of devices, yet none of them compete for a slot in your desktop tower.
That split explains more about the GPU business than the market-share numbers do. It is a story about two different industries that both happen to need graphics processors: PC gaming, where Nvidia and AMD have run the table for two decades, and mobile or edge computing, where a completely different set of companies builds GPUs nobody ever unboxes. Shattered.io has already covered the discrete-card side of this story in detail, including how Intel and Moore Threads fight over the GPU market’s remaining 2%. This piece looks at the other half of the question: who else designs GPUs, and why do they stay out of the box entirely.
Nvidia and AMD’s Grip on the Discrete GPU Market
Jon Peddie Research tracks desktop add-in-board shipments every quarter, and a market-data compilation citing JPR figures puts Nvidia’s share of discrete GPU shipments at roughly 90% in the first quarter of 2026, with AMD near 8% and Intel Arc around 1%. Those numbers bounce a little from quarter to quarter but the shape stays the same. Through 2025, the same compilation lists Nvidia’s discrete share at approximately 92% in the first quarter, 94% in the second, 92% in the third and 94% in the fourth. Outlets citing this data note it should be checked against JPR’s own primary release, since it circulates as a secondary compilation rather than a direct JPR report, but the trend line is consistent across every source: Nvidia and AMD split somewhere between 98% and 99% of every discrete card sold.
Shattered.io tracked the pricing side of that dominance when Nvidia GPU prices surged 41% in Europe and AMD prices rose 28% over a comparable stretch, and again when desktop GPU shipments hit 12.5 million units in a single quarter, the strongest total since early 2022. Memory costs tied to the AI boom and ongoing tariff policy are squeezing every GPU maker’s margins right now, not just the smaller players. When even the two companies that run the table are raising prices, the economics get brutal fast for anyone trying to break in as a third option.
Who Else Actually Designs GPU Silicon
Narrow the question to “who sells a discrete card” and the duopoly looks almost total. Broaden it to “who designs GPU hardware,” and the list grows considerably. Intel builds boxed, retail discrete cards under its Arc brand. Moore Threads, a Chinese company founded in 2020, designs and sells its own cards too, led by the S60, its first consumer-grade graphics card released in 2022. Beyond those two, Apple, Qualcomm, MediaTek and Arm all design GPU silicon that reaches hundreds of millions of devices every year. None of them sell a standalone card you can drop into a desktop, and that distinction is the whole story.
It is worth separating the designers from the board partners here too, since the two groups get confused constantly. Brands like ASRock, Gigabyte and Sparkle sell graphics cards, but they build those cards around Nvidia or AMD GPU chips rather than designing their own silicon. They are manufacturing partners, not GPU architects, and lumping them in with Intel or Moore Threads overstates how many real GPU designers exist.
Intel Arc and Moore Threads: The Two Companies That Sell You a Card
Intel entered the discrete graphics market with its first Arc cards in 2022, after decades of shipping integrated graphics inside its CPUs. Coverage of the lineup describes Arc GPUs as offering decent power for the money, while characterizing the division behind them as unstable, a tension that has followed Intel’s graphics push from day one. Moore Threads took a different path, building a domestic Chinese GPU business from a 2020 founding to a shipping consumer card in just two years with the S60.
Both companies prove the hard part of this business is not designing a chip, it is everything around the chip: drivers, game compatibility, retail distribution and the years of software tuning that Nvidia and AMD have banked. Shattered.io’s earlier reporting on the GPU market’s remaining slice goes deeper into how Intel and Moore Threads are fighting for scraps of a market Nvidia and AMD already own. The four companies covered below never even attempt that fight, and understanding why is the more interesting question.
Apple’s GPU: Built In-House, Never Sold Separately
Apple designs its own GPU cores and bakes them directly into every M-series and A-series chip it ships. There is no standalone Apple graphics card, no retail SKU, no box on a shelf. The GPU exists only as one block among many inside a single piece of silicon that also carries the CPU, neural engine and memory controller. That integration is the entire point of Apple’s chip strategy, and it is why the company has never shown any public interest in selling graphics hardware on its own.
Apple’s GPU performance claims get plenty of coverage on their own merits, including shattered.io’s look at how the M-series GPU scores are closing in on discrete desktop cards in raw benchmark terms. The irony is that Apple’s graphics silicon is powerful enough to draw comparisons to Nvidia’s desktop lineup, yet a MacBook owner can never buy an upgraded Apple GPU separately the way a PC gamer buys a new RTX card. The chip and the computer are sold as one unit, permanently.
Qualcomm’s Adreno: The Most-Shipped GPU You’ll Never See in a Box
Qualcomm’s Adreno graphics architecture ships inside more phones than any single discrete GPU brand ships graphics cards, and the company’s most recent flagship platform shows why mobile GPU design has become its own demanding discipline. Qualcomm announced the Snapdragon 8 Elite Gen 5 on September 24, 2025, claiming a 23% jump in GPU performance over its predecessor alongside a 20% cut in power consumption and a 25% improvement in ray-tracing performance. Qualcomm’s own product page lists the platform’s third-generation Oryon CPU design hitting a peak clock of 4.74GHz, with the GPU claims repeated verbatim from the September announcement.
Interestingly, Qualcomm did not name the GPU in its initial press materials, according to GSMArena’s coverage of the launch. Specialist outlets later identified it as the Adreno 840, distinguishing it from the Adreno 829 used in the standard Snapdragon 8 Gen 5 and the Adreno 830 in the prior-generation Elite chip. The platform also introduces what Qualcomm calls Adreno High Performance Memory, an 18MB cache aimed at game developers that the company says can cut power draw by up to 10%, running at an estimated 1.2GHz versus roughly 1.1GHz on the previous generation, according to Android Authority’s teardown.
Snapdragon 8 Elite Gen 5’s Adreno 840, by the Numbers
Independent testing gives a clearer picture of what those percentage claims mean in practice. TechPowerUp’s report on Geekerwan’s benchmarking found the Adreno 840 running Honkai Impact 3rd at roughly 59.6 frames per second at 1680×770 resolution while drawing about 5.88 watts, and Wuthering Waves at approximately 59.8 frames per second using close to 5.13 watts. In Delta Force, the same testing recorded around 164.9 frames per second at about 4.7 watts. Qualcomm also claims the platform’s broader efficiency gains translate into up to one hour and 48 minutes of additional gaming time per charge, according to GSMArena.
Qualcomm has since moved its flagship lineup forward again with the Snapdragon 8 Elite Gen 6, built on a 2nm process and clocked up to 5.11GHz, which shattered.io covered in detail when that chip launched. The company is also pushing GPU-adjacent silicon upmarket, as shattered.io reported when Qualcomm’s newest AI chip went head-to-head with Apple’s A20 Pro. None of that changes the underlying pattern: every generation of Adreno ships inside someone else’s phone, never as a product Qualcomm sells to you directly.
MediaTek’s GPU Silicon: Volume Over Visibility
MediaTek designs integrated GPUs that ship inside its Dimensity chip lineup, which powers a large share of the mid-range and budget Android phones sold worldwide, along with select laptops and smart TVs. Unlike Qualcomm’s Snapdragon marketing, MediaTek’s GPU work rarely makes headlines on its own, largely because the company competes on value and volume rather than flagship bragging rights. The chips still have to run modern mobile games and increasingly on-device AI workloads, just without the benchmark spotlight that follows Qualcomm’s top-tier releases.
That lower profile does not mean lower stakes. MediaTek’s GPUs reach an enormous installed base simply because Dimensity chips ship in such high volume across budget and mid-tier devices globally. The company has never signaled any interest in building a discrete graphics card, and the business logic tracks with everyone else on this list: a GPU core bundled into a mobile chipset serves a completely different customer than a standalone card sold to PC gamers.
Arm’s Mali and Immortalis: The GPU Blueprint Licensed Worldwide
Arm occupies a different spot in this picture entirely. The company does not manufacture finished chips at all, it designs GPU architectures, Mali and the newer Immortalis line among them, and licenses those designs to other chipmakers who build them into their own system-on-chip products. That licensing model means Arm’s GPU designs end up inside an enormous range of devices without Arm ever touching a factory floor or a retail shelf.
Arm’s broader silicon ambitions have been drawing attention well beyond mobile graphics, including shattered.io’s coverage of how the Arm AGI CPU is taking direct aim at x86 server chips. But the GPU side of Arm’s business stays true to its licensing roots. A discrete Arm graphics card simply does not fit the company’s model, since Arm’s entire value proposition rests on letting partners build and sell the finished silicon under their own brands.
Why None of Them Build a Discrete Card
The pattern across Apple, Qualcomm, MediaTek and Arm is not an accident of timing, it reflects a genuinely different business model from the one Nvidia and AMD run. Integrated GPUs exist to serve a chip that someone else packages into a finished product, whether that is an iPhone, a Snapdragon-powered Android flagship, or a budget tablet built around a Dimensity chip. The GPU is one feature among many, bundled into a single piece of silicon and priced as part of the whole device rather than sold on its own merits.
A discrete graphics card lives in a completely different world. It has to work with Windows, with every major game engine, with a decade of legacy software, and with a buyer who expects to swap it out and upgrade it independently of everything else in their PC. Building that ecosystem from scratch is precisely what has made Intel’s Arc push so difficult, and it is the exact cost that Apple, Qualcomm, MediaTek and Arm have no reason to take on when their existing business already works.
The Software and Driver Problem
Drivers are the part of this equation that rarely gets discussed outside engineering circles, but they explain a lot of why the duopoly has held for so long. Nvidia and AMD maintain driver stacks tuned against thousands of individual PC games, refined over roughly two decades of iteration. A mobile GPU vendor switching to desktop would need to build that entire stack from nothing, then convince developers and reviewers to trust it, all while Nvidia and AMD keep shipping new cards every year. That is a multi-year, capital-intensive bet with no guarantee of payoff, and it is a bet none of these four companies has chosen to make.
Historical Context: How Nvidia and AMD Built a Two-Decade Duopoly
The discrete GPU market was not always a two-horse race. 3dfx helped define consumer 3D graphics in the late 1990s before Nvidia absorbed its assets. S3 Graphics, Matrox and SiS all sold discrete cards at various points over the following decade, and each one eventually exited the consumer graphics business as the cost of competing with Nvidia and ATI, later AMD, grew too steep. By the mid-2010s, Nvidia and AMD were effectively the last two standing in the PC card business, a position the current market-share data suggests they have only tightened since.
Intel’s 2022 Arc launch marked the first serious new entrant in roughly fifteen years, and Moore Threads’ domestic Chinese push the same year marked the first credible attempt from outside the usual US-based competitors. That both companies entered within months of each other, after a decade and a half with no new discrete competitor at all, says something about just how forbidding the barriers to entry had become. It also underscores why Apple, Qualcomm, MediaTek and Arm, despite having the engineering talent to design competitive GPU silicon, have stayed entirely on the integrated side of that line.
Market Impact: Edge AI Is Changing the Calculus
The rise of on-device AI is giving integrated GPU makers a reason to invest heavily in graphics silicon without ever needing a discrete card to show for it. Phones and laptops now run AI features locally, from image generation to voice processing, and much of that workload runs on the same GPU cores that used to exist mainly for gaming and UI rendering. Qualcomm’s emphasis on Adreno’s efficiency gains, and its separate push into dedicated AI silicon that shattered.io covered against Apple’s A20 Pro, both point toward GPU investment increasingly being justified by AI performance rather than gaming benchmarks alone.
That shift raises the stakes for integrated GPU design even as it reinforces why these companies stay out of the discrete card business. A faster on-device AI chip sells more phones and laptops directly. A discrete GPU card would compete in an entirely separate market against two entrenched incumbents, for a return that looks small next to the billions of integrated chips these companies already ship every year.
Competitive Comparison: Discrete vs. Integrated GPU Makers
| Company | GPU Brand | Sells a Standalone Card? | Primary Market | Entered GPU Business |
|---|---|---|---|---|
| Nvidia | GeForce / RTX | Yes | PC gaming, data center | 1999 |
| AMD | Radeon | Yes | PC gaming, data center | 2006 (ATI acquisition) |
| Intel | Arc | Yes | PC gaming, laptops | 2022 |
| Moore Threads | MTT | Yes | PC gaming (China-focused) | 2020 (founded) |
| Apple | Apple GPU | No | Mac, iPhone, iPad SoCs | In-house since 2010s |
| Qualcomm | Adreno | No | Android smartphones, laptops | Acquired from AMD, 2009 |
| MediaTek | Dimensity GPU | No | Budget/mid-range Android, TVs | Integrated SoC design |
| Arm | Mali / Immortalis | No | Licensed to SoC partners | Architecture licensing |
Laid out this way, the split is stark. Four companies sell something you can buy and install, four design GPUs that reach you only bundled inside a finished device. The board-partner brands like ASRock, Gigabyte and Sparkle sit outside this table entirely, since they assemble cards around Nvidia or AMD chips rather than designing their own GPU silicon.
Discrete GPU Market Share: Nvidia and AMD vs. Everyone Else
| Quarter | Nvidia Share | AMD Share | Intel Arc Share |
|---|---|---|---|
| Q1 2025 | ~92% | ~7-8% | ~0-1% |
| Q2 2025 | ~94% | ~5-6% | ~0-1% |
| Q3 2025 | ~92% | ~7-8% | ~0-1% |
| Q4 2025 | ~94% | ~5-6% | ~0-1% |
| Q1 2026 | ~90% | ~8% | ~1% |
These figures come from a market-data compilation citing Jon Peddie Research’s discrete add-in-board tracking, and outlets that cite them flag that the numbers should be checked against JPR’s own primary release rather than treated as a direct quote. Even allowing for quarter-to-quarter noise, Moore Threads does not register as a meaningful slice of global shipments in this dataset at all, underlining just how narrow the opening is for anyone outside Nvidia and AMD, Intel included.
Could Apple, Qualcomm, MediaTek or Arm Ever Go Discrete?
Nothing in current reporting suggests any of the four is planning a discrete graphics card, and the business logic above explains why that is unlikely to change soon. Apple’s model depends on total control over its own hardware and would gain little from selling a standalone GPU that competes against its own integrated chips. Qualcomm and MediaTek could theoretically scale up a mobile GPU architecture for desktop use, but doing so would mean building an entirely new driver ecosystem from scratch to compete against two incumbents with two decades of software investment behind them.
Arm’s position is the clearest of all. Its entire business model is licensing designs to partners rather than selling finished chips, so a discrete Arm-branded card would contradict the company’s core strategy rather than extend it. If any of these four ever does enter the discrete market, the far more likely path is through a partner building a product around their architecture, similar to how Moore Threads and Intel built out their own card businesses rather than licensing someone else’s design.
Predictions: What Happens Next in the GPU Market
- Nvidia and AMD will keep splitting somewhere between 97% and 99% of discrete shipments through the rest of 2026, with Intel Arc making only incremental gains as it works through driver maturity issues.
- On-device AI demand will keep pushing Qualcomm, MediaTek and Arm to invest more heavily in integrated GPU performance, even as none of them show signs of building a standalone card.
- Apple will keep widening the gap between its integrated GPU benchmarks and discrete-card performance claims, using that comparison for marketing even though the two products are not actually interchangeable.
- Moore Threads will remain the clearest test case for whether a brand-new entrant can carve out sustained share against the duopoly, rather than a one-off product cycle.
- Expect continued confusion in consumer searches between GPU chip designers and GPU board partners, since brands like ASRock, Gigabyte and Sparkle will keep selling cards that use someone else’s silicon.
None of these predictions point toward a near-term disruption of the Nvidia-AMD duopoly. The more interesting shift is happening one layer down, where Apple, Qualcomm, MediaTek and Arm keep pouring engineering resources into GPU design without ever needing to sell you the chip directly. That is arguably a more durable business than chasing a sliver of the discrete card market the way Intel and Moore Threads currently do.
Frequently Asked Questions
Who makes GPUs besides Nvidia and AMD?
Intel sells discrete Arc graphics cards, and Moore Threads, a Chinese company founded in 2020, sells its own cards starting with the S60. Apple, Qualcomm, MediaTek and Arm also design GPU silicon, but none of them sell a standalone graphics card.
Does Apple make its own GPU?
Yes. Apple designs its own GPU cores and integrates them directly into its M-series and A-series chips. There is no separate Apple graphics card available for purchase.
What GPU does Qualcomm use in its Snapdragon chips?
Qualcomm’s current flagship GPU architecture is Adreno, with the Snapdragon 8 Elite Gen 5 using what specialist outlets identified as the Adreno 840 after Qualcomm’s September 2025 launch materials did not name the GPU directly.
Is Arm a GPU company?
Arm designs GPU architectures, including Mali and Immortalis, and licenses those designs to other chipmakers rather than manufacturing finished chips or cards itself.
Why doesn’t Qualcomm or MediaTek sell a discrete graphics card?
Building a competitive discrete card requires years of driver development and game compatibility work that Nvidia and AMD have already invested in over roughly two decades. Qualcomm and MediaTek’s GPUs are designed to serve their own mobile chipsets rather than compete as standalone products.
Are ASRock, Gigabyte and Sparkle GPU makers?
No. They are board-partner brands that manufacture and sell graphics cards built around Nvidia or AMD GPU chips. They do not design their own GPU silicon.
How much of the discrete GPU market do Nvidia and AMD control?
A market-data compilation citing Jon Peddie Research puts Nvidia’s share of discrete add-in-board shipments at roughly 90% in the first quarter of 2026, with AMD near 8% and Intel Arc around 1%, figures that outlets note should be checked against JPR’s primary release.
Could Moore Threads challenge Nvidia and AMD eventually?
Current shipment data does not show Moore Threads registering a meaningful global share yet. Reports describe the company’s long-term competitiveness as unconfirmed rather than an established trend.
