GPU Designer
A GPU designer is a company that designs graphics and AI computing chips but usually does not manufacture them itself. The best-known are Nvidia, AMD, and Intel; their designs are fabricated by contract manufacturers such as TSMC.
Inside every computer, alongside the main processor, there is often a second computing unit designed to perform very many similar calculations simultaneously. This unit is called a GPU, short for Graphics Processing Unit, in other words a graphics processor. Originally it was meant to render images for games; today it mainly crunches the vast piles of numbers that make up artificial intelligence. A GPU designer is a company that designs such chips: it determines which computing units exist, how they are wired together, and which software controls them. The design is then manufactured in outside factories, since most of these companies do not have their own chip fabs. The three best-known names are Nvidia, AMD, and Intel.
Why a chip design decides the fate of the AI boom
The rise of AI depends on computing power. Training a large language model means carrying out trillions of computing steps for weeks on end. Whoever designs the fastest chips thereby determines how large and how good the models that can even be built will be. That’s why GPU designers aren’t mere suppliers, but a kind of bottleneck for the entire industry.
This shows up in the numbers. Nvidia was a gaming company for decades and briefly became the most valuable company in the world in 2024. The reason isn’t a consumer product, but the sale of computing accelerators to data centers. Individual cards of this kind cost as much as a small car, and major AI providers buy them by the tens of thousands.
Because there are so few providers, the market is also political. Governments restrict which chips may be sold to which countries. For a GPU designer, this means a new design not only has to be fast, but also has to comply with export rules. That’s why Nvidia deliberately designed throttled-down variants of its chips for China.
From circuit diagram to finished card
A GPU designer initially works only on a computer. Engineers use a special language to describe which circuits the chip should contain. From this description emerges a blueprint with billions of tiny switches, so-called transistors. Up to this point, nothing has yet been made from silicon — only a very precise set of construction instructions exists.
These instructions go to a contract manufacturer, known in industry jargon as a foundry. The global leader is the Taiwanese conglomerate TSMC. This model is called fabless, meaning without a factory: design and manufacturing are separate businesses. The advantage is that a designer doesn’t have to build a factory costing tens of billions of dollars. The disadvantage is dependence on a handful of factories in a politically tense region.
A design involves more than just hardware. For programmers to be able to use the computing units, software is needed as well. Nvidia’s platform CUDA is exactly that: a collection of tools that lets AI programs run on Nvidia chips. Almost all AI software is tailored to it. This software lead often protects the business more than the chip itself.
Where these names turn up in the news
In everyday life, you encounter the result in the graphics card of a gaming PC or in the chip of a game console. When you ask a chatbot something, the answer likewise runs on GPUs — just in a data center somewhere in the world. The graphics card on the store shelf and the computing accelerator in the server rack often come from the same design family.
In financial news, GPU designers mostly show up in quarterly earnings reports. If Nvidia’s revenue rises more than expected, that’s taken as proof that the AI buildout is continuing. If the numbers disappoint, the share prices of many tech companies often drop at once. Conferences such as Nvidia’s GTC are also closely watched, since new chip generations are unveiled there.
A common mistake is to lump GPU designers and chip factories together. Nvidia manufactures nothing, TSMC designs nothing. Intel is the special case, because it does both. And more and more customers are becoming designers themselves: Google, Amazon, and Apple are developing their own AI chips in order to rely less on outside purchases.