Vera Rubin

Vera Rubin is the name of a new computing generation from the US company Nvidia. Nvidia builds the specialized chips that power almost all major AI systems. Such chips are called GPUs because they were originally designed for graphics in computer games. They compute a great many simple tasks simultaneously, and that is exactly what AI needs. Nvidia gives each chip generation a name from the history of science. Vera Rubin was a US astronomer whose measurements of galaxies in the 1970s showed that dark matter must exist.

Why a chip generation moves stock prices

Nvidia dominates the market for AI chips almost single-handedly. As a result, the company has become one of the most valuable in the world. When Nvidia announces a new generation, it involves orders worth many billions of dollars. Companies like Microsoft, Amazon, Google, and OpenAI plan their data centers years in advance. They need to know early on what the next generation will be capable of.

Nvidia has set itself a fixed rhythm: roughly one new platform per year. Hopper was followed by Blackwell, and Blackwell is followed by Vera Rubin. An expanded version has already been announced for 2027. This fast pace puts pressure on competitors like AMD. But it also puts pressure on customers, because whoever skips a generation ends up computing at a higher cost than the competition.

For investors, Vera Rubin is also a signal of how much longer the AI construction boom will last. A data center full of chips costs billions and must pay off over years. Doubts about that cause shares across the whole industry to fluctuate, from chipmakers to power utilities.

Two names, two components

The name consists of two parts because it refers to two chips. Vera is a main processor, meaning the chip that controls the process flow and fetches data. Rubin refers to the actual compute chips for AI calculations. Both sit together on a single board and are connected by extremely fast links. This short distance matters because, in AI, transport latency is often a bigger problem than raw computing power.

A Rubin chip consists of several smaller pieces of silicon installed side by side. Next to it sits especially fast memory, called HBM, which is attached directly to the chip. Nvidia promises that Vera Rubin will deliver several times the performance of Blackwell at a similar power consumption per computing step. Such figures come from the manufacturer itself and should be read with caution.

These chips are rarely sold individually. Nvidia delivers entire racks in which dozens of chips work together like a single large computer. This includes networking technology, cooling with water instead of air, and the software. A single rack requires as much power as several hundred households.

From the data center to the smartphone app

A private individual will not buy Vera Rubin directly. These systems sit in the data centers of cloud providers, who rent out computing time by the hour. Anyone using a chatbot or generating an image is indirectly using hardware like this. Every response is computed on chips of this kind, usually thousands of kilometers away.

The name mainly appears in the news around Nvidia’s quarterly earnings and its in-house trade show, GTC. There, release dates, performance figures, and customer names are announced. It is often also reported how many orders have already been placed before the first chip has even shipped.

A common misconception: Vera Rubin is not an AI model. A model like GPT is software, whereas Vera Rubin is the hardware underneath it. And today the name is used twice over: a new large telescope in Chile is also called the Vera C. Rubin Observatory. Both are named after the same astronomer but have nothing to do with each other.