Nvidia reports strong China demand for H200

Nvidia told analysts on Monday that demand for its H200 chips in China is “strong,” and there’s no shortage of supply from them.

In the live streamed event meeting at CES in Las Vegas, Nvidia CFO Colette Kress said the company has already submitted license requests to U.S. authorities and that “the government is deciding what it wants to do with them.”

Colette added, “We have enough supply to meet China’s demand without affecting shipments to other customers around the world.”

Beijing, though, hasn’t made things simple, as Cryptopolitan previously reported, Xi Jinpong has discouraged agencies and local companies from using Nvidia’s older H20 chip, calling it underpowered.

At the CES meeting, Nvidia also dropped details about robotaxi deals, saying that it is working with ride-hailing platforms to power fleets of autonomous cars using its Drive AV stack and AI hardware. The goal is to roll them out as early as 2027. This includes a partnership with Uber, which was first announced in October.

Since 2015, Nvidia’s been pushing chips under the Drive brand, but it’s automotive and robotics revenue hit only $592 million last quarter, which is barely 1% of total income. To cut time and costs for automakers, the firm is offering its Drive AGX Thor computer for $3,500 per chip. The system includes custom tuning options, like how hard a car accelerates.

Nvidia reveals full AV stack, Vera Rubin system, and chip ecosystem at CES

At CES, Nvidia CEO Jensen Huang also introduced Alpamo, an AI that controls self-driving cars. It’s trained from end to end; meaning cameras to steering, braking, and acceleration. “It tells you what action it’s going to take, the reason behind it, and the path it’ll follow,” he said. The data comes from human drivers, simulations from Cosmos, and hundreds of thousands of labeled examples, according to Jensen.

This AI powers cars like the new Mercedes-Benz CLA, which he said was rated “the world’s safest” by NCAAP. Alpamo connects to a dual safety system, and if the main AI stack can’t handle a situation, a backup system kicks in.

Nvidia claims it’s the only car in the world running both AI and traditional AV stacks at once.

Jensen also named the company’s new system Vera Rubin, already in production. It has 1,152 GPUs spread across 16 racks, each containing 72 Rubin modules. These modules link to Reuben GPUs and Vera CPUs; two huge chips mashed into one setup. Vera doubles performance per watt over earlier CPUs.

Bluefield 4, the new processor, is also baked into every unit. It slices data centers into chunks, so different users can share one space without stepping on each other. All of this is packaged under the MGX standard, made from 80,000 parts. Foxconn, Quanta, Wistron, HP, Dell, and Lenovo are all using it.

Despite doubling the power use, Vera Rubin’s cooling needs stay the same, as it still runs on 45°C water, with no chillers required. Even with the size boost, the whole platform runs on a new silicon photonics process developed with TSMC, delivering 512 ports at 200 gigabits per second, pushed right to the chip.

Sign up to Bybit and start trading with $30,050 in welcome gifts