ITHome October 20 news: According to Reuters, NVIDIA has won a central processing unit order from yet another fast-growing AI startup. Previously, this company relied entirely on chips from Intel and AMD. This is an early sign that NVIDIA's strategy of entering this market long dominated by its two major rivals is beginning to pay off.
Cloud computing startup Modal provides compute rental services for AI developers. The company said it plans to use NVIDIA's brand-new Vera central processing unit for the first time to host AI agents, that is, AI systems capable of carrying out long-cycle tasks on behalf of humans. Before this, businesses like Modal's ran only on Intel and AMD chips.
Last week, Modal completed a US$750 million funding round (ITHome note: about RMB 5.02 billion at the current exchange rate) led by Accel, with a post-money valuation of US$15.75 billion (about RMB 105.425 billion at the current exchange rate). The company said it expects revenue to reach US$1 billion (about RMB 6.694 billion at the current exchange rate) next year. Modal said its service usage has tripled, with year-over-year revenue growth of more than 15-fold.
Modal said on Monday, local time, that early testing showed this new NVIDIA chip can double the task-processing throughput of AI agents, so it decided to adopt NVIDIA's chip solution for the first time, which required Modal to make deep changes to its underlying software.
However, CEO Erik Bernhardsson specifically emphasized that the company will not abandon its existing suppliers. Modal said it will use hybrid compute clusters, deploying NVIDIA chips in parallel with its current Intel and AMD equipment, and scheduling different tasks to the most suitable hardware.
Bernhardsson explained that the benchmark for this test was Intel and AMD chips already in mass production on the market. Both of NVIDIA's rival vendors are developing next-generation products that are expected to narrow the performance gap with NVIDIA.
"When the new products hit the market, we will test them," he added, noting that the hybrid hardware architecture means the company "can deploy the best-performing hardware at any time."
NVIDIA has told investors that its sales target for Vera chips this fiscal year is at least US$20 billion (about RMB 133.873 billion at the current exchange rate). This is NVIDIA's opening move into this market of roughly US$200 billion (about RMB 1.34 trillion at the current exchange rate). NVIDIA previously said that AI startups such as OpenAI, Anthropic, Oracle, and Perplexity also plan to adopt the chip.