March 2, 2024
Nvidia's Q2 earnings prove it's the big winner in the generative AI boom


In August last year, a ban on Nvidia’s chip export to China sent the country’s budding artificial intelligence startups scrambling for alternatives. A momentary sense of relief came when Nvidia unveiled chips with reduced performance to bypass export restrictions. But this respite was short-lived.

On Tuesday, the Biden administration announced a slew of measures to curb Beijing’s military ambitions, including a further restriction on Nvidia’s AI chip shipments to China. A800 and H800, the two AI chips Nvidia designed specifically to continue shipping to China, will be hit by the fresh round of new rules.

The chip bans are originally targeted at China’s military use, but the more visible victims are arguably the country’s raft of startups riding the rapid advancement of large language models. Many of them rushed to stockpile Nvidia’s A100 and H100 before the bans went into effect, shelling out millions of dollars for the inflated costs. Alibaba, Baidu, ByteDance, and Tencent, collectively ordered $5 billion of A800 chips this year and the next, according to a report from the Financial Times.

Nascent startups, meanwhile, are driven to raise venture capital hastily to support their costly AI dreams.

The U.S. chip bans have not stopped Chinese giants from pursuing their AI ambitions. On Tuesday, Baidu unveiled the latest version of its flagship foundation model, Ernie 4.0, and claimed that it now matches GPT4. To date, Ernie has amassed 45 million users, the company claimed.

The challenge of entering China’s AI fray goes beyond increasingly limited semiconductor access. The country’s regulations require that LLM-based services obtain a license before serving public users, a test of companies’ government relations and their ability to navigate the red tape.

Ultimately, a shortage of high-end chips and the intricate nature of Beijing’s censorship requirements created an environment for generative business intelligence services to flourish, as they require less computational power (data sources are internal rather than the entire internet) and are comparatively easier to control as prompts are more scenario-based. Qianfan, Baidu’s enterprise-facing AI platform built on Ernie, has amassed some 17,000 customers.





Source link