Spicy Competition May Cause Heartburn for Nvidia
OpenAI showed off its aptly named new semiconductor at its Hot Chips conference Tuesday, the day before Nvidia’s earnings.
Sign up for smart news, insights, and analysis on the biggest financial stories of the day.
Nvidia snapped a seven-day losing streak fueled by concerns about its outpouring of capital into the AI industry one day before CEO Jensen Huang and CFO Colette Kress shared the chipmaker’s blowout second-quarter results.
Now, it can point cash to back up its budget. The world’s largest company reported record revenue yesterday of $96 billion, fueled by still-rising demand for AI chips and data centers from hyperscalers including Meta and Amazon. Nvidia has its sights set on 12-figure revenue in the third quarter, but competition for its chips is heating up.
Don’t Forget the Tilde
OpenAI showed off its aptly named new semiconductor at the Hot Chips conference Tuesday, the day before Nvidia’s earnings. The Jalapeño chip was developed by OpenAI and Broadcom, using OpenAI’s models to customize the chip while Broadcom took care of the manual tinkering. OpenAI, which expects the chip to debut on a small scale by the end of this year, isn’t the only Nvidia customer trying to make its own chips:
- Google plans to launch a tensor processing unit specifically tailored to training AI models, and another TPU that’s customized for inference work. Meta, similar to OpenAI, has been developing chips in collaboration with Broadcom. The social-media-slash-AI company expanded its deal in April, agreeing to deploy 1 gigawatt of the Broadcom chips.
- Anthropic, meanwhile, plans to spend $100 billion scooping up chips from Amazon’s already robust lineup of AI chips (they’re called Trainium). A slew of startups, including Cerebras and Fractile, are also creating AI chips.
Masters of One: Nvidia’s competitors are mostly trying to do one thing, and do it well. That’s unlike Nvidia, whose chips are popular in part because of their broad applications (and top-of-class ability to handle the most compute-intensive tasks). OpenAI said its Jalapeño chip, which was made specifically for the inference work that is OpenAI’s bread and butter, beat Nvidia’s Blackwell in quick and efficient response to prompts. But that doesn’t mean it can perform the same work Nvidia’s chips can. While the new wave of custom chips could put a dent in Nvidia’s domination of the market, they seem unlikely to cause Nvidia to fall far enough in the near term to get hurt … especially when it’s always wearing a leather jacket.
