NVIDIA might’ve added more than $200 billion to their market cap overnight after their fresh revenue projections, but some in the AI community can’t wrap their heads around the hype.
Soumith Chintala, who co-founded and currently leads the Pytorch project at Meta, has said that he’s puzzled with the latest rise in the price of NVIDIA’s stock. “I’m fairly puzzled by $NVDA skyrocketing. GenAI inference and fine-tuning will significantly outweigh GenAI training in overall compute,” he wrote on Twitter.
“When it comes to inference and fine-tuning, NVIDIA’s advantage in software won’t hold much significance. They will inevitably have to face competition from AMD and custom silicon in terms of performance per dollar. As a result, their high-profit advantage will inevitably diminish,” he added.
“The software work that enables this moat-shedding is
Mosaic ML Foundry, HuggingFace Optimum, OpenAI Triton, PyTorch 2.0,
Microsoft ONNXRT (and maybe eventually Modular_AI). These stacks abstract hardware out without change in functionality or reliability. I believe NVIDIA will enjoy their high-margin advantage for another 12 to 18 months before they are forced to make adjustments,” he continued.
“The assumption is that the inference workloads are going to be homogeneous enough that the software and inference will be successfully commoditized (and centralized to a few software stacks that I listed) without needing a lot of customization or consulting,” he clarified.
“Some people confusing this take with some kind of dunk on NVIDIA. To be clear, NVIDIA will likely dominate the industry for a while — they are the best in the business. I think they have some of the finest talent at every level, all the way up to Jensen. I’ve worked with many hardware companies over the years and its easy to see the talent and direction gap between NVIDIA and the rest. However, when they are priced at a P/E above 200, its worth analyzing the relative magnitude of their moat, and whether that kind of P/E is sustainable or justified. Only time will tell,” he added.
Chintala’s contrarian take quickly went viral, and has been shared more than a hundred times on Twitter. Just yesterday, NVIDIA’s stock had zoomed more than 30 percent after its chips were thought to be powering most of the AI revolution. There’s an AI gold rush that’s happening, and NVIDIA is one the biggest suppliers of shovels, and appears poised to take advantage of the new AI-first paradigm. But like Chintala pointed out, software methods that’ll lead to computationally cheaper inference could limit the size of NVIDIA’s market, and the rise in its stock price might not necessarily hold in the long term.