Cathie Wood finally said something real this time: Jensen Huang isn’t Nvidia’s only engineering genius. She noted that people tend to attribute nearly all of Nvidia’s success to Huang, but the company’s technological moat is also built by a group of people who have spent years working on its underlying architecture, including Chief Scientist Bill Dally, CUDA key figure Ian Buck, and GPU architecture head Jonah Alben. In other words, Nvidia isn’t a “one-man company.” Its engineering organization, depth of talent, and operational capabilities are what make it harder for others to replicate.

It’s interesting to hear her say this. After all, ARK’s models have consistently been quite bullish on NVDA: in ARK’s public framework, demand for AI computing power is assumed to grow rapidly over the long term, and Nvidia is positioned as a key beneficiary, with high assumptions for revenue growth, gross margins, and market share. According to public reports, ARK’s flagship fund made a major reduction in its Nvidia holdings in early 2023, before the stock surged. Nvidia’s share price rose about 239% in 2023, and the move has been repeatedly discussed by the market ever since. That’s also why her current acknowledgment that “once competition picks up, the return expectations in the model have to be revised downward” carries such significance. Put simply: the moat isn’t as wide as people thought.

The competition she’s talking about is real. AMD’s MI300X offers 192GB of HBM3 memory for large-model training and inference. Google has TPUs, Amazon has Trainium and Inferentia, Microsoft has Maia, Meta has MTIA, and Broadcom is developing custom ASICs. Broadcom’s AI-related revenue reached $12.2 billion in fiscal 2024, up about 220% year over year, showing that the path of “big customers developing their own chips plus custom accelerators” is gaining traction. Nvidia reported fiscal 2024 revenue of $60.9 billion, including $47.5 billion from data centers, and a gross margin of 72.7%. Those numbers are certainly impressive, but they also mean that the bigger the base and the more competition there is, the harder it becomes to keep exceeding growth expectations. The CUDA ecosystem, NVLink, and rack-scale systems remain Nvidia’s strengths, but those advantages aren’t limitless. Once customers have a second supplier, assumptions about pricing power and market share will be reexamined.

Personally, I think this is her way of admitting she was wrong, even if she didn’t say so outright. She didn’t say Nvidia was a bad company; she acknowledged that previous models had set return expectations too high and that they would have to be revised downward as competition picks up. That statement is more informative than simply being bullish or bearish. If you disagree, share your reasons.