Hello everyone, and welcome back to our weekly positioning update.

Right now, everything seems to revolve around tech sentiment and the direction of U.S. equities. Naturally, this leads us to discuss the DeepSeek model and its implications for NVIDIA, which, as of writing, makes up 12% of the Nasdaq. You can’t analyze broad equities without forming a view on DeepSeek and NVIDIA.

While the initial narrative around DeepSeek and NVIDIA looked incredibly bearish for U.S. tech, sentiment has shifted over the past 24 hours following OpenAI’s claims that it has evidence DeepSeek used outputs from the O1 model to train its own. This “distillation” raises doubts about some of the low training cost claims—training O1 is a vastly different challenge than training a model that piggybacks off it. It also brings into question the natural limits of R1’s methodology: if any advancement depends on improving a “parent” model, we are essentially back to compute-intensive training, where scale and resources matter.

The key takeaway: on Monday, the market narrative was that DeepSeek had innovated independently and massively outperformed OpenAI in terms of efficiency. However, these latest developments—along with the research paper (which discusses distillation in a broader sense)—suggest that models like O1 remain essential. The logic follows: if you’re saving 95% on costs, it’s likely because you’re using a distilled version of a premium model.

Markets, however, aren’t buying into the bearish tech story at all. On Monday, QQQ recorded its largest inflow in history—clear evidence that tech sentiment remains strong. Inflows into tech ETFs continue to outpace those of other sectors, reinforcing the impression that U.S. equities and tech are becoming “too big to fail.” Notably, we’ve seen no inflows into Chinese equities in response to the DeepSeek story, which is surprising if the story were truly market-moving. However, this could be due to the Asian holidays.

Chart 1.a: QQQ recorded its largest inflow in history Monday