Victor
Muraviev

I published my DeepSeek R1 analysis on 21 January 2025, six days before the market sell-off. My commentary subsequently received over 1M combined views on X. I anticipated the decline in prices across compute-exposed stocks, whilst acknowledging long-term bullishness due to factors such as Jevons paradox.

My release coverage addressed R1’s multi-stage RL/SFT pipeline, emergent reasoning behaviours and API pricing 20–30× below o1 at comparable performance. In discussion with Garry Tan on 25 January, I argued for long-term acceleration whilst also anticipating short-term crashes, pivots and fears of resource allocation changes regarding compute. That evening, I was setting new SLs on $NBIS, extending the downside view to $NVDA and the wider market despite the likelihood of a Jevons outcome.

Replying to my 25 January warning, Eddie Jones (fmr Bridgewater, running own book at the time) argued that R1 was not leading-edge and that compute requirements were increasing exponentially. Although I was of a similar opinion (as someone bullish on compute, AGI and further, since at least 2018), I had gauged market sentiment and my world model of market actors was indicating for me to anticipate negative price action from investors acting on inferior information or selling in anticipation of others doing so, whilst remaining bullish on the underlying long-term demand; “What we think ≠ what the rest of the market thinks.” When prices fell overnight on 27 January, I returned to the discussion with “Look at the overnight price action. Told you so…” and, replying to Eddie again, “I’m still long. I’m just accurately predicting the price action.”

Two days before industry leaders such as Satya Nadella (Microsoft chairman and CEO) publicly invoked Jevons paradox to explain why greater AI efficiency and accessibility would increase usage, I had already been applying that thesis around DeepSeek and correctly anticipated the short-term crash despite the long-term bullishness. As one of the earliest public NBIS investors and commentators, already heavily long since 2024, I had covered R1 six days before the sell-off and anticipated the adverse repricing of the same infrastructure on which I remained bullish long term. I published that assessment on 25 January; by the time of Nadella’s 27 January post, the overnight sell-off I had predicted was well underway. In my response to him, I linked the earlier warning. Before the US open (amidst market panic), I was already considering the extent of re-entry: “Waiting for reaction to open to see how much to re-enter with.”

Nvidia closed at $142.62 on Friday 24 January and $118.42 on Monday 27 January: a 16.97% decline. It closed at $128.99 on Tuesday, an 8.93% rebound from Monday. NBIS closed at $41.69 on 24 January and $26.08 on 27 January: a 37.44% decline. It closed at $30.09 on 28 January, a 15.38% rebound from Monday.

Unlike most of the bears and the bulls, I correctly identified that the trade would still be bullish, but that the market would likely react negatively because of inferior information being too prevalent amongst buyers and sellers. I then anticipated and predicted that negativity, re-entered at a lower price, and proceeded to benefit from massive increases in the price of Nebius Stock from that point on, which had 10X’d since the DeepSeek low.

USD closing prices; percentage changes calculated close to close. These are stock-price moves, separate from the portfolio and option returns elsewhere on this page.