The AI Complex Faces Another Efficiency Scare
The biggest development today was the accelerating selloff across AI infrastructure beneficiaries, including the miners held within our crypto equities portfolio. While the broader rotation out of AI CapEx receivers has been underway for several weeks, today’s move was likely catalyzed by Chinese AI startup Moonshot releasing its new open-weight model, Kimi K3. Ahead of launch, the model was expected to outperform Claude Opus 4.8. Early benchmark results have been even stronger, with certain rankings placing it ahead of Fable and GPT-5.6.
The market’s concern is not simply that another capable model has emerged, but rather that it appears to have achieved those results with meaningfully greater efficiency. Moonshot claims K3 delivers roughly a 2.5x improvement in scaling efficiency versus its prior generation. If models require less compute to achieve comparable intelligence, investors naturally begin questioning how much incremental spending hyperscalers ultimately need to make on GPUs, networking, power infrastructure, and data centers. That concern was reflected across the entire AI CapEx receiver complex today.
We’ve Seen This Movie Before
That said, I think investors should be careful about extrapolating today’s reaction too far. We’ve seen a remarkably similar setup before.
When DeepSeek released its highly efficient open-weight model in January 2025, AI infrastructure beneficiaries sold off sharply as investors questioned whether improving model efficiency would structurally reduce demand for AI hardware. However, those stocks subsequently staged a violent tactical rebound before eventually rolling over several weeks later. Importantly, I think that later weakness was driven primarily by the broader tariff-induced macro selloff rather than the DeepSeek announcement itself. In other words, the market’s initial reaction to greater AI efficiency ultimately proved much more severe than the underlying fundamentals justified.
My base case remains that improving model efficiency ultimately expands, rather than contracts, aggregate demand for compute. Lower inference costs reduce the cost of intelligence, allowing entirely new applications to become economically viable. As I often describe it, when you expand the number of lanes on a highway, more traffic eventually shows up. While efficiency gains may reduce compute requirements for any individual task, they also broaden adoption and create new sources of demand that can ultimately offset much of those savings.
That is not to say AI infrastructure spending will continue compounding at today’s pace indefinitely. At some point, hyperscaler CapEx growth will almost certainly mature as the market becomes more saturated and returns on incremental investment begin to normalize. The AI CapEx “music” will eventually stop. I simply would not extrapolate today’s news as an indicator of the music stopping.
Positioning Also Suggests the Selloff Is Becoming Exhausted
Beyond the fundamental debate, today’s price action also appears consistent with a broader momentum unwind that has been unfolding across the AI complex for several weeks. The rolling 17-day decline in BlackRock’s Momentum Factor ETF (relative to SPX) has now surpassed the unwind experienced during March 2021, making it one of the sharpest momentum reversals on record. Historically, these unwinds have generally occurred within broader bull markets and have often been followed by constructive forward returns over the subsequent 1-2 month periods. The lone exception was the tariff-driven correction during early 2025, when macro conditions deteriorated materially.
Credit markets provide another reason for cautious optimism. Despite today’s sharp selloff in AI infrastructure equities, neither investment-grade nor high-yield credit spreads widened meaningfully. Because credit markets often deteriorate before equities during periods of genuine financial stress, the relative stability in spreads suggests today’s move may have reflected positioning and multiple compression more than a material reassessment of long-term fundamentals.
A Tactical Bounce Looks Increasingly Plausible
Taken together, I think the ingredients for a tactical rebound are beginning to come into place. The historical precedent following DeepSeek, an increasingly washed-out momentum backdrop, and resilient credit markets all suggest the recent selloff may be becoming overextended.
That said, I am not yet increasing exposure to the miners. While I think the tactical risk-reward has improved meaningfully, I would still prefer to see evidence that selling pressure is beginning to exhaust itself before becoming more aggressive. For now, I think the appropriate takeaway is to expect a bounce rather than assume today’s selloff marks the beginning of a structurally weaker environment for AI infrastructure demand.
Clarity Remains the Primary Crypto Catalyst
Turning back to crypto, President Trump met with several Senators today to discuss the remaining ethics provisions within the Clarity Act, which continue to represent the primary obstacle to passage. Prediction markets responded positively, with implied odds moving from roughly 40% to 45%. While I continue to have reservations about the quality of price discovery in that market, the move is nevertheless directionally encouraging.
My own view remains largely unchanged. I continue to think the probability of passage is slightly better than a coin flip. At this point, the remaining disagreement appears to be political rather than policy-related. If lawmakers can ultimately reach agreement on the ethics language, I think the path toward passage remains intact. President Trump is also scheduled to speak this evening. While the speech is not expected to focus on crypto, it would not surprise me if he referenced today’s meeting or highlighted progress on the legislation, which could provide another incremental catalyst for sentiment.
Bottom Line
Today’s selloff in AI infrastructure appears to have been driven by a familiar concern. Rapidly improving model efficiency will ultimately reduce demand for compute. History suggests investors should be cautious about drawing that conclusion. Following DeepSeek, the market initially reacted in much the same way before AI infrastructure beneficiaries staged a sharp recovery. While I fully expect the AI infrastructure investment cycle to mature eventually, I do not believe today’s announcement marks that inflection point. Instead, I think the combination of an extreme momentum unwind, resilient credit markets, and historical precedent argues for an increasing probability of a tactical bounce.
Within crypto, I continue to view the Clarity Act as the most important near-term catalyst. My base case remains that the odds of legislation passing are slightly better than a coinflip. Taken together, I think that we could see a near-term bounce in miners and the tactical opportunity for crypto is still alive.
Tickers in this video: ^SPX
______________________________
PS: If you are enjoying our service and its evidence-based approach, please leave us a positive 5-star review on Google reviews —> Click here.