Key Highlights

  • Echoes of the DeepSeek R1 Selloff: Kimi 3's release triggered a selloff compared to DeepSeek R1 last year, when the Nasdaq fell roughly 4% and NVIDIA (NASDAQ: NVDA) dropped nearly 10%, before a subsequent 45% tech rally.
  • Distillation Explains the Efficiency Claim: The efficiency gains behind models like Kimi 3 are attributed to distilling American frontier models rather than a genuine lower-cost breakthrough in model training.
  • Selective Valuation Debate: While some AI and semiconductor names are widely viewed as overvalued, commentary points to NVIDIA (NASDAQ: NVDA) around 20x earnings, CoreWeave (NASDAQ: CRWV), and Nebius (NASDAQ: NBIS) as comparatively reasonably priced, though this view is not universally shared.

The release of Kimi 3, a new large language model out of China, has triggered a now-familiar reflex across markets: a sharp bout of selling in semiconductor and AI infrastructure names on fears that a cheaper, more efficient Chinese model undermines the entire American AI capital expenditure story. According to market commentary circulating since the release, this reaction is largely a repeat of a pattern the market has already lived through once before, and one that did not hold up over time.

The comparison being drawn is to DeepSeek R1's release last year, when a similar wave of panic sent the Nasdaq down roughly 4% in a single session while NVIDIA (NASDAQ: NVDA) shares fell nearly 10%. Commentary now circulating among some market participants argues that Kimi 3 does not fundamentally change the AI capex outlook any more than DeepSeek did, drawing a parallel to a prior episode that did not result in a lasting repricing of the sector.

A Rerun of the DeepSeek Playbook

The core of the bullish counter-argument rests on what actually happened after DeepSeek R1 rattled markets. Despite the initial shock and the sharp declines in the Nasdaq and in NVIDIA specifically, capital expenditure estimates from the major American hyperscalers kept climbing in the months that followed rather than contracting. Technology stocks recovered from the DeepSeek-driven drawdown and went on to deliver a rally of approximately 45% from those lows, according to the commentary, a recovery that undercuts the idea that a cheaper foreign model release represents a lasting threat to the American AI infrastructure buildout.

Whether Kimi 3 follows a similar trajectory remains an open question. If the pattern repeats, the current selloff in AI-linked equities would represent a sentiment-driven overreaction rather than a genuine repricing of the industry's long-term capital spending trajectory. But that outcome is not guaranteed, and the key question for investors is whether this time is meaningfully different from the DeepSeek episode, or whether the same underlying dynamics that drove the eventual recovery are still in place.

Distillation, Not a Capex Breakthrough, Explains the Efficiency Gap

Central to this argument is a specific claim about how Chinese labs achieve their reported training efficiencies in the first place. The contention is that if American labs had not made the enormous capital investments they have in frontier model development, Kimi 3 would not have been possible at all, because distilling knowledge from already-trained American frontier models is how Chinese developers achieve most of their reported training efficiency gains.

Under this framing, the narrative that Chinese researchers have discovered some fundamentally different, lower-cost method of building frontier-level models that American labs are simply unable to replicate is misleading. Instead, the efficiency gains reported by labs behind models like Kimi 3 and DeepSeek are seen as a byproduct of building on top of the very capital-intensive American research those investments funded, rather than evidence that the underlying capex-heavy approach to frontier model development is obsolete. This is offered as the explanation for why capital expenditure estimates kept rising even after DeepSeek's release: the market ultimately recognized that the underlying capex-driven model development approach in the US remained necessary and unthreatened.

Overvaluation Exists, But Views on Specific Names Diverge

The commentary circulating around Kimi 3 is careful to draw a distinction between broad skepticism toward the AI trade and a more selective view of where genuine overvaluation exists. Some companies across the semiconductor and broader AI ecosystem are described as extremely overvalued, and this analysis does not dispute that characterization for parts of the market.

Within that same commentary, a further distinction is drawn between speculative, richly valued corners of the AI trade and core infrastructure names such as NVIDIA, described in the commentary as trading around 20 times earnings, and neocloud providers including CoreWeave (NASDAQ: CRWV) and Nebius (NASDAQ: NBIS), which the commentary characterizes as not particularly expensive at current levels relative to their exposure to hyperscaler capital spending.

Not all market participants share this reading. Skeptics of the AI infrastructure trade point to the same underlying risks the selloff reflects: elevated capital intensity, uncertain returns on AI capex relative to hyperscaler spending guidance, and the possibility that efficiency gains of the kind claimed for Kimi 3, regardless of their true source, could eventually reduce the volume of compute purchases needed per unit of AI capability delivered. Whether the DeepSeek-era recovery pattern repeats this time, or whether Kimi 3 marks a genuine inflection point in the capex narrative, remains an open question that will likely only be resolved by subsequent hyperscaler earnings and capital expenditure guidance in the coming quarters.

Investors evaluating NVIDIA, CoreWeave, Nebius, or any other AI infrastructure name in the wake of the Kimi 3 selloff will need to weigh these competing interpretations against their own assessment of valuation, capex durability, and the underlying source of Chinese model efficiency gains, rather than relying on the outcome of a single prior episode as a guide to how this one will unfold.

Disclaimer: This article is for informational purposes only and does not constitute financial, investment, or trading advice. Readers should conduct their own research or consult a licensed financial advisor before making investment decisions. Past performance is not indicative of future results.