TSMC and ASML Raised Guidance the Same Week Chip Stocks Blamed Kimi K3 for a Selloff Already Underway

The July 17 chip rout was already loaded before Kimi K3 landed — and the two companies that control advanced-chip capacity both expanded into it.

What Happened

On July 17, according to Bloomberg, chip stocks sold off hard across Asian markets. TSMC fell approximately 7.3% in Taipei, SoftBank dropped roughly 9%, and equipment makers Tokyo Electron and Advantest lost around 8% and 7% respectively. The dominant narrative in financial media pointed to a single trigger: Kimi K3, the newest open-weight model from Chinese startup Moonshot AI, which promised performance close to the top US frontier models at a materially lower cost — roughly $15 per million output tokens against approximately $25 to $30 for the leading US model. The framing was familiar: a cheaper Chinese AI model surfaces, chip demand looks fragile, semiconductor stocks crater.

The problem, as Bloomberg’s own reporting noted, is that the selloff was already underway before Kimi landed. It followed TSMC’s earnings the prior day — July 16 — and reflected weeks of accumulated worries about rising infrastructure costs and fatigue over a years-long AI capex boom. CNBC described the rout as multi-factor, with Kimi’s debut making things worse rather than starting them. The July 17 drop of roughly 7% is a materially different event from the muted earnings-day reaction on July 16; conflating the two misstates both the timeline and the causal chain.

And there is a deeper fear the tape is expressing — one that is real, legitimate, and unresolved: if near-frontier AI capability becomes cheap or openly downloadable, the roughly $700 billion that hyperscalers are committing to AI infrastructure may not pay back at the returns the market has priced in. That is not a question one week of trading settles. It is the central unresolved tension in AI-infrastructure investing, and it deserves to be stated plainly rather than laundered through a single model release.

The Structural Read

Here is the tell: in the very same week that chip stocks were being blamed on a Chinese model release, the two companies that actually control the world’s advanced-chip capacity both raised their guidance and expansion plans. TSMC pushed 2026 capex to $60–64 billion. ASML raised its 2026 revenue outlook to €43–45 billion and announced plans to add roughly 30% more lithography capacity. (See our full ASML/TSMC capacity analysis here.) If a cheaper Chinese model were structurally destroying chip demand, the chokepoints — the foundry and the sole EUV supplier — would be the first to flinch. Instead they expanded into the selloff. That is the counter-evidence, and it is exactly what you would expect if underlying demand is still expanding. (Our TSMC capex-flip analysis here.)

The lens we develop at Business Engineer for this pattern is what we call attribution substitution — and it is laid out in full in the essay Frontier AI and the Kimi Delusion. The framework describes what happens when a market swaps a visible, nameable catalyst — in this case, a Chinese model release — for the actual causal driver: months of accumulated supply-chain, cost, and payback concerns that finally crystallized into a story the tape could tell itself. The visible catalyst is not the causal one; it is the story the rout needed to have. Kimi K3 is a real, large open-weight model and a genuine Chinese competitive milestone (see our Kimi K3 model analysis), but conflating a competitive milestone with a structural threat to AI infrastructure spending misdiagnoses both the causation and the implication.

The bearish inference, even if you grant Kimi as a trigger, also runs backwards on the economics. Cheaper capable models have historically tended to expand the addressable market for compute rather than compress it — lower inference costs pull more use cases into economic viability, which drives more training and more deployment cycles. An open-weight release at the frontier also pressures closed labs to accelerate, not retreat. Single-day stock moves are noisy and multi-factor; the honest read is that the payback-on-$700-billion fear is legitimate and the timing risk is real, but those are structural questions about the AI investment cycle, not verdicts delivered by one model’s pricing sheet.

Three Implications

The Bottom Line

The July 17 chip selloff was a multi-factor event — TSMC earnings, accumulated cost-and-payback fatigue, broader macro pressure — that found a convenient villain in Kimi K3’s debut; Moonshot AI’s open-weight model is a genuine competitive milestone and a real pricing challenge for closed frontier labs, but the structural counter-evidence is that both TSMC and ASML raised guidance and capacity into the exact week the market blamed a Chinese model for gutting chip demand, which is not what you would expect if the foundry cycle were breaking. The deeper fear — that $700 billion in hyperscaler infrastructure commitments may not pay back at priced-in returns as capable AI becomes cheap — is legitimate, unresolved, and worth tracking with discipline; it just is not what the tape was actually measuring on July 17.

Sources: Bloomberg — Chip Stock Selloff Deepens in Asia as TSMC Fails to Impress · CNBC — SoftBank, Chip Stocks: Asia/Wall Street AI Rout · FourWeekMBA — ASML/TSMC Q2 2026 Capacity Expansion · FourWeekMBA — TSMC Q2 2026 Capex Flip · FourWeekMBA — Kimi K3 / Moonshot AI Model Analysis · 91,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.