Kimi K3 reopened the question embedded in chip valuations: what if better AI needs less compute? The PHLX Semiconductor Index finished the week down 10% and in bear-market territory as Moonshot AI’s release revived the same fault line DeepSeek exposed in 2025: frontier-level capability may not require the compute budget Wall Street has embedded in semiconductor valuations. Moonshot says its 2.8 trillion-parameter open-weight model beat GPT-5.5 and Claude Opus 4.8 across parts of its coding suite, while Arena ranked K3 first in front-end coding. Bank of America estimates K3 costs roughly half as much as OpenAI’s GPT-5.6 Sol, although Moonshot concedes it still trails the strongest proprietary systems overall. THE REAL QUESTION IS COMPUTE INTENSITY The chip bull case assumes three things keep compounding together: model capability, AI usage and infrastructure spending. K3 challenges the third link. A cheaper open model can let enterprises route ordinary workloads away from premium systems, pressure closed-model economics and force hyperscalers to defend the return on every new GPU cluster. That does not mean Nvidia demand suddenly disappears. It means investors may pay a lower multiple for that demand because the long-term amount of compute required per unit of useful AI output is less certain. Markets reprice at the margin, and a small cut to terminal growth can do more damage to an expensive stock than a modest cut to next quarter’s shipments. WHY “DEEPSEEK 2.0” STICKS DeepSeek’s January 2025 shock pushed the semiconductor index down 9.2% and erased $593 billion from Nvidia in one session, not because AI adoption broke, but because the scaling-at-any-cost narrative briefly lost credibility. Kimi K3 hit the same pressure point, and it arrived when the trade was already vulnerable. By Friday, the SOX was down 18.6% over one month despite remaining up 64.8% year to date, which tells me K3 was an accelerant for a crowded valuation reset, not the sole cause. The next test is not another benchmark chart. It is whether megacap tech can show that rising AI capex is producing proportional revenue, utilization and margins rather than simply buying insurance against falling behind. BOTTOM LINE AI can keep winning while chip shareholders lose. Kimi K3 does not need to make GPUs obsolete; it only needs to make Wall Street question how many GPUs the next dollar of AI revenue actually requires.