A quiet day for fresh trade ideas, with Serenity — the stock-picker who builds his portfolio by back-mapping the AI supply chain and buying the small upstream chokepoint suppliers that the hyperscalers and Nvidia quietly depend on — turning his attention to a single macro story rather than any individual name.

The catalyst was a Wall Street Journal report describing how China's Zhipu AI has reportedly matched Anthropic's frontier model on certain benchmarks, including the ability to exploit cybersecurity vulnerabilities. (Anthropic is the US lab behind the Claude family of models; Zhipu is one of China's leading domestic AI developers.) The framing — that China is closing the gap and "resetting" the AI race — is the kind of headline that bears directly on the capital-spending backdrop underpinning Serenity's entire thesis.

His take pushed back against the instinct to fault US policy. Rather than criticizing the administration's export and competitiveness stance, which he judged to be the right call, he laid the blame on Anthropic itself for what he characterized as insufficient safeguards against model distillation — the practice of training a cheaper model by extracting the behavior of a more capable one. He noted that even before the relevant model release there had been chatter that Chinese players had found ways to distill from it, and suggested that handing that capability over via paid API access amounted to giving away hard-won work for relatively modest revenue.

The thread he drew is consistent with how he reads the broader buildout. In his view, the moat in AI rests on two legs: the trillions in capital expenditure required to keep pushing the frontier, and the discipline not to leak that advantage cheaply. That capex leg is precisely what feeds the demand he plays through upstream hardware — the optical-interconnect chokepoints (the indium-phosphide lasers and substrates that supply co-packaged optics), the memory and HBM packaging bottlenecks, and the power and grid suppliers he expects to re-rate as data-center electricity demand climbs. A narrative of intensifying US-China competition, in his framework, is broadly supportive of the capex trajectory rather than a threat to it, because it raises the stakes for staying ahead.

No tickers were named and no positions were discussed today, so this reads as thesis-reinforcement rather than a trading signal. Serenity's point was essentially that the competitive pressure visible in the Zhipu report strengthens, not weakens, the case for the sustained hardware spending his upstream names are positioned to capture — provided the leading labs do a better job protecting what they build.

It was a light session overall, with the commentary confined to the AI race rather than the specific lasers, substrates, memory and power suppliers that usually populate his notes.