
Rare earths in the age of AI
Until about 2030, AI uses only ~5% of rare-earth magnets. These stocks move on China’s export rules, not on AI news.
China’s export-rule pause ends
Market Storm · all 13 reports
Not stock tips — a look at where AI is really going by following its money. Each report is produced by STORM, a multi-agent AI research method, pointed at a real market catalyst: earnings, a major deal, an industry move. The finance is the payload; the method is the point.
3 reports

Until about 2030, AI uses only ~5% of rare-earth magnets. These stocks move on China’s export rules, not on AI news.
China’s export-rule pause ends

ARKG’s 96% rise, an 8.8% fall, and a month of new AI biology tools
ARKG holds about 33 gene and biology companies. Scored on AI use today and on gain if AI takes off, the two often disagree, and the best positioned own what AI can’t make: patient data, lab work and proven medicines.
ARK Genomic Revolution ETF


Rebuilt after NVIDIA’s latest quarterly report — checked against company filings through 31 August 2026
The gap is not an accounting quirk. NVIDIA now waits 60 days to be paid instead of 45, and for the first time it told investors in writing that some big buyers can take up to a year. It has also promised to cover $108.5 billion of somebody else’s rent, and set aside nothing against it. Forty-five claims went to agents told to disprove them, checked against 94 filings. Two came back wrong. The rest say the same thing: the company at the centre of the AI build has started financing its own customers.
14 agents · all 45 claims refuted-tested
Fundamentals analyst · Short-seller · Industry engineer · Valuation watcher — and a fifth on the thesis pieces. They interview each other grounded in live web search, then a separate pass tries to refute the load-bearing claims against primary sources. What survives is written up with the caveats it earned — and every report shows its own roster and how deep the refutation went.
10 reports
Each of these reads a single company’s filing. Four agents rather than five, and the top load-bearing claims go to the refutation pass rather than all of them — every line says which, and every report names its own roster.
Earlier reports, each at its own link. I don’t update them.
13 agents · 8 of 32 claims refuted-tested · caveated
12 agents · all 22 claims refuted-tested
12 agents · 6 of 6 claims refuted-tested
12 agents · 0 of 10 claims refuted-tested · caveated
12 agents · 6 of 12 claims refuted-tested
12 agents · 5 of 11 claims refuted-tested · caveated
12 agents · 6 of 13 claims refuted-tested
12 agents · 6 of 13 claims refuted-tested
9 claims tracked · caveated