Follow the AI story.

Listed for reference, not as recommendations, and not investment advice. Membership is a judgement call, not a definition — there is no official “AI sector”, and most of these companies earn plenty of revenue that has nothing to do with AI. Prices and returns are in US dollars for every row, including the Korean, Japanese, Taiwanese and European listings: each daily close is converted at that day’s rate before anything is derived from it, so a return here is what a dollar investor actually earned, currency move included. An arrow marks the currency a row trades in.

NVIDIA Data Center and Google Cloud are reported segments, though neither is AI-only. TSMC’s high-teens share is management’s company-defined accelerator revenue mix — GPUs, ASICs and HBM controllers, excluding CPUs — not a standalone segment. Amazon’s run rate is call commentary; Anthropic and OpenAI are private. Run rate annualises a recent period — it is not revenue earned. Microsoft stopped disclosing an AI-specific figure after FY26 Q2, which is why it is absent.

The backlog total combines management-reported Google Cloud and AWS commitments. It is not AI-only, and the two companies’ definitions and recognition timing differ. The capex total is also company-wide rather than a standalone AI budget.

SK hynix’s card pairs company-wide quarterly revenue and operating margin with management’s HBM4 shipment commentary; the margin is not an HBM-only measure. TSMC’s monthly growth is also company-wide and is shown only as a high-frequency foundry-demand proxy, not an AI revenue breakout. SK hynix and ASML monetary figures are translated to USD using the official report-date Federal Reserve and ECB reference rates and rounded to the nearest $0.1B.

This tracks employer attribution, not independently measured displacement. Challenger classifies the reason cited in announced US cuts; multiple factors may shape a restructuring, while unannounced hiring freezes and role changes are outside the count. Each monthly percentage uses that month’s announced cuts as its denominator. Source: Challenger, Gray & Christmas.

AI ranked fourth among stated reasons in August, while remaining the leading reason year to date. Total announced cuts rose 58% from July to 52,881. Challenger tracks employer attribution, not independently measured displacement. Its technology-sector total covers cuts for every stated reason.

Q1’s ≈80% share and total preserve Crunchbase’s Mar 31 snapshot; Q2’s >70% share and total preserve its Jul 1 snapshot. The later revised H1 total is not backfilled into the quarter view because Crunchbase did not publish a revised split. Deal amounts are announced or committed, not necessarily funded on the announcement date. Crusoe’s amount is the anticipated full round at its initial close. August and September deals are excluded from the H1 concentration calculation. A post-money valuation is the negotiated price for preferred stock with its own terms, not a daily market capitalisation. This is the reported headline set, not a complete private-market league table.

Combining the three H1 OpenAI and Anthropic rounds in the deal table with Crunchbase’s later revised H1 total gives about 42%. That is a derived, cross-vintage comparison; the Q2 share is a separate July 1 snapshot, and funding totals can be revised as later deals are disclosed.