Research

Three years and ~¥31.8bn of losses, then a first-ever profit in 2025 and an H1-2026 guidance (¥50–57bn net) that would erase all of it — a company the US blacklisted in Dec 2024, no controlling shareholder, pricing the biggest A-share IPO of the year (~¥29.5bn).

Google pays Apple ~$20B a year for its front door; Apple pays Google ~$1B a year for Siri's new brain. Capex of $13B against the hyperscalers' $660B+ — fifty to one — yet the group's slowest grower wears 37.3x trailing, its richest sober multiple. What the certainty premium actually buys, and five dated events inside nine months that could reopen it.

BofA's sum-of-the-parts says autos are 12% of the price; robotaxi, Optimus and FSD promises are the other 88%. 42 driverless cars against Waymo's 500K paid rides a week, a $1T comp package that doubles as management's own probability map, and ~196x forward earnings — the control case of the receipts economy.

From 125 won and a $7B bailout to a $1.12T market cap and a 72% operating margin — six days before it prices the largest ADR offering ever ($29B, Nasdaq, July 10). Two-thirds of NVIDIA's HBM4 runs on its memory. Trailing 23x, forward 5.5x: the market is paying trillion-dollar prices while pricing the earnings to die.

It invented the Transformer and watched every author walk out to build its competitors. Then: a $462B backlog (roughly 40% from Anthropic), TPUs selling as merchant silicon, Apple paying ~$1B a year to put Gemini inside Siri — while 68% of searches end without a click. The comeback is real, and at 27.5x/24.7x it is fully narrated.

In 1980 it licensed DOS to IBM non-exclusively and inverted the giant. In April 2026 OpenAI deleted the AGI clause, capped the revenue share and went non-exclusive — the same clause, pointed the other way. RPO of $627B (~45% OpenAI), Copilot's 20M seats against an 11.5% usage share, and a ~$135B stake carried nowhere on the books.

February: a $200B capex guide, stock -11%. April: the same plan with receipts attached, +27% in a week — the market issued a permission slip. AWS +28%, a $364B backlog before the new $100B Anthropic deal, >$225B of Trainium commitments, trailing FCF of $1.2B, and shareholder returns of exactly zero — by design.

Record single-day value swings in both directions within months, a buyback that went to exactly zero — the tell — and $125B of 2026 capex flowing into Meta Compute. At 21.2x trailing / 15.8x forward, the cheapest member of the group is making its most concentrated bet, funded with what used to be the shareholders' money.

A $4B castoff in April 2025; a $258B giant fourteen months later, gross margin 22.5% → 78.4% in five quarters. Then −26% in a week on zero company news. Structural re-rating of NAND, or the most synchronized cycle top memory has ever printed?

A $99.4B backlog against a $46.7B market cap — and a 13% one-day drop because Meta, its $21B customer, may start selling its own surplus compute. Contracted infrastructure compounder, or a leveraged spread trade on a commodity that deflates 30% a year?

Forty years a brutal commodity, near death more than once — then AI made its dullest product the scarcest thing in computing. An 84% gross margin against a 7.6× forward P/E: a generational bargain, or the most spectacular cycle top ever?

A 2020 brush with bankruptcy, a $9.7B chemical-arm sale to Berkshire in 2026, and a moonshot to pull carbon from the sky. Why Buffett owns a quarter of it — and why that isn't the thesis.

Three decades, three bets, and a moat that isn't the chip. A panoramic read on the world's most valuable company — and why it trades at ~15× forward.

Three rockets that blew up, a near-monopoly on space, and a $2T bet on AI compute in orbit. The most legendary company of its era — at 100× sales.