**TL;DR:** GOOGL lost $200B in about 30 hours because a model is late and some engineers changed badges. This started the morning after Buffett told CNBC the $31B stake was his idea, including a $10B check written straight to Google funding the exact capex everyone is scared of. META meanwhile gained $125B this month on a leaked memo about maybe renting out spare GPUs someday. Google's actual cloud did $20B last quarter, up 63%, record margins, $462B backlog, and they're turning away customers because they can't build compute fast enough. Earnings Wednesday, 25x forward, expectations conveniently reset. Sold the machine, bought the memo.
July 15, Buffett goes on CNBC and says the thing nobody expected. The $31 billion Alphabet position everyone assumed was Greg Abel flexing his new job? "I initiated it." Said not buying sooner was a mistake, said Google is more likely to succeed than 90-95% of Wall Street's picks. Stock ran from the 358 area to close near 372 and kept climbing the next morning, touching 380.
Then Bloomberg landed midsession on the 16th. Gemini 3.5 Pro months behind schedule, coding benchmarks came in soft. The stock collapsed the same day, $200 billion gone, and kept bleeding through the 17th down to a 344 low. From 380 to 344 in roughly 30 hours of trading. Revenue estimates moved zero dollars.
The delay stacked on a month of exodus headlines. Shazeer back to OpenAI. John Jumper, the Nobel guy, off to Anthropic with two more researchers behind him. Except 3.5 Pro finished its training runs before any of them cleaned out their desks. Whatever those departures cost shows up in 2027's models. The market sold this quarter on next year's org chart.
Two weeks earlier Meta showed what headlines are worth when they point the other way. Stock jumps 9%, $125 billion in a session, on a report with unnamed sources and nothing announced that they might build a cloud business to sell excess compute. Zuck's own framing of the opportunity: we may have overbuilt. A hundred twenty five billion dollars for "we bought too many GPUs and might rent out the extras."
Google can't relate. Their cloud did $20 billion last quarter, up 63%, fastest of any major provider, operating margin at a record 32.9%, backlog nearly doubled to $462 billion with over half converting to revenue inside 24 months. And they're turning deals away. Not losing them, declining them, because they can't build capacity fast enough. Monday's bounce was a report that they're etching Gemini directly into next-gen silicon for 6-10x better power per token, built specifically to fix the shortage that's forcing them to say no to money. They're also selling TPUs to third parties now, with multiple gigawatts committed to Anthropic and Broadcom. Meta wants to enter the cloud business someday. Google is rationing theirs.
The delayed model canceled none of that backlog. Not one contract.
The leaderboard fear has the shelf life of milk anyway. Chinese labs keep compressing the frontier within weeks through distillation, Kimi and friends proved model rank doesn't hold. The durable war is cost per token and distribution, and Google owns every layer: silicon, data centers, models, three billion users. The AI subscription revenue people wring their hands over is a rounding error next to cloud and ads.
The capex is the real bear case and it deserves respect. $180-190 billion guided this year. Q1 free cash flow down 47%. Buybacks paused. Anthropic concentration in the backlog. If cloud growth prints below roughly 55% Wednesday the conversion story cracks and this whole thing gets rethought. Watch that one number.
That June tranche of Buffett's stake, though. It wasn't open market buying. It was a $10 billion private placement, shares purchased directly from Alphabet, cash going straight into the AI buildout. The spending that terrifies the market has Omaha's signature on the check. The man does not fund bonfires.
One more wrinkle in Wednesday's print. Alphabet marks its SpaceX and Anthropic stakes to market. SpaceX went public in June, repricing that stake off its old private mark. Anthropic closed a $65 billion round in late May at a $965 billion valuation, up from $380 billion in March. Both revaluations land inside Q2. Last quarter's paper gains already made the headline EPS look absurd. This quarter's paper is bigger, on both barrels, and the algos reading the wire at 4:01 don't read footnotes.
Stock's at 351 today, closer to last week's low than its high, coiled at the same level it sat before the Q1 print. That one went 350 to 408 on the report before round tripping the whole move. Same coil. Bigger catalyst.
Badge counts and memos got traded all month. Wednesday the machine reports.
At the end of the day: calls.
Not financial advice. The author is highly regarded and drives his own short bus.
Position: GOOGL 380C 7/24, many contracts.