Alphabet beat everything. It fell 6% anyway

AI & robotics investing — explained so you can actually act on it.

ISSUE 20 // Friday, 24 July 2026 // 6–9 min read

THE HOOK

Three companies reported this week. Alphabet grew 24%, beat every estimate, and lost more than $200 billion of market value in a day. Tesla posted record revenue and record deliveries and fell 14.5%. Intel printed its fastest growth since 2011 and jumped 12% within the hour.

Same market. Same 48 hours. The difference wasn’t performance — it was which side of the compute invoice each company sits on. Two of them buy compute. One of them sells it. That split is the trade for the rest of Q3 — and Tuesday’s Fed meeting decides how violent it gets.

WEEKLY SCOREBOARD

Prices and moves are Thursday’s close and session — the entire week happened on Thursday this week. VIX = the “fear gauge.” Measures expected S&P 500 swings over the next 30 days. Above 25 = real anxiety; 18.7 = paying attention.

Bottom line: Alphabet grew 24% and lost $200 billion in a day; Intel grew 25% and gained 12% in an hour. Growth stopped being the question this week. The invoice is.

TOP HEADLINES

1. Alphabet’s monster quarter met the capex wall.

Revenue $119.8B, up 24%. Google Cloud up 82% at a 35.6% operating margin, on a $514B backlog. TPU chips sold externally recognised as revenue for the first time — over $1 billion. Berkshire Hathaway disclosed a $10 billion stake. And the stock fell 6.2% anyway, because capex guidance went to $195–205B for 2026 — with 2027 set to “increase significantly” — and free cash flow turned negative for the first time in Alphabet’s history as a public company. One line of guidance outweighed the best revenue quarter in five years.

2. Tesla grew 26% and fell 14.5%. That’s not a contradiction.

Record revenue ($28.24B), record deliveries (480,126, up 25%) — and adjusted EPS of $0.33 against $0.50 expected, a 1.4% operating margin, negative free cash flow, and a confirmed $25B capex bill for robotaxi and Optimus. The robot itself slipped: Optimus Gen 3 has roughly 10,000 unique parts, production will be “extremely slow at first,” and high-volume output moved to 2027. What it means: the physical-AI complex trades on Optimus timelines, and the timeline just moved again. The validation event Signals has flagged since the spring is now a 2027 story.

3. Intel is what the market wants AI to look like.

Flip the implication first: sellers of compute are supply-constrained price-setters right now, and Intel just proved it. Q2 revenue $16.1B, up 25% — the fastest since 2011. Data centre and AI up 59%; the company says it cannot keep up with orders. Foundry up 31% to $5.8B, 18A yields at 85%, gross margin recovered to 40.4% from 27.5% a year ago, ten long-term server CPU contracts signed, Q3 guidance above consensus. Stock +12% after hours — in a July where INTC had fallen 28%. Earlier in the week Fortinet became the foundry’s first named customer under Lip-Bu Tan, and The Information reported a Google order for 3 million TPUs.

4. China’s open-weight one-two punch.

Moonshot’s Kimi K3 — a 2.8-trillion-parameter open-weight model competitive with US frontier systems — is what tipped the chip index into a bear market last Friday. This week added two twists: a US official said Moonshot trained on Nvidia silicon despite the export ban, and a leaked investor call attributed to DeepSeek’s Liang Wenfeng claims roughly 16,000 Huawei chips and a working path off Nvidia’s CUDA software. Two reports, one week, and they can’t both be true: either the export regime leaks, or the Chinese stack genuinely doesn’t need Nvidia. Decide which you believe before the next leg down.

5. The macro turned hostile to long-duration AI bets.

Brent closed above $100 for the first time since May — fifth straight up session — after Houthi attacks on two Saudi tankers and Kazakhstan suspending pipeline exports. The 10-year yield hit its highest level in about eighteen months. Jobless claims came in at 187,000, the lowest reading since 1969. Fed funds futures now price an 82% chance of a September hike, up from 52% a week ago, with July odds at 38% into the 28–29 July FOMC. What it means: every “spend $200B now, monetise in 2028” story just got a higher discount rate. The market didn’t change its mind about AI this week. It changed its price of time.

OUR INVESTING ANGLE

Everyone’s watching the AI trade for signs of a bubble popping. The smarter watch is which direction the cash is flowing.

The thesis: the AI trade is inverting. Capital is rotating from the buyers of compute to the sellers of compute — and rising rates are the accelerant. This is not a demand problem. Intel literally cannot fill its order book (headline 3), and NVDA finished the week up 2.9% through Thursday’s wreckage. What changed is the cost of waiting: with the 10-year at an eighteen-month high and hike odds at 82% (headline 5), a dollar of profit in 2028 is worth meaningfully less than it was a month ago. Alphabet (headline 1) and Tesla (headline 2) both asked the market to fund that wait. The market said no.

So yeah. One side’s anxiety is the other side’s income statement. But here’s the second-order move nobody priced yet: if this punishment persists, boards will start splitting the capex story — expect more Alphabet-style moves to monetise infrastructure externally (TPU sales, foundry deals, compute leasing) so the spending line can be reframed as a product line. The first hyperscaler to convincingly do that gets its multiple back. Watch the language on next week’s calls for it.

Who gets hurt? Named, not vibes. Tesla — negative FCF funding two moonshots that both slipped. Meta — the biggest compute buyer with no cloud business to monetise it; it reports next week and fell 3.4% Thursday in sympathy. Oracle and CoreWeave — the debt-funded compute renters from Issue 09’s thesis; higher rates hit levered buyers twice. Alphabet sits in the middle — it’s the biggest spender, but with TPU revenue crossing $1 billion externally it’s also becoming a seller. That tension is the whole GOOGL debate now, and it’s Idea 2 below.

→ Read the deep dive: Who pays for the AI buildout — and who gets paid

THREE IDEAS TO RESEARCH THIS WEEKEND

Not recommendations — starting points for your own research. One primary, one contrarian, one watchlist hold.

Idea 1 — Intel (INTC): the primary research target this weekend.

I did not expect to write this name in 2026. But Thursday made it the cleanest expression of the thesis: the only large cap that is simultaneously supply-constrained, margin-expanding (27.5% → 40.4% gross in a year), and freshly re-priced — +12% after hours into a stock that had dropped 28% in July.

The risk: a lot is already paid for — up roughly 170% this year after 84% last year — and the GAAP loss ($12.5B CHIPS-escrow charge) shows the accounting under the comeback isn’t clean. Fortinet, the first named foundry customer, is trailing-edge.

Tripwire: a named top-five hyperscaler on 18A by the Q3 print (late October) = the foundry story is real. Another quarter of “discussions continue” = you own a products cycle at a foundry multiple.

How to research: ticker INTC. Last night’s release plus the 18A yield commentary; cross-check the reported Google TPU foundry order.

Idea 2 — Alphabet (GOOGL): the tripwire tripped. The stock fell anyway.

In Issue 09, the GOOGL tripwire was TPU revenue breaking out as its own line by Q2 earnings. It did — over $1 billion, recognised for the first time. The stock fell 6.2% the same day. Both things are true, and that’s exactly why this is the contrarian idea.

Why now: you’re being offered Cloud growing 82% at a 35.6% margin, a $514B backlog, and Warren Buffett arriving with $10 billion — at mid-April prices, because of a capex guidance line.

The case: Alphabet is the one company migrating from buyer to seller of compute. If TPU external revenue compounds, the capex everyone fears becomes the moat everyone missed.

The risk: FCF just went negative for the first time ever, and 2027 capex is guided “significantly” higher. If Cloud margin bends while the spending rises, there is no valuation net under the story.

Tripwire: Q3 print (late October): Cloud operating margin at or above 33% while capex runs $50B+ a quarter = the spend funds itself, and the fear was the entry. Margin below 30% = the bears were early, not wrong.

How to research: ticker GOOGL. Q2 release plus the capex commentary; watch for Berkshire’s position in the next 13F round.

Idea 3 — Japanese industrial robotics: still a watchlist hold. This week is why.

Honestly? Third issue running for this one, and I’m still not pulling the trigger — but Tesla just made the argument for me. Optimus volume slipped to 2027 with 10,000 unique parts and “extremely slow” initial output; Thursday’s 14.5% drop is the cost of paying humanoid-demo multiples. Meanwhile FANUC (6954.T), Keyence (6861.T) and Yaskawa ship profitable robots today at roughly half the multiple, with BOTZ as the wrapper. The obvious caveat: if the whole physical-AI layer de-rates on Tesla’s miss, Japan falls with it — cheap is not the same as immune. Issue 09’s trigger was serialised Optimus units by end-August; Tesla has now told you that won’t happen fast. So flip it: August passes with no serialised units, and this graduates from watchlist to primary. That’s the whole idea. No sub-headings required.

AI INVESTMENT FRAMEWORK

Living portfolio framework by layer. Not financial advice — research starting points only.

Per-layer notes

Infrastructure — the sellers of compute. INTC added this week (see changes); NVDA +2.9% on the week through a 2.2% Nasdaq drop; Intel says it cannot meet demand; Micron near $1,000 with Morgan Stanley forecasting memory prices up 25%. Capacity is sold out and the buyers just told you they’re spending more.

Platforms — the repriced spenders. GOOGL fell 6.2% on a beat; MSFT −2.2% in sympathy before it even reports. The layer’s problem isn’t revenue — it’s that capex guidance is now a liability until someone proves the return maths. Conviction cut.

Applications — quietly improving. ServiceNow raised its subscription forecast for the second time and money is rotating toward companies with earnings today rather than 2028. Not enough for a conviction change — yet.

Physical AI — the validation event moved. Optimus volume production is now 2027; robotaxi is live in seven metros but the market wanted robots, not promises. TSLA at its lowest close since August 2025. Japan (Idea 3) is the hedge inside the layer.

Cybersecurity — no news is no change. The layer did nothing notable this week; that itself is information after a quarter of AI-security noise. Holding MEDIUM.

Global — the live bear case for the Nvidia premium. Kimi K3 open-weight, DeepSeek claiming 16,000 Huawei chips and a CUDA exit — but also a US official saying Moonshot trained on smuggled Nvidia silicon. Five risk dots earned.

What we’re watching (next 2 weeks)

Changes this week

● INTC added to Infrastructure (replacing AVGO in the headline three — the custom-ASIC story is intact, but the fastest incremental compute dollar this quarter is Intel’s data centre + foundry ramp, and AVGO has lagged the layer all year).

● Platforms cut from HIGH to MEDIUM ↓ — a 6% drop on a beat quarter is the market telling you what it thinks of unbounded capex guidance. The layer stays investable; the conviction doesn’t survive “significantly higher in 2027” unexamined.

● Physical AI signal flipped to DEVELOPING ↓ — Optimus volume production moved to 2027. Conviction level unchanged; direction isn’t.

Disclaimer: This newsletter is for informational and educational purposes only and does not constitute financial advice. iPrompt Signals is not a registered investment advisor. Always conduct your own research and consult a qualified financial professional before making investment decisions.

YOUR MOVE

The week’s three takeaways, traceable to the headlines and ideas above:

1. Next week is the framework’s first real test, not its victory lap. Meta reports with no cloud to monetise its spending, and the FOMC meets Tuesday–Wednesday into $100 oil and 82% September hike odds. Those two events either entrench Thursday’s rotation or unwind it — apply the lens before the prints, not after.

2. Two of Issue 09’s tripwires resolved this week — against the stocks. TPU revenue broke out as its own line (bullish trigger) and GOOGL fell anyway; Optimus slipped past its window and the humanoid premium finally paid for it.

3. Watch for the capex-reframing play. The first hyperscaler that convincingly turns its spending line into a product line — external chip sales, compute leasing — gets its multiple back. The language on next week’s calls will tell you who’s trying.

Now run the ledger yourself. Take one company you actually own — not this issue’s three — and answer two questions: which side of the compute invoice is it on, and what single observable event would flip it? Reply with your one-line entry — company, side, tripwire. I’ll feature the three sharpest in next Friday’s issue, credited or anonymous, your call. The replies are the part of this job that actually keeps me curious.

Stay curious — and stay qualified.

— R. Lauritsen

Editor, iPrompt Signals

Know someone building an AI position? Forward this — they’ll thank you by Friday.

P.S. — Two tripwires from Issue 09 fired this week, and both fired “correctly” while the stocks did the opposite of what the triggers implied. That’s not the tripwires failing — that’s what they’re for. They tell you when the facts changed. The market tells you when it agrees. The gap between those two dates is where the returns live, in both directions.

QUICK GLOSSARY

18A — Intel’s next manufacturing process (“1.8-nanometre class”). Yields — the share of chips that come out working — hit 85%, the level where external customers can commit.

Bear market — A drop of 20% or more from a recent high. The chip index entered one this month, from its late-June record.

Capex — Capital expenditure — money spent on long-lived assets like data centres. Alphabet will spend $195–205B this year; the cost hits profits over years via depreciation, but the cash leaves now.

CUDA — Nvidia’s software layer that AI code is written for. The real moat — chips can be copied, a 15-year software ecosystem can’t. DeepSeek claims it’s building an exit.

Discount rate — The maths that converts future profit into today’s value. Higher rates shrink the value of profits arriving in 2028 — which is why rate-hike odds hit AI spenders hardest.

FCF — Free cash flow — cash generated minus everything spent, capex included. Alphabet’s went negative for the first time in its public history this quarter.

FOMC — The Federal Reserve committee that sets US interest rates. Meets 28–29 July; futures price 82% odds of a September hike.

Foundry — A factory business that manufactures chips designed by others. TSMC is the giant; Intel is trying to become the Western alternative.

Open-weight model — An AI model whose trained parameters are published for anyone to run — like Kimi K3. Undercuts the pricing of closed rivals.

iPrompt Signals

Published Fridays by FrontWave Media Ltd · Limassol, Cyprus