The Answer of the Other Four

Two days ago, before Alphabet's earnings, I made a prediction by synthesizing several pieces of data. The 100-year bond and the preemptive equity offering, the raised guidance, the CFO's earnings-call remark — the public data pointed in one direction. On the 22nd, Alphabet raised capex, the stock did not react as strength, and the core business ex-equity-gains fell below expectations. The prediction was right. More precisely, the direction in which the data was read was right.

Now the other four come. Microsoft on the 28th, Meta on the 29th, Amazon on the 30th. Within three days, the rest of the hyperscalers report. By the same method — not a hunch, but the data these companies already left on the record last quarter — I read the direction.

First, one number. These four companies' combined 2026 AI capex is $725–785 billion, up about 77% year over year. And the share of that spend in the group's operating cash flow rose from 33% in 2023 to about 93% this year. Most of the cash they earn goes into building. Here is how that structure shows up in each of the four.

Microsoft — the 28th

Microsoft has already said it. Across recent earnings calls, CFO Amy Hood put it this way — "We know we're behind. We do need to spend." And she stated that "total spend will increase sequentially, and we now expect the FY26 growth rate to be higher than FY25." In the most recent quarter (April), she flagged next-quarter capex above $40 billion. CEO Nadella said he would raise AI capacity by more than 80% and roughly double the data center footprint within two years, and the remaining performance obligations (RPO) rose 51% to $392 billion.

The direction reads like this. Microsoft is likely to raise, or at least hold and accelerate, capex — because it has already said so. And the burden of that spend is already showing: last quarter's gross margin was 67.6%, the narrowest since 2022, which the company attributed to rising depreciation. Over the past four quarters it spent $97 billion on infrastructure to earn $37 billion of AI annual recurring revenue — short of an obvious return. Of the raised $190 billion guidance, $25 billion is due to higher component prices, including memory, as Hood herself said. What to watch is whether Azure's growth rate justifies the spend.

Meta — the 29th

Meta may become a replica of Google. Last quarter, CFO Susan Li said, "We have continued to underestimate our compute needs even as we have been ramping capacity significantly." An admission that it cannot but spend more. And she declined to give a 2027 capex outlook — calling it "a very dynamic planning process." JPMorgan cut its rating from Overweight to Neutral on that uncertainty.

The direction. Meta already showed this whole structure last quarter. It raised capex to $125–145 billion, and that day the stock fell 10%, erasing $175 billion in market value. Zuckerberg attributed the raise to "component costs, particularly memory pricing," and Li explained the rising infrastructure cost as "depreciation and data center operating costs." And the reported earnings per share of $10.44 included an $8 billion one-time tax benefit — strip it out, and growth drops to 14%. The same earnings-quality illusion seen in Google. What to watch on the 29th is not whether it raises again, but how the market receives the spend it already raised, and whether it keeps refusing 2027.

Amazon — the 30th

For Amazon, the CEO said this observation out loud. On last quarter's earnings call, Andy Jassy said, "In times of very high growth like now, where the capex growth meaningfully outpaces the revenue growth, the early years free cash flow is challenged." He effectively read this piece's thesis word for word. Amazon had already flagged $200 billion of 2026 capex back in February.

The direction. Amazon is the one of the five most candid about the cash pressure. AWS free cash flow collapsed 95% in a year, from $26 billion to $1.2 billion — because purchases of property and equipment rose by $59.3 billion. And yet Jassy's posture is closer to "good, keep going." Trainium2 is sold out, 3 is nearly sold out, 4 is already being reserved, and $225 billion in commitments are on the line. A signal that there is no reason to stop. And here too is the same illusion — last quarter's EPS of $2.78 was essentially flat once the $16.8 billion Anthropic equity gain is stripped out. What to watch on the 30th is whether AWS's growth rate supports the spend, and whether its own chip, Trainium, changes the cost structure.

The one thread through all four

Three companies, different businesses, but the data they left on the record last quarter points to the same five things. First, a CFO or CEO already said "we spend more" — not a prediction, but their own words. Second, all three named memory prices as a reason for the capex rise. Third, depreciation has begun to press on margins. Fourth, one-time gains (tax benefits or equity marks) inflated EPS, and stripping them reveals the core business. Fifth, a raise was not read as strength — Meta fell 10%, and Google just did.

So the direction reads like this. Within three days, these three are likely to raise, or at least hold and accelerate, capex — because they have said so. And that raise is not a sign of strength, but an extension of the structure all five share — spending faster than they earn, financing that has shifted from own cash to debt and component prices. How the market receives that raise is a separate matter, and the precedent from Google and Meta is on the "did not celebrate" side.

Of course each company differs. Microsoft's $392 billion backlog is a shield, but the ROI gap of spending $97 billion to earn $37 billion is a weight. Meta's advertising core is strong, but its refusal to give 2027 leaves the greatest uncertainty, and it has already fallen 10% once. Amazon has been the most candid about the cash pressure, but its own chip, Trainium, is a variable the others don't have. Same structure, different exposure.

This piece was written before the earnings. The data is public — the three companies' last-quarter earnings-call remarks, the flagged guidance, memory prices, the depreciation pressure. This piece reads that data as pointing in one direction. If it is wrong, the data remains and the direction in which it was read is what was wrong. That the direction was right for Google is no guarantee it will be right for the four. But once again, the direction is written down first, in the open.

This piece records the structure. It does not tell you which side to stand on.

As of 2026-07-23 · written and published before the MS (7/28), Meta (7/29), and Amazon (7/30) earnings. This is a structural prediction based on public data, not individual investment advice.

Sources and verification

Common

• Combined 2026 AI capex of the four $725–785B, ~+77% YoY · multiple reports (IG·Stocktwits synthesis, 2026-07) ○

• capex / group operating cash flow 33% (2023) → ~93% (2026) · multiple reports (NextWave et al.) ○

Microsoft (fiscal year starts July — Q1 FY26 = Oct 2025 results, Q3 FY26 = Apr 29 2026 results. Below spans two calls)

• CFO Hood "We know we're behind. We do need to spend" · MS Q1 FY26 call (2025-10-29) ◎

• "total spend increases sequentially, FY26 growth rate > FY25" · Hood, same call ◎

• Next-quarter capex flagged "$40B+" · MS Q3 FY26 outlook (2026-04-29) ◎

• 2026 capex $190B (+61%), $25B of it component pricing · Hood, Q3 FY26 call (2026-04-29) ◎ · CNBC·Yahoo ○

• Q3 gross margin 67.6% (narrowest since 2022), due to depreciation · CNBC (2026-04-30) ○

• $97B infrastructure spend over 4 quarters → $37B AI ARR · The Register (2026-04-30) ○

• RPO $392B (+51% YoY), AI capacity +80%, DC footprint 2x in 2 years · Nadella, Q1 FY26 call (2025-10-29) ◎

Meta (Q1 2026 = 2026-04-29 results)

• CFO Li "We have continued to underestimate our compute needs" · Meta Q1'26 call (2026-04-29) ◎ · Benzinga ○

• Declined 2027 capex outlook ("very dynamic planning process") · Li, same call ◎ · JPMorgan OW→Neutral (Anmuth: 2027 capex $202B, FCF −$24B) ○

• capex raised to $125–145B → stock −10%, $175B market value erased · multiple reports (CNBC·Fortune·IndexBox, 2026-04-29~30) ○

• Zuckerberg on raise: "component costs, particularly memory" · Meta Q1'26 call ◎

• Li: rising infra cost = "depreciation and data center operating costs" · same call ◎ · Fortune ○

• Reported EPS $10.44 incl. $8.03B one-time tax benefit; ex-benefit growth ~14% · multiple reports (IndexBox·Doolpa) ○

• ~$107B new infrastructure commitments in one quarter, 8,000 layoffs · Global Data Center Hub (2026-05-13) ○

Amazon (Q1 2026 = 2026-04-29 results)

• CEO Jassy "capex growth meaningfully outpaces revenue growth, early years FCF challenged" · Amazon Q1'26 call (2026-04-29) ◎ · Investing.com·TechCrunch transcript ○

• $200B 2026 capex flagged (February) · multiple reports (CNBC) ○

• Q1 capex $43.2–44.2B (+60%) · CNBC·FactSet (2026-04-30) ○

• AWS/company TTM FCF $26B→$1.2B (−95%), property & equipment +$59.3B · multiple reports (CNBC·TechCrunch·MindStudio) ○

• Trainium2 sold out / 3 nearly / 4 reserved, $225B commitments · Jassy, same call ◎ · Alpha Spread ○

• Headline EPS $2.78, flat ex-$16.8B Anthropic equity gain · finsee.ai analysis ○

• Analyst Q on "memory & storage prices impacting this year's and next year's capex" · Amazon Q1'26 call transcript (Investing.com) ◎

Linked pieces

• Google verification: "Alphabet Answered" (2026-07-23, Silicon Nexus)

• Full structure / financing thesis: "The End of Building With Your Own Cash" (Silicon Nexus)

※ ◎ = earnings-call primary remark, ○ = confirmed by multiple reports. Call remarks are per each company's IR transcript; results figures may differ by fiscal calendar and aggregation method (TTM/quarter), so original context is noted. Items in negotiation or planning are unconfirmed.