Nvidia's Real Risk Isn't the AI Chips — It's a $700 Billion CapEx Checkbook Outrunning the Cash Behind It

Meta began 2026 guiding $115–135 billion of capital spending. After first-quarter earnings in April, it raised that to $125–145 billion, pointing at higher component prices and additional data-center capacity. By July the range had tightened again, to $135–145 billion — two guidance increases inside a single year, resting on the $72.2 billion Meta actually spent in 2025. Put those on a matched, full-year basis and the number speaks plainly: at the top of the raised range, $145 billion is roughly double the capital Meta laid out the year before. A company that commits to nearly doubling its own capital budget inside twelve months is not being cautious. Two raises in one year is not hedging. It is a bet that the returns arrive before the cash runs out.

Period	Low (USD B)	High (USD B)
2025 actual	72.2	72.2
2026 initial guide (Feb 2026)	115.0	135.0
2026 raised guide (Apr 2026)	125.0	145.0
2026 narrowed guide (Jul 2026)	135.0	145.0

The market reads the wave as simple good news — more hyperscaler building, more GPU sales — and mostly shrugs it off as bullish for Nvidia. Meta is far from the outlier. Goldman Sachs put Amazon, Microsoft, Alphabet and Meta on pace for roughly $757 billion of combined 2026 capital spending; CNBC had the same four names near $700 billion as early as February. Two dates, two estimates, one conclusion: the group is on track to spend somewhere around three-quarters of a trillion dollars this year, a large share aimed at data centers, GPUs, networking and power. That money has to land somewhere. Nvidia is the primary GPU supplier to nearly every major AI build, so its data-center revenue is, in effect, the hyperscaler capex line moved across into Nvidia's income statement. Strip the roadmaps away and the whole AI-infrastructure thesis reduces to a single proposition: the hyperscalers keep raising the number. Nvidia, at a $5.55 trillion market cap and roughly 18 times trailing sales per Ainvest data, already has an enormous amount of that continuity priced in.

Here is the part the bull case steps over: the race is being paid for with money these companies are not generating. Bain & Company's read is that capital spending is rising faster than operating cash flow, pushing hyperscalers toward debt markets and off-balance-sheet structures — Bain calls much of the investment a "circular bet": each new data center and model compels a competitor to build yet another, while many firms fall short of the promised cost savings and productivity gains. Meta, on that account, has shifted parts of its infrastructure obligations off its balance sheet, making the liabilities less visible in conventional debt metrics. Hold those financing details at the right distance. They are the analysts' read on how the build is being carried — not a settled accounting conclusion, and not a realized downturn. No hyperscaler has announced a cut; the warning is a forward risk premised on diminishing returns and a financing squeeze. CNBC put the immediate cost plainly when it flagged a big drop in free cash flow, projecting Amazon to turn cash-flow negative this year. So the failure mode for this trade is not engineering. Nvidia's execution is not the risk; the risk is that the people buying its chips run low on money that costs them nothing to hand over. The binding constraint is the checkbook, not the fab. And the cleanest place to see it break is the single number Meta has already moved twice. If Meta holds or raises guidance while free cash flow sags, the race rolls on and Nvidia's revenue line holds. The moment Meta cuts below the $135 billion low end it set in July, the self-reinforcing race is broken — and Nvidia's data-center line is the first place the market will feel it.