Big Tech’s Q3 2026 Capex Keeps Climbing. Cash Flow Just Turned Negative

Three trading days in late October 2026 delivered the clearest read yet on where AI infrastructure spending is actually headed — and the message was messier than either the bulls or the bears expected. Alphabet, Microsoft, Meta and Amazon all reported third-quarter results within 48 hours of each other. Three of the four posted negative free cash flow. Two got punished by the market anyway, and two didn’t. None of them cut capital spending. If you’re planning to lease colocation space, sign a PPA, or lock in GPU capacity over the next 12 to 24 months, this week of earnings calls is the single best data point you’ll get all quarter on whether the hyperscalers behind most of that demand are still spending — and on what terms.

Four Earnings Calls, Four Different Markets Reactions

Alphabet reported Q3 capital expenditure of $51.3 billion, up from $44.9 billion the prior quarter, and said it still expects to spend up to $205 billion for full-year 2026 — with a “significant increase” flagged for 2027. Free cash flow came in at negative $9.1 billion, Alphabet’s second straight negative quarter, and the company issued $28.7 billion in senior unsecured notes plus roughly $12.4 billion through at-the-market stock sales to help fund the buildout. Revenue was strong — $127.58 billion, up 25% year over year, with Google Cloud revenue more than doubling to $30.71 billion — but an $8.7 billion loss on equity investments dragged EPS below consensus, and the stock fell about 11% the next morning.

Microsoft told almost the opposite story. Capex for the September quarter came in at $45.2 billion, with guidance pointing to more than $50 billion in the current quarter, yet free cash flow stayed positive at $18.6 billion. Azure grew 48% against a 44% consensus estimate, and commercial remaining performance obligations — contracted, unbilled revenue — hit $677 billion, up 73% year over year. The stock rose roughly 2% on the print. Meta narrowed its 2026 capex range to $135–145 billion after its own Q3 results and reported negative free cash flow of about $2.1 billion, a smaller shortfall than Alphabet’s or Amazon’s, and shares were roughly flat to slightly up.

Amazon had the roughest reaction. AWS revenue grew 39% to $45.98 billion — its fastest growth in nearly five years — and the company held its 2026 infrastructure spending plan at roughly $220 billion. But free cash flow swung sharply negative, and EPS of $1.89 missed the roughly $1.96–1.97 consensus. Shares fell more than 4% into the report and extended losses afterward, erasing on the order of $100 billion in market value in a single session, even as CEO Andy Jassy told investors AWS growth should stay above 40% through 2027 and 2028.

Why Investors Flinched at the Cash-Flow Math

The pattern across three of the four companies — Alphabet, Meta and Amazon all reporting negative free cash flow in the same window, with Alphabet turning to debt and equity markets to help cover the gap — is what moved the market more than any single capex number did. Spending roughly $700 billion a year collectively on data centers and GPUs while generating less cash than you’re spending is sustainable only as long as investors believe the resulting capacity converts into durable, high-margin revenue fast enough to justify it. Microsoft’s result is the reason that belief hasn’t collapsed: its capex is growing just as fast as anyone else’s, but its free cash flow stayed solidly positive, which is largely why its stock went up while Alphabet’s and Amazon’s went down. The market isn’t rejecting AI capex broadly — it’s starting to price hyperscalers differently based on how each one is funding it, a distinction buyers should start tracking as closely as the headline capex figures. This builds directly on the financing dynamics covered in TechInfraHub’s look at the hyperscaler depreciation gap, where the mismatch between capex and reported depreciation was already flashing early warning signs.

The Backlog Counterargument: Demand Still Outstrips Supply

None of this week’s numbers suggest hyperscalers are pulling back. Microsoft’s $677 billion RPO figure — roughly 30% of which is expected to convert to revenue within 12 months — means a large chunk of next year’s Azure growth is already contracted, not projected. Amazon’s Jassy reiterated that AWS has been capacity-constrained, not demand-constrained, for several quarters running, and Alphabet’s own guidance for a “significant increase” in 2027 spend signals the current buildout cycle has at least another full year to run, market jitters notwithstanding. For buyers, that’s the practical takeaway underneath the stock-price noise: the volatility investors are reacting to is about financing structure and cash-flow timing, not about whether hyperscalers still need the gigawatts. Capacity scarcity on the ground — in colocation, in interconnection queues, in power-purchase agreements — isn’t going to ease just because Wall Street spent a rough week second-guessing capex math.

What This Means If You’re Negotiating Capacity Right Now

A few concrete implications follow for anyone actively buying or planning data center capacity this quarter. First, don’t expect this earnings volatility to translate into pricing leverage at the negotiating table — the companies driving colocation, power and transceiver demand are, with the partial exception of Amazon’s guided spend, not actually cutting budgets, so the scarcity dynamics that have been pushing up PPA and lease pricing are unlikely to reverse in the near term. Second, pay closer attention to how a given hyperscaler or neocloud counterparty is funding its buildout, not just how much it’s spending: a provider leaning on debt issuance and equity sales to cover a capex-cash gap carries different counterparty risk than one funding growth out of operating cash flow, which matters if you’re relying on that counterparty’s balance sheet to honor a multi-year capacity commitment. Third, build financing-health checks into supplier diligence for any deal that routes through an intermediary with exposure to the kind of circular financing arrangements already flagged elsewhere — if a neocloud or GPU lessor’s revenue depends on investment from the same hyperscaler it’s reselling capacity to, a cash-flow wobble upstream can show up fast downstream. Finally, treat the spread between this week’s winners and losers as a signal about where capacity will actually get built: capital tends to follow the balance sheets the market rewards, and Microsoft’s results suggest Azure-linked buildout has more room to accelerate than its cash-flow-strained peers.

The Real Signal for 2027 Planning

Strip away the day-to-day stock reaction and the underlying message for anyone planning capacity through 2027 is straightforward: aggregate hyperscaler capex isn’t slowing down, backlogs are growing faster than capacity can be delivered, and the financing side of the industry is under more scrutiny than at any point since this buildout began. That combination argues for locking in capacity and power commitments sooner rather than waiting for a pullback that the backlog data doesn’t support, while simultaneously tightening the diligence you apply to any counterparty whose own spending is outpacing its cash generation. The buyers who come out ahead over the next four quarters will be the ones who treated this earnings week as a stress test of their suppliers’ balance sheets, not just a headline about capex totals.

Get the full framework for vetting counterparty risk, benchmarking lease terms and modeling capacity costs in the Data Center Buyers’ Toolkit.

Written by

Raajeev Ratra

Data Center Infrastructure Expert | 15+ Years in DC Design, Operations & Project Management

Raajeev is a seasoned data center professional with hands-on experience in hyperscale facilities, colocation design, power & cooling infrastructure, and global DC operations. He shares practical insights to help engineers and IT leaders build better infrastructure.

Connect on LinkedIn →