Big Tech Pushes Past $700bn in AI Capex While Cloud Waste Rises for the First Time in Five Years

Week of 28 September 2026 | Compiled by Team Paringa

The hyperscalers are spending more than ever on AI infrastructure, and FinOps teams are managing more AI spend than ever — but the waste numbers are moving the wrong way for the first time since 2021.

What Happened This Week

🔴 Hyperscalers raise 2026 capex past $700bn, and Wall Street is nervous

At their most recent earnings (reported through late July and August 2026, the latest actuals available ahead of Q3 reporting in late October), the four largest hyperscalers all raised or held full-year 2026 capital expenditure guidance. Alphabet lifted its 2026 capex midpoint by roughly 8.1% to $200 billion; Amazon raised its forecast 10% to $220 billion; Meta raised guidance for the second time this year, to $125–145 billion, citing higher component pricing and additional data-centre costs; Microsoft held its calendar-year figure at $175 billion. Combined estimates for Big Tech's 2026 AI infrastructure spend now sit in the $700–760 billion range depending on methodology (Reuters-cited estimates put it at $700bn; Statista's chart-based tally reaches $760bn) — either way, a step-change from 2025.

The spending is buying real growth: Google Cloud posted 82% year-on-year revenue growth in Q2 2026 with operating income up 212%, ahead of Azure's 43% and AWS's 37%. But the market reaction split sharply on capex discipline rather than growth: Alphabet's stock fell 7% on the day of its report despite the cloud numbers, on investor concern about whether the spending will show adequate return, while Microsoft rose 8% specifically because it held capex flat. That divergence — punishing the raise, rewarding the hold — is the clearest signal yet that AI infrastructure ROI, not top-line AI revenue, is now the metric investors are actually pricing.

Sources: The Motley Fool | Yahoo Finance

For clients running FinOps against hyperscaler bills, this is the moment to renegotiate committed-use discounts and reserved capacity terms — hyperscalers under investor pressure to show capex discipline are more willing to deal on unit economics than they were twelve months ago. It's also the moment to stop treating "our AI costs are growing" as self-evidently bad news to leadership: the hyperscalers themselves are being rewarded or punished based on efficiency of spend, not its absolute size, and that's the framing FinOps teams should be taking into their own budget conversations.

🟡 Cloud waste rises for the first time in five years, even as FinOps goes mainstream on AI

Flexera's 2026 State of the Cloud Report, surveying more than 750 cloud decision-makers, found that 29% of IaaS and PaaS budgets are now wasted spend — the first year-on-year increase in waste in five years. At the same time, the FinOps Foundation's sixth annual State of FinOps survey (1,192 respondents representing over $83 billion in annual cloud spend) found that AI cost management has gone from a niche concern to near-universal: 98% of practitioners now manage some AI spend, up from 31% just two years ago, and AI cost management ranks as the number-one skillset organisations say they need. Both reports converge on the same underlying story: FinOps functions have scaled fast to cover AI, but the discipline hasn't caught up with the waste it's creating.

The reporting-line data reinforces this: 78% of FinOps teams now report to CTO/CIO rather than finance (up 18 points), and teams with VP-level executive sponsorship have 2–4x more influence over technology purchasing decisions — meaning the organisations best positioned to fix AI waste are also the ones increasingly empowered to do so, if they act on the data rather than just collecting it.

Sources: FinOps Foundation | Flexera / Accelerate Partners

A 29% waste rate reversing five years of improvement is the number to put in front of any CFO who thinks FinOps maturity is a solved problem. The Foundation's own data shows why: teams "hit the big rocks of waste" years ago and are now chasing smaller, harder-to-find AI-era waste — idle GPU reservations, unmonitored inference spend, agent sprawl — with tooling and governance built for the old, simpler cloud-waste problem. Clients should treat any FinOps programme that hasn't explicitly re-scoped for AI workloads in the last twelve months as running on borrowed maturity.

🟢 AI budget overruns: vendor surveys disagree on the number, agree on the direction

Two September surveys reached different headline figures but the same conclusion. The Futurum Group's 2H 2026 CIO survey (1,636 global enterprise decision-makers, published 10 September) found 46.9% of organisations report AI spend exceeding planned budget, split between moderate (35.6%) and substantial (11.3%) overruns; of those over budget, 47.6% requested additional funding and only 17.2% paused or cut initiatives. A separate DoiT-commissioned Sapio Research survey (500 finance leaders, published in June) put cost overruns at 79% over the prior 12 months — and, counterintuitively, found organisations with the most mature FinOps practices reported the highest overrun rates (89%), which the report attributes to better detection catching overruns that less mature organisations simply aren't measuring.

Sources: Futurum Group | DoiT

Flag these as vendor-commissioned surveys with different methodologies and samples rather than a single reconciled industry figure — but don't let the discrepancy obscure the shared signal: whether the true rate is closer to 47% or 79%, a plurality-to-majority of enterprises are overspending on AI and choosing to fund the overrun rather than cut scope. That's a governance gap, not a budgeting error, and it's the gap Atmoz-style cost visibility tooling is built to close before the CFO conversation turns adversarial.

What FinOps Leaders Are Dealing With

The structural shift underneath all three stories is that FinOps has stopped being a cloud-cost function and become a technology-value function — the Foundation's own mission language now says "Value of Technology" rather than "Value of Cloud." That expansion (90% of practitioners now also manage SaaS, 64% licensing, 57% private cloud) is happening at the same time as AI workloads introduce a genuinely new cost-management problem: agentic AI. IDC data cited at Flexera's FinOps Forward 2026 event put AI agent deployments at 28.8 million in 2025, projected to grow roughly 80x by the end of 2026 — and IDC's Jevin Jensen warned bluntly that "shadow IT is coming back with a vengeance" as agents become trivially easy to spin up outside any governance perimeter. FinOps teams that scaled to cover AI compute costs are now being asked to cover a fleet of autonomous agents making their own resource decisions, with only 7.5% of enterprises reporting FinOps embedded into AI projects from the start.

Three Things Worth Watching Next Week

1. Q3 2026 hyperscaler earnings — Alphabet, Amazon, Meta and Microsoft report calendar/fiscal Q3 results in late October; watch whether capex guidance rises again or whether investor pressure (visible in Alphabet's 7% post-earnings drop) forces a more cautious tone.

2. Agent-driven shadow IT tooling — Expect more FinOps and cloud-security vendors to launch AI-agent cost-visibility and governance products in Q4, responding directly to IDC's 80x growth projection.

3. Reconciliation of overspend survey methodologies — With Futurum and DoiT/Sapio publishing conflicting overrun rates, expect a third analyst (Gartner, IDC or Forrester) to weigh in with its own figure before year-end, likely settling public discourse on a single benchmark number.

Could not verify: Combined 2026 hyperscaler capex is cited variously as ~$700bn (Reuters-sourced) and ~$760bn (Statista); the discrepancy reflects differing company sets and rounding rather than a factual conflict, but no single authoritative combined figure exists yet. The 46.9% vs. 79% AI-overspend figures are both vendor-commissioned surveys with different samples and time windows and should not be quoted as a single industry consensus number.

Questions? hello@paringa.ai

Paringa works with FinOps and infrastructure teams navigating exactly this gap — capex racing ahead of cost governance, and AI agents outrunning the tooling built to track them. Atmoz's cost-visibility approach is built for the AI-era waste problem Flexera's data shows is already reversing five years of progress. | 28 September 2026 | © Paringa Ltd