AI Workloads Just Reversed Five Years of Cloud Waste Reduction

Week of 10 August 2026 | Compiled by Team Paringa

Cloud waste was trending down for five years. AI broke the trend - and this time nobody can agree who owns the bill.

SECTION 1 - What Happened This Week

🔴 Flexera 2026: AI workloads push cloud waste to 29%, ending a five-year decline

Flexera's 2026 State of the Cloud Report - a survey of 753 cloud decision-makers, 62% based in the US - found that AI workloads and new cloud services pushed estimated waste to 29% of cloud spend, reversing a downward trend that had held for five years. This isn't a marginal regression: it's the first time in half a decade that the industry-wide waste number has moved in the wrong direction, and AI is the named cause.

The mechanics are visible elsewhere in the same wave of 2026 research. GPUs now make up 18% of spend at AI-forward organisations, up from just 4% in 2023 - but statically provisioned GPU fleets are running at only 30-40% utilisation. Multi-cloud enterprises are separately estimated to waste around 29% of IaaS/PaaS spend on idle, overprovisioned, or redundant resources. Almost all IT leaders (97%) now believe some portion of their cloud spend is wasted, and 52% put that waste above 25% of total spend.

The pattern across every 2026 dataset is the same: organisations provisioned AI infrastructure fast, under pressure to not fall behind competitors, and skipped the rightsizing and commitment-based purchasing discipline that took years to build for traditional cloud. They're now paying for GPU capacity sitting idle two-thirds of the time.

Sources: Flexera Finds Cloud Value is Rising While AI Waste Grows, TechInformed - AI workloads drive estimated cloud waste to 29%

Every FinOps programme built before 2024 was tuned for compute and storage waste, not GPU waste. Static provisioning is the default because nobody's built the rightsizing muscle for AI infrastructure yet - the visibility → rightsizing → commitment sequence that worked for traditional cloud hasn't been rebuilt for AI spend. That gap is exactly where a 90-day recovery engagement pays for itself several times over before Q4 budgets lock.

🟡 Harness: 26% of AI spend wasted, 52% have no named owner

Harness's 2026 State of AI in FinOps report, based on 700 engineering leaders and practitioners across five countries, found that enterprise AI spending is outpacing organisations' ability to track, own, or justify it. The headline number: 26% of AI spend is estimated as wasted. For the 1 in 5 organisations spending over $1 million a month on AI, that's roughly $260,000 monthly with no measurable return.

The ownership gap is the more damaging finding. 52% of respondents say there is no clear owner of AI cost - split across engineering, FinOps, finance, and IT with nobody accountable end to end. 72% report an unexpected AI cost spike or bill in the past year, a third of them more than once, and only 20% could identify the cause of a spend doubling within hours. This lines up with the FinOps Foundation's own 2026 data: 98% of FinOps teams now manage some AI spend, up from just 31% two years ago - practice has scaled far faster than governance has caught up.

Sources: Harness - New Report Reveals Enterprise AI Spend Has Outgrown the Systems Built to Track It, Digitalisation World - Enterprises waste 26% of AI spend amid unclear cost ownership

"No clear owner" is the tell. Waste survives not because nobody notices it, but because nobody's job is to fix it. Before touching tooling, fix the org chart problem: assign a named owner for AI spend with the authority to kill idle capacity, then layer visibility and rightsizing on top. Tooling without ownership just produces a nicer dashboard of the same unowned waste.

SECTION 2 - What Finance and Infrastructure Leaders Are Dealing With

The FinOps discipline itself is being redefined mid-flight: the FinOps Foundation renamed its mission from "advancing the people who manage the value of cloud" to "advancing the people who manage the value of technology" in 2026, explicitly folding in AI, SaaS, licensing, and data centre spend. That's not cosmetic - it reflects that CFOs and infrastructure leads are now fighting the same battle (unowned, unpredictable, fast-growing spend) across four categories at once rather than one. Shadow AI compounds it further: enterprise AI tool spend is growing 40% year-over-year, with an estimated 25-35% happening entirely outside IT visibility, and over a third of that shadow spend duplicates tools the company already pays for.

Three Things Worth Watching Next Week

1. Q3/Q4 budget lockdowns - Finance teams building 2027 budgets off 2026 AI run-rate numbers without correcting for the 26-29% waste baked in; expect overcorrection or under-budgeting in either direction.

2. GPU utilisation tooling - Watch for vendors moving fast into the static-provisioning-to-dynamic-allocation gap Flexera flagged; this is the fastest-growing wedge in the FinOps tooling market right now.

3. AI cost ownership mandates - With 52% of organisations lacking a named AI spend owner, expect a wave of CFO-driven "assign an owner" initiatives through Q3 as boards start asking pointed questions about AI ROI.

Questions? hello@paringa.ai

Paringa pairs Atmoz's cost intelligence with hands-on FinOps discipline - turning "nobody owns this" AI spend into a governed, rightsized line item before the next budget cycle locks it in. | 10 August 2026 | © Paringa Ltd