That $1.1 Trillion Cloud Backlog Isn't What You Think It Is
Tyler
Co-Founder & CEO
A wake-up call for CIOs, CFOs, CTOs, and every FinOps team watching the scoreboard from the wrong side of the field.
Most people see momentum. They see validation. They see proof that AI is real, cloud is the future, and the big three Amazon, Google, Microsoft are printing money faster than they can deliver it.
I see something else. I see a warning label that most enterprise leaders are walking right past.
I spent years inside these machines. I watched how contracts get structured, how commitments get sold, how capacity gets allocated when demand exceeds supply. And right now, the story that's being told to the market is very different from the story that will be written in your next board presentation.
So let's talk about what's actually happening.
The Scoreboard People Are Reading Wrong
Here's the headline as it's being reported: The major cloud providers are sitting on staggering backlogs of signed but undelivered contracts. As of their most recent earnings (Sherwood News, GeekWire, Windows Forum):
-
Microsoft Azure reported an RPO (Remaining Performance Obligation) of approximately $368 billion, up 37% year-over-year (LinkedIn analysis, GeekWire).
-
Google Cloud is carrying roughly $155 billion in committed future revenue, growing at a breakneck pace off the back of Gemini deals and enterprise AI commitments.
-
AWS holds around $195–200 billion in backlog.
-
Add it all up and you're looking at well north of $1 trillion in contracted, undelivered cloud services.
The financial press treats this as proof of strength. And in one narrow sense, it is. These companies have locked in future revenue. Investors love it. Analysts praise "visibility."
You're the Financing Mechanism
Here's a number that should cause you to pause: Hyperscalers are projected to spend over $443 billion in capital expenditure in 2025, scaling toward $602 billion in 2026. Goldman Sachs estimates the total hyperscaler capex bill from 2025–2027 will hit $1.15 trillion — more than double what they spent in the three years prior.
About 75% of that is going to AI infrastructure: GPUs, custom silicon, liquid-cooled data centers, and the fiber and power to connect them.
Now ask yourself: where is that money coming from? In large part, it's coming from the multi-year enterprise contracts your team signed. The committed spend you negotiated with your hyperscaler rep last year. The EDP (Enterprise Discount Program) deal your CFO approved because the per-unit economics looked compelling.
You signed a commitment. They took that capital certainty to the bond market and the boardroom, and they allocated it to building AI infrastructure that is, first and foremost, designed to serve AI-native workloads - OpenAI, Anthropic, Databricks, and the hyperscaler's own internal AI products.
Your traditional enterprise workloads? You're in a different queue.
This isn't a conspiracy. It's just economics. When demand exceeds supply and your best customers are AI companies burning through GPU clusters at a pace that dwarfs legacy enterprise workloads, you allocate accordingly. I watched this prioritization logic play out in real time from inside the machine. It's rational. It's just not disclosed.
The result: enterprises are, in some cases, being steered to secondary regions, facing longer provisioning timelines, or discovering that the capacity they thought they were buying into is pre-committed elsewhere. Capacity reservation clauses in your contracts are no longer a nice-to-have. They're table stakes.
The FinOps Illusion: Why Your Governance Model Is Already Obsolete
Let me be blunt about something that will make some FinOps practitioners uncomfortable.
The tools and practices that got you here aren't built for where you're going.
The 2025 State of FinOps report - representing organizations responsible for more than $69 billion in cloud spend - surfaces a picture of an industry in governance crisis:
-
89% of practitioners say that lack of cloud cost visibility impacts their ability to do their job - and nearly half describe that impact as significant.
-
Only 2% of CIOs report spending less on cloud than they projected. Two percent. That means 98% of CIOs either hit their target exactly (unlikely) or overspent.
-
A Harness report estimates enterprises will waste roughly $44.5 billion in cloud infrastructure spend in 2025 - unused compute, idle resources, misconfigurations, forgotten instances quietly billing against accounts nobody audited.
-
Broader estimates put cloud waste between 28–55% of total spend, depending on organizational maturity.
Now layer in AI. Because AI workloads don't play by the same rules.
AI infrastructure costs - GPU reservations, training runs, inference endpoints are notoriously opaque compared to traditional compute. Pricing is more variable. Consumption is harder to predict. And the governance frameworks most FinOps teams built over the last five years were designed for a world of VMs, storage buckets, and data egress not token throughput, model training jobs, and reserved GPU capacity.
63% of organizations are now tracking AI spend - up from 31% just a year ago. But the keyword there is tracking. Tracking is not governing. Tracking is not optimizing. Tracking is just watching the meter run.
The FinOps Foundation's 2025 framework has begun integrating AI cost management, but the honest reality is that most enterprise teams are still several maturity levels away from having the controls, tooling, and organizational muscle to govern AI spend the way they govern traditional cloud.
Meanwhile, the contracts keep getting signed.
The ROI Gap Nobody Wants to Put in the Deck
Here's a number to think about: AI-related cloud services are expected to deliver approximately $25 billion in revenue in 2025 to the hyperscalers against a capex investment of over $443 billion.
That's roughly a 10% return on infrastructure invested. In any other industry, that math would be scrutinized relentlessly. Here, it's celebrated as a sign of how early we are.
Maybe it is. But for the enterprises on the other side of those contracts, the ROI picture is similarly sobering:
-
Only 25% of AI initiatives have delivered their expected ROI to date.
-
Fewer than 20% have been scaled across an entire enterprise.
-
Meta - a company with AI capabilities most enterprises can only dream about — has acknowledged that their current AI ROI is, in their own words, "terrible."
And yet enterprise AI spending is accelerating. Commitments are being made. Contracts are being signed. Backlogs are growing.
I'm not saying AI isn't real or that the investment isn't justified. I spent enough time building cloud infrastructure to know that the early economics of transformative platforms always look painful before they look prescient.
But I am saying this: the gap between what enterprises are committing to and what they can currently govern, measure, and extract value from is concerning. And the people who need to close that gap, the CIOs, CFOs, CTOs, and FinOps leads reading this - are working with tools and frameworks that were not designed for this moment.
What You Should Actually Be Doing Right Now
I'm not here to just drop the problem on your desk. Here's what I'd be doing if I were running a FinOps function or sitting in a CIO chair today:
1. Audit your commitments with fresh eyes.
Pull every cloud commitment you have - EDPs, Reserved Instances, private pricing agreements and map them against actual utilization. Not projected utilization. Actual. If you're carrying unused commitment, understand why, and determine whether the discount justifies the waste. At current waste rates, it often doesn't.
2. Get capacity clauses in writing.
If your enterprise cloud contracts don't specify capacity availability by region and workload type, you have a relationship, not a contract. As hyperscalers prioritize AI infrastructure, enterprise workloads need legal teeth to ensure they're not left waiting in line.
3. Build a downtime governance strategy.
Here's something almost no enterprise FinOps team has operationalized: tracking cloud downtime and claiming the SLA credits they're contractually owed. Research by Oxford Economics suggests the Global 2000 spends an average of $200+ million per year on downtime. AWS, Azure, and GCP who collectively control 68% of the global cloud market - have SLA credit mechanisms built into their contracts. Most enterprises never claim them, because no one owns the process.
And if you think this is a "nice to have," look at what's happening in Europe. The EU's Digital Operational Resilience Act (DORA), which became fully enforceable on January 17, 2025, now requires all financial entities to maintain comprehensive registers of cloud provider contracts, enforce SLA accountability, retain audit rights, and document exit strategies. In November 2025, the EU went further formally classifying AWS, Azure, and Google Cloud as critical ICT infrastructure, subjecting them to direct regulatory supervision.
DORA is a European financial regulation today. It is a global governance template for tomorrow. North American CIOs and CFOs who treat it as "someone else's problem" are underestimating how fast operational resilience requirements travel. Building your downtime governance framework now - tracking SLA performance, automating credit claims, and documenting vendor dependencies - is not just good FinOps hygiene. It's regulatory future-proofing.
4. Renegotiate with more leverage than you think you have.
I've sat across the table from enterprise buyers and hyperscaler account teams. Enterprises chronically underestimate their negotiating leverage, especially at renewal time. The backlog is large. Competition between providers is intensifying. Google, Microsoft, and AWS are all fighting for enterprise share. Use that.
The Bottom Line
That $1.1 trillion backlog is real. The demand is real. The AI buildout is happening at a scale that will reshape computing for decades.
But the narrative being told - that this is simply proof of hyperscaler strength and enterprise cloud adoption - misses the more important story for the people actually running these budgets.
You are, in part, financing an AI infrastructure race you may not be the primary beneficiary of. Your governance frameworks are not keeping pace with the complexity of what you're committing to. And the contracts being signed today, in the heat of AI excitement and competitive pressure, will define your cost structures for years.
The $1.1 trillion number doesn't scare me. What concerns me is the absence of any real discussion about who's truly footing the bill for this expansion.
Sources
Sherwood News — Big Tech's $1.1T Cloud Backlog
GeekWire — Microsoft Beats Expectations, Cloud Tops $50B
LinkedIn — Microsoft's $625B RPO Backlog
FinOps Foundation — State of FinOps 2025
Goldman Sachs — AI Infrastructure Outlook
Harness — Cloud Cost Report 2025
Windows Forum — Q4 2025 Cloud AI Earnings
Cloud Computing News — Hyperscaler Infrastructure Impact