Your Cloud Economics Pulse For April 2026

Welcome to April’s Cloud Economics Pulse, CloudZero’s monthly look at cloud spend as AI moves from cost problem to strategic commitment.

March’s Pulse called 4.01% a record. It lasted all of 31 days.

Why? February’s billing data came in at 4.84% aggregate AI/ML share. That’s another high, another acceleration. You’ve heard it before and it’s getting a bit boring now, but the story isn’t in the numbers; it’s now in the behavior. Something has shifted in how organizations are approaching AI spend. 

For months, the conversation in FinOps circles has centered on control; how to manage costs that move fast, hide in unexpected places, and compound quietly. That framing rested on the assumption that AI spend was something happening to organizations.

That’s changed now for some orgs. A growing cohort is approaching it differently. They’re pushing harder, deliberately maximizing AI consumption on the logic that more input leads to more output. 

It also connects to a pattern our FinOps in the AI Era report documented across 475 executives: organizations deliberately accepting margin compression in the near term on the bet that AI investment pays off later. Strategic unprofitability, defined as spending ahead of provable return, is increasingly a conscious choice, not a planning failure. An “all in” vibe is what that looks like at the infrastructure level. 

We’ll get more into that below, but first, the billing data we’re seeing in March is a reflection of this shift. Both ends of the market accelerated into March, not just the median catching up to heavy users but also heavy users pushing further. It’s a simultaneous escalation of median and average, not a convergence.

That’s where the conversation is headed. The question of whether to spend on AI is largely settled. The harder question is whether organizations can connect what they’re spending to what they’re getting. The bill is clear and the return isn’t. Yet.

How We’re Looking At Data (And Why It Matters)

For the Cloud Economics Pulse, we track monthly cloud spend trends using anonymized, aggregated data from CloudZero’s network.

Together, these views show not just where cloud dollars go, but how spending patterns shift as new technologies — especially AI — reshape the cost landscape.

One note on methodology this month: You may find that the AI/ML percentage in the service category section differs from the average in the dedicated AI/ML section. Both are correct; just measured differently. The service category is money-related; essentially, total AI/ML dollars divided by total cloud dollars across all customers. Meanwhile, the average/median figures are org-weighted — every customer counts equally regardless of size. 

Research Report

FinOps In The AI Era: A Critical Recalibration

What 475 executives told us about AI and cloud efficiency.

Main Highlights For April 2026

1. Cost By Provider

Here, we’re looking at how overall cloud spend is distributed across providers: 

March’s provider mix is a study in two speeds. At the top, the Big Three are locked in — barely moving, collectively dominant. The real action is a layer below.

AWS ticked up to 69.20%, a 62-basis-point gain from February. Year-over-year it’s down 175 basis points from March 2025’s 70.95%, when it was coming off its mid-2025 peak above 72%. That pullback has stabilized. AWS has settled into the 67–70% band, and within that range, the workload mix continues to deepen rather than diversify away.

Meanwhile, Azure slipped to 10.81% — its lowest point in the dataset, down 78 basis points year-over-year. Nothing dramatic has happened MoM during that time span; it just looks like consistent drift without a floor (yet?). GCP, on the other hand, pulled back slightly to 6.18% but is actually up 46 basis points YoY. It’s holding ground in a way Azure isn’t.

AWS Marketplace hit 3.72% in March — up 78 basis points month-over-month and up more than two full percentage points year-over-year. That’s the largest year-over-year gain of any provider in this dataset, and it has more than doubled since January 2025. Third-party tooling is consolidating into the AWS ecosystem at scale. This is no longer a trend worth watching. It’s the new default. 

Datadog, meanwhile, continued its post-January slide to 1.65%, well below its 2.25% from a year earlier. The verdict is still out on that steady decline. On one hand, you could view it as normalization, but on the other hand, we should watch this closely and determine over time whether observability spend is actually consolidating into other categories.

OpenAI reached 0.73% in March — nearly triple its January 2025 level of 0.28%. Anthropic hit 0.10%, up from effectively zero a year ago, with uninterrupted month-over-month growth since September 2025. Neither number is large at the moment, but both are only moving in one direction: upwards.

Key takeaways

2. Cost By Service Category

Here, we’re looking at how overall spend is distributed across cloud services:

March’s service mix tells a familiar story with sharper edges. The structural rebalancing underway since mid-2025 continued. Compute is yielding share, AI/ML is absorbing it, the data layer is holding its new configuration. 

Compute fell to 47.93%, down 58 basis points from February and 158 basis points year-over-year. Usage isn’t shrinking — everything else is growing faster. Compute still anchors the stack, but it’s a bigger, more complex one than one year earlier. 

The data layer comes in two: Storage at 11.02%, down slightly from February and still up YoY, while Databases is essentially flat MoM but down a dramatic 2.29 percentage points from March of last year. What’s likely driving all this is AI workloads, which generate persistent storage demand in the form of embeddings, retrieval indexes, and training artifacts while traditional database growth is plateauing. In short: the data layer is reshaping around AI, not the other way around.

AI/ML hit 4.84% in aggregate, up 86 basis points from February’s 3.98%. That’s nearly triple September 2025’s 1.74%. We’ll dig deeper into this in the next section.

Key takeaways

3. Cost Of AI/ML

Here, we’re looking at how AI and machine learning costs are growing as a share of total cloud spend — shown as both average and median percentages to capture the full distribution of adoption across organizations:

The org-weighted average reached 3.94% in March, up 63 basis points from February. As we saw in the service section, the aggregate dollar-weighted figure — total AI/ML dollars divided by total cloud dollars across all customers — hit 4.84%, up 86 basis points. Meanwhile, the median reached 1.06%, up 23 basis points from the previous month. 

All three are new highs. Year-over-year, the average is 2.3x March 2025. The median is 4.6x.

For most of 2025, median AI/ML spend climbed steadily but modestly. Then, in February, it jumped 22 basis points. In March, another 23. 

The average tells the same story: grinding single-digit monthly gains through mid-2025, then 60 basis points in February and 63 in March — or a growth of full 1.2 percentage points in just two months. Both ends of the market accelerated simultaneously, and they’ve now done it two months in a row.

This is where “all in” enters the picture: the practice of deliberately maximizing AI consumption on the premise that more input produces more output. 

The logic is straightforward: if AI creates value, more AI creates more value. It’s an offensive posture toward spend, not a defensive one. That’s what the billing data increasingly suggests — the upward average isn’t spiraling costs, it’s a deliberate strategic choice to push harder. For this cohort, AI is an investment, and the bet is on consumption. 

The avg/median ratio tells the same story in compressed form. In January 2025, the average was 8.56x the median — a wide gap driven by a small number of heavy users. By March 2026, that ratio compressed to 3.72x. The gap is closing not because heavy users are pulling back, but because the middle of the market is accelerating toward them. 

Just as in poker, this is a bet. The input side is well documented. It’s in the bill. The output side measured in features shipped, customers retained, margins held doesn’t appear anywhere in a cloud bill. 

That visibility gap isn’t new to those in FinOps. AI makes it wider and faster-moving than anything before it. The organizations that connect consumption to production will have a structural advantage over those still reading the bill as the final word. 

AI/ML bills are surging and the commitment is clear, and ROI still needs more tangible data to be accurately calculated.

Key takeaways

Actionable Guidance

The data is clear on the input side. AI spend is accelerating, the commitment is real, and for a growing cohort it’s intentional. The harder problem, and the one this month’s data puts in sharp relief, is building the output side of the equation. 

Here are four areas to focus on:

1. Map your AI workloads to a business outcome before you add another one

The impulse to scale AI consumption is understandable — and increasingly, it’s deliberate. But scaling without a defined output metric is just spending. 

Before the next workload goes to production, identify what it’s supposed to move. Revenue, retention, deflection, velocity — pick one. It doesn’t need to be precise. It needs to exist.

2. Track cost per output, not just cost in aggregate

Aggregate AI/ML spend tells you the size of the bet. Cost per inference call, per feature, per customer, or per outcome tells you whether the bet is paying off — that’s granular allocation. 

Even a rough unit cost — total AI spend divided by a meaningful output denominator — gives you something to trend over time. That trending is where the insight lives.

3. Treat inference costs as a product cost, not an infrastructure cost

The output side of the AI equation lives in your products and so do the costs that generate it. Inference calls, retrieval, and token volume aren’t abstract infrastructure line items. They map to features, and features map to customers. 

Organizations that start allocating AI costs at the product or feature level will have a fundamentally cleaner view of return than those still reading aggregate cloud bills. Start with one product line. The methodology scales once you prove it out.

4. Pressure-test your costliest AI workloads quarterly

All in makes sense when the return is real, but can also compound if you don’t regularly evaluate. 

Pick your top three AI spend categories by dollar volume and run a simple test: what does this workload produce, and is that output still worth the current run rate? 

The discipline of asking questions regularly is what separates intentional AI investment from accumulated AI spend.

Your Takeaway For This Month

The acceleration in March’s data confirms what we already know. The question of whether to spend on AI is maturing to a question of how much and especially: how do you know it’s paying off? Orgs are leaning into AI because they know there’s value in it. There’s a healthy appetite for AI-fueled workloads and products.

The conviction is real and growing. What’s lagging is the infrastructure to validate it; not the tools to spend, but the tools to evaluate. Assumption is not a measurement strategy. 

The teams that pull ahead won’t necessarily be the heaviest AI spenders or the most conservative ones. They’ll be the ones who can read both sides of the equation — the input and the output — and make faster, better-grounded decisions because of it. 

Thoughts, comments, disagreements? Reply to this Pulse or email [email protected] with “CEP” in the subject heading. We’ll feature the best feedback in an upcoming issue. Watch for our next Cloud Economics Pulse on May 12, 2026, and on the second Tuesday of every month.

Author: Keith MacKenzie

Keith is CloudZero's Content Marketing Manager with a specialty in topics around FinOps, SaaS, AI, and overall work optimization. He also brings more than a decade's worth of background as an editor and writer in the mainstream media and content marketing industries. He's also been to Chernobyl twice – and (sometimes) still has all his marbles.