Your Cloud Economics Pulse For March 2026

Welcome to March’s Cloud Economics Pulse, CloudZero’s monthly look at cloud spend as AI moves from cost line item to cost problem.

Two things landed in February that are worth holding together.

Last month’s Pulse established a new baseline, calling it a floor, not a ceiling. Then CloudZero’s FinOps in the AI Era: A Critical Recalibration benchmark survey report documented something interesting: FinOps maturity improved sharply across the industry while cost efficiency declined across every segment and every quartile. 

In other words, we saw more standardized process but less actual control. The variable breaking the relationship between the two is, of course, AI.  

Last month’s Pulse billing data connects both, and shows how the floor is moving at break-bank speed.

This month’s Pulse numbers made an impression with us — not because we were surprised, but rather, this tracked with much of what we’re hearing from customers and prospects. Namely, the data layer crossed a threshold it hasn’t crossed before, with the AI/ML line posting its biggest single-month move since we started tracking the data. 

This tracks with our survey report findings. The adoption side is widening in real billing data, whereas the efficiency side isn’t keeping pace. That’s this month’s story, and the numbers make the case better than we can in an intro.

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. 

Main Highlights For March 2026

1. Cost By Provider

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

February’s provider mix shows entrenchment at the top and slow, steady movement underneath.

AWS ticked up to 67.71% after January’s dip to 67.16%. Nothing dramatic — AWS has held the 67–68% band since Q4 2025, well off its August–September peak above 70%. It remains the dominant platform for compute, data, and AI workloads, and nothing in February’s data suggests this will change anytime soon.

Azure slipped again to 11.6%, continuing a pattern of gradual erosion that’s been running most of the year. It hasn’t held above 12% since May 2025. No single-month move is alarming, but the downward track is consistent.

GCP held at 6.65% for the second consecutive month — its highest sustained level in the dataset. A year ago, it was at 5.52%. That’s more than a full point of share gain over 12 months. The analytics and data workload story remains intact.

Outside the Big Three, a few moves are worth looking at.

AWS Marketplace bounced back to 3.06% after January’s pullback to 2.58%. That recovery, combined with December’s 3.22% high, reinforces our consolidation thesis: third-party tooling is migrating into the AWS ecosystem, and the underlying trend is up even when individual months pull back seasonally.

Datadog gave back most of January’s spike, falling to 1.97% from 2.56%. January’s jump now reads clearly as contract timing noise — annual renewals hitting in Q1 — not a structural shift. Let’s watch the March data unfold to confirm normalization.

OpenAI reached 0.81%, another incremental high. That’s nearly triple since January 2025 (0.28%). Not a large number in absolute terms, of course, but it’s worth noting for the trajectory and consistency.

Anthropic hit 0.08%. Like OpenAI, the number is still small — but Anthropic is the only provider in this dataset showing uninterrupted month-over-month growth since September 2025. That persistence is worth more than the number itself.

The overall picture: provider consolidation continues. The Big Three are locked in as usual. The real action is in the layer below, with observability, marketplaces, and AI API vendors slowly but ultimately claiming share.

Key takeaways:

2. Cost By Service Category

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

Two numbers in February’s service mix stand out immediately, and they tell a connected story.

Storage hit 11.17% — up 54 basis points from January, a new high in the dataset even if it’s ultimately just over half a percentage point. Databases fell to 10.6%, down 55 basis points, also a new extreme albeit the other way. Most compelling: storage share now exceeds database share for the first time in our records. That marks a potential structural rotation.

The likely driver? AI workloads. Embedding storage, vector databases, retrieval layers, and training artifact retention are all driven by AI and generate persistent storage demand. Meanwhile, traditional relational database workloads are growing more slowly than everything around them. That means the data layer is reshaping itself.

Meanwhile, AI/ML hit 4.01% of aggregate cloud spend, again at its highest point in the dataset. We’ll unpack that fully in the next section, a dedicated dive into AI/ML.

Compute creeped up slightly to 48.39% from January’s 48.06% — a minor uptick that interrupts the recent softening streak but doesn’t reverse it. The two-year trend of compute share compression remains intact. What might look like compression at first is actually just the surrounding stack growing and taking up more of the service pie.

The “Other” category dropped to 15.34% from 16.48% in January, unwinding some of December’s elevation. That’s a pretty hefty drop of 1.14 percentage points in a single month. This is likely due to stabilization in container orchestration and platform overhead usage and spend after a brief spike.

And last but not least, security crept up to 1.26%. Not large in absolute terms, but a persistent year-long climb that hasn’t reversed.

The broader story in all of this for February is that the data layer is rotating from relational to retrieval. AI is claiming an accelerating share of aggregate dollars. And like snow thawing after a long winter, compute is continuing its very slow yield of the stack to everything that is built around it.

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:

In last month’s Pulse, we called 2.67% a floor. What little we knew. The latest data makes that call look conservative.

Let’s start with averages. The org-weighted average reached 3.32% in February, up 61 basis points from January’s 2.71%. The aggregate dollar-weighted figure from the service category hit 4.01% — the largest single-month jump in the dataset. And median AI/ML spend reached 0.83%, up 22 basis points from January’s 0.61% — also the largest single-month move on record.

What’s more: all three figures are new highs. The average has more than doubled year-over-year (1.53% in February 2025), and the median has practically quadrupled (0.21% in February 2025 to 0.83% now).

The gap between the aggregate (4.01%) and the org-weighted average (3.32%) tells you something specific: the heaviest cloud spenders are leaning into AI harder than the middle of the market, pulling the dollar-weighted figure above what the typical org is running. 

All the same, both numbers are real. They’re just measuring different things.

The median is the one to watch most closely. Unlike averages, medians don’t get distorted by a few large outliers. A 22-basis-point median jump in a single month marks a structural shift, and not just at the top. Last month’s Pulse noted that the average-to-median ratio had narrowed from 8.6x (January 2025) to 4.2x (January 2026), signaling AI adoption spreading beyond heavy users. February’s data shows continuation of that compression: the ratio now sits at 4.0x. In plain terms, the middle of the market is catching up.

But catching up in spend without catching up in visibility is exactly how the adoption-efficiency gap widens — and February’s billing data is that gap in motion.

The FinOps in the AI Era report documented it at the survey level: maturity up, efficiency down, AI the variable that broke the relationship between the two. The details are in the report. The short version: organizations are spending faster than they can see, attribute, or forecast. February’s numbers are what that looks like in an actual bill.

Last month’s Pulse recommended auditing your real AI cost footprint beyond the AI/ML line item. February’s 4.01% aggregate versus the 2.67% that felt like a record just 30 days ago is the urgent case for doing that now. The true AI-driven share of cloud spend is much, much higher than what’s explicitly attributed here, as AI costs are often buried in compute, storage, and databases. 

AI/ML has been climbing steadily since early 2024. And the rate is rising even faster, and that’s where finance leaders lose sleep and budgets break.

Key Takeaways

Actionable Guidance

It’s not all doom and gloom. There are action items you can pursue to stem the flow until you have a more established process in place. The AI Era report found that code optimization delivers the highest efficiency returns of any cost-cutting tactic yet is the least used, at just 29% adoption

Commercial levers (i.e. commitment discounts, right-sizing, enterprise negotiations) are well-understood and widely deployed. Engineering leverage isn’t. In AI-heavy environments, that imbalance is expensive.

February’s data reinforces why. If AI spend is accelerating at the rate the billing data shows, commercial levers alone won’t bend the curve. The efficiency opportunity lives in the code, prompt, and pipeline architecture, not the contract.

Five things you can do:

1. Separate your AI cost visibility from your cloud cost visibility

Why: 78% of organizations fold AI costs into overall cloud costs. That’s fine if you have the granularity to distinguish what’s driving spend within that total. Most don’t. The report’s top-ranked AI cost challenge: lack of visibility, cited by 60% in their top three. 

Do this: Track token volume, model usage, and inference costs as distinct dimensions — not a subset of compute. If your AI costs only appear as EC2, S3, or “Other,” you’re diagnosing from a blurred image. 

Result: Visibility that actually maps to what’s driving the bill.

2. Treat the storage-database rotation as an architectural signal, not a line item

Why: Storage hit a new high (11.17%), databases a new low (10.60%) — the first time storage has exceeded databases in this dataset. Embedding storage, vector retrieval layers, and training artifact accumulation are structural, not seasonal. 

Do this: Audit what’s living in storage that used to live in databases, and whether it’s being managed with the same rigor. Implement lifecycle policies for embeddings, training artifacts, and retrieval indexes — not just raw data. 

Result: Storage costs that grow with intent, not by default accumulation.

3. Build cost-to-price alignment before you need it

Why: The report found 84% of organizations price AI into their products but only 43% track costs by customer. That’s pricing on aggregate trends, not unit economics. As AI spend accelerates, the margin exposure compounds. 

Do this: Start with one product line or customer segment. Map the AI cost inputs — inference calls, token volume, retrieval, storage — to the revenue they generate. Even rough attribution is better than none. 

Result: Pricing decisions grounded in data, not approximation.

4. Apply engineering efficiency to inference, not just infrastructure 

Why: Code optimization is the highest-return cost lever and the least-used one. In AI systems, prompt efficiency, context window defaults, caching, and batching can cut inference costs significantly without touching contracts or architecture. 

Do this: Treat prompt and context size like payload optimization. Add response caching for repeated query patterns. Benchmark cost per feature, not cost per model. 

Result: Lower run-rate without slowing delivery velocity.

5. Set rate-of-change alerts, not just budget thresholds

Why: A third of organizations discover AI cost overages after the invoice. Budget thresholds catch absolute spend. Rate-of-change alerts catch acceleration — which is where February’s story lives. 

Do this: Flag any category showing >15% MoM growth for review, regardless of absolute size. AI/ML’s jump from 2.76% to 4.01% in a single month is exactly the signal a threshold-only system would miss until it’s too late. 

Result: Earlier intervention, fewer end-of-quarter surprises.

Your Takeaway For This Month

The FinOps in the AI Era report documented the paradox. February’s billing data demonstrates it.

AI spend is accelerating — across the aggregate, across the median org, across the data layer. The efficiency playbook that worked for cloud is showing its age against AI’s consumption model, fragmented infrastructure, and attribution complexity.

It’s not about whether you have AI or not. It’s about whether you can see AI costs clearly enough to make good decisions. You can’t watch the needle move without knowing what’s influencing its movement.

The trajectory is clear: AI spend is accelerating, not stabilizing. Close the gap between the bill and your visibility before that gap closes your business.

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 April 14, 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.