Your Cloud Economics Pulse For February 2026

Welcome to February’s Cloud Economics Pulse, CloudZero’s monthly look at cloud spend as AI moves from experiment to expectation.

Last month, we closed out 2025 with a settling: provider shares locked in, compute softened, and AI claimed more of the mix (big surprise there). 

January confirmed those patterns weren’t year-end hustle and bustle. They signify a new baseline.

Also, the Big Three (AWS, GCP, Azure) barely moved. They’re as entrenched as can be. Compute hit a two-year low while AI/ML hit a new high. And the gap between what shows up as “AI spend” and what AI actually costs keeps widening.

In this month’s Pulse, we take a look at when (and why) the real work shifts from choosing platforms to understanding what you’re running within them.

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.

Main Highlights For February 2026

1. Cost By Provider

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

As we mentioned in the opening hook, January’s provider mix tells a story of settling rather than shifting. The Big Three held steadfast for the most part. The exciting stuff is in the supporting cast after that.

Still, because of who they are, the Big Three need a quick look.

First, AWS dipped slightly to 67.37%, down 32 basis points from December. Essentially, a mere fraction, nothing too dramatic. When a provider holds at 67%-68% in our network data for a full year and more, that shows us they’re the big kid on the block for compute, data, and AI workloads. 

Meanwhile, Microsoft’s Azure showed a minor uptick to 11.37%, its first MoM gain since October. Is it decisive? No. But it does interrupt a multi-quarter pattern of erosion, which makes it notable even if not yet meaningful.

And finally, Google Cloud Platform (GCP) continues in its humble consistency, relatively unchanged at 6.82% from December’s 6.83%. It is worth noting that it’s more than a full percentage point higher than it was in January 2025 (5.69%), and the December/January percentages show its highest share of provider spend since it hit 6.98% in August 2024.

Now let’s look at the prime movers – they may be small players in the big race, but their shifts are worth looking at. Datadog was the largest shapeshifter, jumping nearly half a percentage point to 2.67% in January from December’s 2.23%. Whether driven by annual contract renewals, expanded observability footprints, or both, observability spend is becoming harder to ignore.

AWS Marketplace went in the other direction, falling to 2.67% from December’s 3.33%. It’s probably seasonal with EOY procurement surges now winding as we enter Q1. Still, AWS Marketplace is growing as a piece of the spend pie, surging to 2.76% in August 2025 after years of sub-2.0 percentages, and staying above 2.4% since then. Consolidation of third-party tooling into AWS continues.

GCP Marketplace, meanwhile, jumped to 1.01% in January from 0.68% in December. It remains a volatile category, but it’s something we should watch precisely because of its competition at AWS Marketplace.

Overall? The start of the year looks like one of consolidation. There’s a stable orbit of the big three planets, with marketplace absorbing third-party sprawl and observability growing slowly. 

Key Takeaways

2. Cost By Service Category

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

Unlike the Provider category, we’re seeing continued rebalancing in service share that was a highlight of late 2025. Compute is still softening, the data layer is becoming more durable, and AI/ML of course is continuing its upward trajectory.

Let’s go through the highlights.

Compute fell to 48.06%, down a third of a percentage point from December. That’s also its lowest share in more than two years. This doesn’t mean compute usage is shrinking. It just indicates that the overall system is growing in complexity. In short, the stack is broadening, not contracting.

AI/ML hit a new high, but we’ll get into that in detail in the next section. 

Meanwhile, Storage held elevated at 10.63%, up 27 basis points MoM from December. The step change that began in September 2025 is proving to be a durable one — Storage hasn’t dipped below 10% of total service share since. Storage entails data retention, retrieval layers, and embedding, and it’s becoming more structural.

Databases slipped to 11.15%, down from December’s 11.84%. This gradual compression has been steady after peaking near 13% in early 2024. This isn’t to suggest a collapse — databases remain essential, but growth has plateaued.

The “Other” category, a myriad of unassigned categories, held at 16.48%. This is a combination of container orchestration, platform overlays, management services, etc., which all remain a meaningful slice (or, let’s admit, are not adequately tagged in our network). This area isn’t growing or shrinking in any meaningful way.

This is minimal, but it’s worth attention: Security edged up to 1.31%, the latest update of a gradual climb throughout 2025. It’s nothing sexy, but it is persistent.

The big story for the start of 2026 is simple: AI/ML and storage are claiming more of the mix, leading to a compression of compute share. Data, storage, inferences are all expanding, and it’s potentially the sign of a new cost structure taking shape.

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:

Let’s face it; AI/ML is what you came here for. It’s growing in prominence everywhere, and if that’s what you’re seeing in your own spend, you’re not alone. 

In short: January saw a new record for AI/ML spend and that acceleration isn’t slowing down.

Let’s crunch the numbers: the average reached 2.67% for January, up from December’s 2.54%. That’s the fifth straight month of growth, starting with October’s 2.26% and continuing through January. (And, if you ask us, we don’t think that’s going to change anytime soon.)

We don’t just like to look at averages because that can easily be skewed by a few big players. Having the median in front of us gives us sharper perspective into what’s really happening with AI/ML.

So — median AI/ML spend rose to 0.63%, up from 0.57% in December. The median has more than tripled over the past year (0.18% in January 2025 to 0.63% now). This tells us something we’ve already repeated in past Pulse reports: AI adoption isn’t just deepening among heavy users. It’s spreading across the middle of the market.

Also, the gap between the average and the median continues to narrow. In January last year, the average was 8.6x the median (1.55% vs. 0.18%). In January 2026, that lowered to 4.2x (2.67% vs. 0.63%). The median is catching up to the average, meaning AI/ML spend is becoming far more of a mainstay across sectors, industries, company sizes, etc. than nestled in a few large companies with undue influence on our dataset.

The compounding growth pattern also tells us that AI/ML is entering production more than experimentation, including inference workloads, RAG pipelines, embedded AI features all generating steady and predictable cost growth.

But wait, you ask, what gives with this unusually low number? If you’ve been paying attention to the AI/ML landscape, you’ll know that industry surveys often report AI as 30% or more of total IT budgets. This metric we’re tracking is a little different. It tracks explicitly attributed AI/ML spend in cloud bills. Recognized, dedicated services like SageMaker, Bedrock, Vertex AI, and direct API costs like OpenAI, Anthropic, and Gemini are all clearly categorized AI/ML spend.

But what isn’t being captured is the AI costs that are folded into everything else in the cloud stack. S3 buckets storing training data, RDS clusters holding vector embeddings, and networking for model serving. All that stuff shows up as compute, storage, and databases, not AI/ML.

This makes the 2.67% of AI/ML a floor, not a ceiling. The real AI-driven share of cloud spend in this report is unquestionably higher, but many organizations can’t see it because the attribution layer for AI/ML is in its nascent stages. AI is so baked into modern architectures that isolating its true cost requires deeply intentional tagging, allocation, and instrumentation that only a few teams and companies have built.

So, the growth in AI/ML percentages we see (both median and average) tells us two stories: first, yes, AI/ML is climbing as a portion of overall spend. And second, AI/ML unit economics is starting to emerge. Both are absolutely happening at once.

Key Takeaways

Actionable Guidance

What we see here points to a theme: consolidation is complete, rebalancing is underway, and the organizations gaining clarity are the ones instrumenting for the stack they’re actually running right now, not the one they planned for a year ago.

So, what should you do as an org to survive these rough seas and get ahead of your competition?

1. Audit your real AI cost footprint — not just the AI/ML line item.

Why: Don’t let that small number fool you. The 2.67% you see on the bill is the floor. It’s only what you can see right now. AI costs hide in compute (inference instances), storage (embeddings, training data), databases (vector stores), and the data transfer you never see (model serving, retrieval calls, API round-trips). Frankly, most teams are blind to a huge portion of their actual AI spend.

Do this: Map the infrastructure supporting each AI feature. Track as much as you can at the granular level: your instances, storage buckets, database clusters, networking. Tag or allocate them to AI workloads, even if the service category says “compute.”

Desired result: Visibility into what AI actually costs, not just what’s explicitly labeled.

2. Treat observability as a cost category, not overhead.

Why: Datadog jumped 44 bps in a single month to match AWS Marketplace. Observability is no longer a rounding error. It’s now becoming a top-five line item for many organizations.

Do this: Review observability contracts and consumption patterns quarterly. Correlate spend growth with actual coverage expansion. Ask: are we paying for visibility we’re using?

Desired result: Observability spend tied to value, not sprawl.

3. Assume storage growth is permanent and plan accordingly.

Why: Storage has stayed above 10% since September. The step change isn’t reversing. Data retention, retrieval layers, and embedding storage are structural now.

Do this: Implement lifecycle policies with teeth: default retention windows, automated tiering, deletion schedules for training artifacts and stale embeddings. Treat storage like compute — something that requires active management.

Desired result: Storage costs that grow with intent, not by default.

4. Stop benchmarking provider share — start optimizing within your ecosystem.

Why: The Big Three barely moved. AWS, Azure, and GCP are locked in a stable orbit. The reshuffling is over. Chasing provider arbitrage yields less than deepening commitment to the ecosystem you’ve already chosen.

Do this: Shift focus from “which provider” to “how efficiently are we using this provider.” Reserved capacity, savings plans, Marketplace consolidation, committed use discounts — the gains are inside the ecosystem now.

Desired result: Cost optimization tied to architectural reality, not theoretical multi-cloud flexibility.

5. Watch for Q1 observability and Marketplace normalization.

Why: Both Datadog and AWS Marketplace showed unusual January moves. Datadog spiked while Marketplace pulled back from December’s high. Most executives know that Q1 often brings contract resets, true-ups, and procurement normalization.

Do this: Flag any >20% MoM swings in observability, Marketplace, or SaaS-layer spend for review. Distinguish structural growth from seasonal/contractual noise.

Desired result: Fewer surprises when Q1 closes and the real run-rate becomes clear.

Your bottom line: The stack is rebalancing whether you’re watching or not. Compute’s share is compressing, AI and storage are expanding, and observability is claiming real budget. The teams that win 2026 aren’t the ones chasing provider deals — they’re the ones building attribution into their architecture so they can see where cost is actually compounding.

Your Takeaway For This Month

Providers have settled, compute is yielding share, and AI is spreading through the stack in ways the AI/ML line item doesn’t capture — yet. 

The game has changed. It’s about seeing where the action is headed and futureproofing for that, especially as AI/ML continues its upward trajectory.

In short: The bill tells you what you spent. Attribution tells you why and for what, and lets you tie costs back to outcomes.

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 March 10, 2026, and on the second Tuesday of every month.