Your AI Economics Pulse for September 2026

It’s still the same (old) story: AI’s share of the cloud bill goes up, nearly everywhere, nearly every month. But August complicates it.

The median company’s AI share saw its smallest gain since last December and roughly one fifth of July’s move. If you read that in isolation, you’d think the AI spend boom finally hit its limit.

It didn’t. The 75th percentile crossed a tenth of the cloud bill for the first time, and the number of organizations at that level posted its biggest monthly jump on record.

Adoption is spreading outwards and concentrating upwards at the same time, while the median is sitting in the area between those two changes. Which means the number you’d quote in a board deck this month could well be the one least likely to describe your own company.

Let’s start with share of AI spend, then intensity and breadth.

Download Now

Get your own copy of the AI Economics Pulse now

The September Pulse digs into where AI spend is accelerating, who’s pulling back, and what that means for AI ROI.

Share: the top quartile crossed a tenth of its cloud bill

Here, we’re tracking AI spend as a share of each organization’s total cloud bill, reported as the panel median with the 25th and 75th percentiles around it. The median is the midpoint of the panel; the percentiles show how far apart the light and heavy spenders are. Every figure is relative to a company’s own bill, so it reads the same at any size.

As noted above, the median company in our panel spent 2.66% of its total cloud bill on AI in August, up from 2.61% in July. That is the 13th consecutive monthly increase in this metric, and the smallest of the 13.

Plus, if you step back and look at the 75th percentile, another element emerges. That line jumped from 9.63% to 11.12% in a single month, a gain of nearly 1.5 percentage points. A quarter of the panel now sits above a tenth of its cloud spend, at 4.2 times the median. That’s the widest median-to-75th percentile multiple since we began publishing this index.

At the other end, the 25th percentile grew from 0.11% to 0.17%. Small numbers, but it’s still a 51% increase and the largest absolute gain in our records for that line. The bottom quarter of the panel is still measuring AI spend in fractions of a percent.

We’ve mentioned this in past Pulses; this is spend we can cleanly attribute to AI as a category. Tying it to what it produced is a separate exercise, and one that takes deliberate instrumentation rather than a better invoice.

So, treat this category as a floor, and see why AI ROI is a question of allocation.

Intensity: the 10% tier grew faster than it ever has

Here, we’re tracking something different: not how much AI costs the typical company, but how many companies have crossed given thresholds. We count the share of organizations in our panel with at least 1%, 5%, 10%, or 25% of total cloud spend attributed to AI. Each threshold is relative to that company’s own bill, so the tiers read the same whether you run $50 million or $500 million through the cloud.

This is where August moved. The 10% tier jumped to 28.2% from 23.9%. That 4.3 percentage point gain is the largest single-month move this tier has recorded, beating the 3.7 points it added in April.

The 5% tier crossed 40% for the first time, reaching 40.6% from 39.6% in July.

The 25% tier fell from 9.1% to 8.2%, its first decline since January after six consecutive months of gains.

Last month we called the 25% cohort the one worth watching. We have two potential explanations for the drop.

The first is rationalization. We surmise that organizations reaching a quarter of their cloud budget in AI hit the point where someone senior asked what it returned, and the answer was thin. Gartner’s September 1 survey of 1,303 organizations found 11% cannot say what their function spent on AI in 2025 at all, and a company that can’t see the number is a company that can’t defend it. On this reading, the 25% tier is where AI spend meets AI governance, and the decline is a healthy correction.

The second is arithmetic. Intensity is a ratio, and a company can drop out of the 25% tier either by cutting AI spend or by growing everything else faster. Some of these organizations run large AI workloads on infrastructure that grows alongside them: storage, data movement, and the databases holding embeddings. If total cloud spend outran AI spend for a month, the ratio actually falls while the AI bill keeps climbing.

We can’t separate the two from one month of data. The 10% tier gained 4.3 points in the same month, so the 10-to-25% band swelled. A band can swell from companies climbing into it or from companies dropping into it, and the panel doesn’t distinguish those. We’ll report which way it resolves rather than guess now.

Breadth: two-thirds of the panel is now spending on AI

You’ll notice in past Pulses that we were tracking AI spend growth. That measured how fast existing AI budgets are compounding, but this month we want to look at breadth, which measures how many companies are spending on AI at all. Right now, at this stage of the market, that’s a more useful metric to monitor.

Here, we’re tracking the share of organizations in our panel spending at least $1,000 on AI in a given month. Unlike the other two readings, this one is an absolute dollar floor rather than a percentage of anything. It doesn’t care how big your cloud bill is. The floor is deliberately low: it is the line between running a pilot and running nothing.

In August, 67.7% of the panel cleared the $1,000 baseline, up two percentage points from 65.7% in July, and the 10th consecutive MoM rise. 

A third of the panel is still below the line. Not below 10%, or below 1%, but rather, below a thousand dollars a month. That means whatever AI has done to cloud budgets so far, it’s only done that to two-thirds of the market. The other third is just getting started or hasn’t started at any scale. At the pace of the last few months, three quarters of the panel clears the line before spring.

Which shifts the question: when most of the market has an AI budget, having one stops being the differentiator.

Capital allocation: who can show what the spending bought

Two-thirds of our panel now has an AI budget. That makes the interesting question less about who is spending and more about who can account for it. 

This summer, the public markets ran a live experiment on exactly that, on the companies selling the compute. Alphabet and Amazon both reported in late July, eight days apart, with similar spending postures and opposite outcomes.

Two companies, same spend, opposite verdicts

First, Alphabet. Google Cloud revenue grew 82% to $24.77 billion in the second quarter, capital expenditure hit $44.9 billion, and management raised full-year capex guidance to a range of $195 billion to $205 billion, according to the company’s July 22 release and the earnings call that followed. The stock closed down 7.1% the next day.

Eight days later, Amazon got the reverse. AWS revenue grew 37% to $42.2 billion, its fastest in 18 quarters, against a reported backlog of $496 billion. Andy Jassy raised capex guidance to roughly $220 billion and told analysts the company still would not have enough capacity to meet 2026 demand, as Fortune reported that day. The stock rose more than 9% after hours.

The market did not punish the spending. It sorted on whether the revenue was already committed. Amazon’s backlog was revenue contracted before the capacity was built. Alphabet grew faster, at 82%, but spent $44.9 billion in the quarter against $24.77 billion of cloud revenue and free cash flow that had turned negative. Combined 2026 capex guidance across the four largest hyperscalers now runs between $720 billion and $745 billion, according to TMT Finance’s August 18 analysis of those earnings calls, and every one of them faces the same test.

Growth was not the variable. Proof of forward demand was. 

The market settled for two adjacent numbers

Neither company published anything as precise as a return on those dollars. The capex figure is company-wide, the revenue is segment-level, and the two are only adjacent on the page.

That was enough for investors to reach a verdict. It is also more than most companies can assemble about their own AI spend. CloudZero’s June survey of 260 finance leaders found that 87% need to tie AI spend to business outcomes within the year, and 22% can do it today. 

The same skew, in everyone else’s data

The unevenness shows up at market scale too. Gartner’s July 27 forecast put 2026 worldwide IT spending at $6.37 trillion, up 14.2%, with data center systems growing 62.5% to $822 billion. Distinguished VP Analyst John-David Lovelock put it directly: “Despite the strong growth in spending, this is not a ‘rising tide lifts all boats’ market trend.”

Vendor data finds the same skew. Ramp, analyzing corporate-card spend across U.S. businesses, reported in August that the top 1% of firms spent a median of $7,400 per employee on AI in July, against $11.95 for the median firm. Card data captures what companies buy directly from model providers. It cannot see the AI spend that arrives inside a cloud bill, i.e. Bedrock line items, GPU instances, inference running on infrastructure the company already pays for. That layer is what our panel measures, and it is skewed the same way: a 75th percentile at 11.12% of the cloud bill against a median of 2.66%.

Neither dataset can see whether the spending is working. Writing in Mostly Metrics on September 1, CJ Gustafson walked through a $61,656 monthly model invoice against its usage export and found five separate leakage patterns: stale model versions, missing prompt caching, standard rates where batch pricing applied, oversized models on simple work, and one job eating a disproportionate share. The invoice itself had four lines on it.

The spend is easy to total and hard to explain.

That gap is the problem, not the spend behind it. The companies in the 22% didn’t get there with better bills. They’re getting there by allocating spend against the units their business actually sells, then reading margin off it. That capability is available to anyone willing to do it, which is why the 87% gap reads as a lag rather than a limit.

What this means for September

If you benchmark yourself against the median this month, you are benchmarking against the least informative number in this report. Check your tier instead. Above 10% of your cloud bill puts you in the fastest growing cohort in our panel, past the point where AI is a line item somebody watches rather than one somebody owns.

Your tier tells you how much. Whether it was worth it is a separate question, and the invoice will not answer it. Alphabet and Amazon each got a verdict from the market in an afternoon, on two numbers printed side by side. Most AI spend has no such pairing.

That pairing is buildable. Allocate AI spend to the customers and workloads it serves, and cost per outcome falls out of the allocation. Measuring AI ROI turns “was it worth it?” into a number you can put in front of a board, which is the difference between defending an AI budget next quarter and growing one.

See where you land

CloudZero’s AI Benchmark Tool shows where your AI spend sits against your peers, so you can walk into the next board meeting with a number instead of a shrug.

Frequently Asked Questions

Author Spotlight

Keith MacKenzie

Keith is CloudZero's Content Marketing Manager with a specialty in topics around AI & cloud spend. 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.