AI Spend Hit $177 Billion This Year. Only 22% of It Came Out of Software Budgets. The 4th Annual State of Enterprise AI Report from Menlo Ventures,

The $177 Billion Illusion

We are seeing some massive numbers coming out of the enterprise sector lately. The latest report from Menlo Ventures claims that enterprise AI spending has rocketed to $177 billion this year. On the surface, that looks like a victory lap for anyone building in the space. If you are a founder, you might see that figure and think the floodgates have finally opened. But if you look closer at where that money is actually coming from, the picture gets a lot more complicated.

The most telling detail in the report is that only 22% of this AI spend is coming out of traditional software budgets. For those of us who have spent years selling SaaS or building infrastructure, that is a jarring statistic. It means that nearly 80% of the money flowing into AI right now is being pulled from other places—mostly experimental pools, hardware reserves, or one-time strategic allocations. It is not "business as usual" yet.

Where the Money is Actually Hiding

In the traditional software world, budgets are predictable. You have your CRM seat costs, your cloud storage fees, and your security stack. AI is currently behaving like a disruptive force that hasn't found a permanent home on the balance sheet. Because only a small fraction is coming from software budgets, we know that enterprises are treating this as a capital expenditure or a massive R&D bet rather than a standard operational cost.

For builders, this is a double-edged sword. It means there is a lot of "new" money available that isn't tied up in legacy contracts. However, it also means this spending is fragile. If a company pulls $50 million from a "strategic innovation" fund to buy GPUs or LLM credits, that money can vanish the moment the board decides the ROI isn't showing up fast enough. Software budgets are sticky; innovation budgets are volatile.

The Infrastructure Tax

Another massive takeaway from the Menlo data is that AI deployment spend now equals about 30% of public cloud revenue. A year ago, that number was sitting at a measly 6%. This tells us that the vast majority of that $177 billion isn't going to clever wrapper apps or niche productivity tools. It is going to the giants—the companies providing the compute, the chips, and the raw horsepower required to keep these models running.

If you are a founder building at the application layer, this is your biggest challenge. You are competing for a sliver of the remaining budget while the infrastructure providers take the lion's share. The "AI tax" is real, and it is currently being paid to the cloud providers and hardware manufacturers. To survive, builders have to prove that their specific application provides more value than the raw compute it consumes.

The Founder's Dilemma: Utility vs. Hype

I speak to founders every week who are trying to figure out why their sales cycles are lengthening despite the massive headlines about AI spending. The answer is in that 22% figure. Most enterprise buyers are still in the "window shopping" or "pilot" phase. They are throwing money at the problem to see what sticks, but they haven't yet integrated AI into their core operational workflows.

If you want to capture part of that 22%—the stable, recurring software budget—you have to stop selling "AI" and start selling "solved problems." A CTO might have a million dollars in a slush fund for an AI experiment, but they have a billion dollars in a recurring budget for things that make the company run. Moving your product from the experimental bucket to the essential bucket is the only way to build a sustainable business in this climate.

The Shift in Market Leadership

We are also seeing a shift in who is winning the enterprise race. Early on, it was all about the model providers. Now, the focus is shifting toward data infrastructure. You cannot have a functional AI strategy without clean, accessible data. This is why we are seeing so much movement in vector databases and data pipeline tools. The enterprise is realizing that buying a shiny new model is useless if their internal data is a mess.

For those of us in the builder community, this is a call to focus on the boring stuff. Everyone wants to build the next world-changing agent, but the companies making real money right now are the ones helping enterprises organize their data so the models actually work. The unsexy parts of the stack are where the most resilient budgets are being formed.

Why We Should Be Skeptical of the Totals

It is easy to get blinded by a $177 billion figure. But we have to ask: how much of this is actual value creation, and how much is just the cost of entry? Much of this spend is defensive. Large corporations are terrified of being left behind, so they are over-provisioning and over-spending on tokens and compute just to say they are doing something.

Eventually, the bill comes due. When these companies realize that 80% of their AI experiments aren't yielding a 10x return, the spending will contract. The companies that survive that contraction will be the ones that found a way to tap into the core software budget by providing undeniable utility. We are currently in a period of artificial inflation driven by FOMO.

What Builders Should Do Now

If you are building in this space, stop looking at the $177 billion as a sign that everything is easy. Instead, look at the 22% as your target. Your goal should be to become a line item in the permanent software budget, not a temporary experiment funded by a one-time grant.

• Focus on Integration: If your tool doesn't talk to the existing stack, it will be the first thing cut when the experiment ends.
• Watch the Margins: With infrastructure costs taking up 30% of cloud revenue, you need to be extremely disciplined about your own COGS.
• Solve for Data: Help your customers get their data ready for AI. That is a pain point that isn't going away.
• Avoid the Hype Cycle: Don't build features just because they are trending on social media. Build what a CFO will agree to pay for in three years.

The Long View

The transition from experimental spending to core operational spending is going to be painful for a lot of startups. We are going to see a lot of companies that raised money on the back of the $177 billion figure fail because they couldn't cross the chasm into the 22% of stable software budgets.

The founders who win will be the ones who treat AI as a tool, not a destination. The technology is impressive, but the economics are what matters. We are moving out of the era of "AI for AI's sake" and into the era of "AI that pays for itself." If you can't prove the latter, the massive spending numbers won't save you.

Takeaway for Builders

The enterprise is spending heavily on AI, but most of it is temporary, experimental capital. To build a lasting company, you must migrate your product from the volatile "innovation" budget into the stable "software" budget by solving core business problems rather than just providing a cool interface for a model.

Read the original at SaaStr →