Revealed: The AI Tools Startups Are Actually Spending Money On

Andreessen Horowitz's First AI Spending Report

On Thursday, Andreessen Horowitz (a16z) released its first AI Spending Report in partnership with the fintech firm Mercury. Using transaction data from Mercury, the report analyzes the top 50 AI-native application layer companies that startups are spending money on, similar to the previously published Top 100 Gen AI Consumer Apps.

a16z partners Olivia Moore and Seema Amble say the data shows companies are still adopting a range of different AI products for certain tasks — and new apps are rising and falling very quickly.

“There’s a proliferation of tools,” Amble said. “It hasn’t just coalesced around one or two in each category.”

The report also shows a lot of spending on “human augmentors” or “copilots” that can help boost productivity among the workforce, suggesting startups aren’t ready to fully shift into agentic workflows.

“As computer use becomes more of a mode and there’s more of the ability for there to be end-to-end agentic flows built, I think that shift will happen, where we’ll see fewer copilots and more end-to-end agent tools,” Amble said. “Especially as people are really eager to give them a try.”

Top AI Companies Startups Are Paying For

Unsurprisingly, the top of the list was dominated by major labs, with OpenAI taking the top slot and Anthropic following up at No. 2. Vibe-coding tools were also well represented, with Replit at No. 3 and Lovable at No. 18. Cursor landed at No. 6 and Emergent at No. 48. Cognition, which operates more enterprise-oriented coding tools like Devin and Windsurf, was at No. 34.

When a16z produced a similar list for consumer habits, Lovable ranked much higher than Replit on pure traffic alone because a lot of people were using it to create projects. But startups are not spending as much money on Lovable as they are on Replit, in part because of the lack of enterprise features. But the variety of companies on the list seems to suggest there’s room for plenty of different companies at once.

“It’s an open question going forward on vibe coding,” Moore said. “Does the space start to consolidate, and one place becomes the best platform to vibe code? Or is it the case where there’ll be four or five more vibe coding companies that are really big businesses for different types of applications? We don’t have the answer to that yet.”

Moore was also surprised to see startups adopting consumer-oriented tools like CapCut and Midjourney.

“We’re seeing that a lot of these [consumer] companies are getting yanked into enterprise faster and faster because they make such delightful consumer tools that then people adopt and use as individuals and bring into their teams and workplaces,” Moore said.

Horizontal vs. Vertical AI Applications

Horizontal applications overall made up at least 60% of the names on the list, and 40% consisted of vertical applications. The most popular vertical software companies fell into three buckets: sales, recruiting, and customer service. But the report also found AI making progress in many sectors that previous generations of startups had struggled to crack.

“What maybe previously would have been, like, service firms or consultancies are now software companies in the age of AI,” Moore said.

Amble gave Crosby Legal as an example, which can quickly review a legal contract for a user, replacing what at one point would have been a meeting with an in-house general counsel, rummaging through thoughts and research. She said right now, most of the tools are used to aid employees (like a co-pilot) in making decisions faster, rather than replacing entire workforces and talent suits with automated workers (end-to-end agents).

“As the tech gets better, and we’re actually able to build out full agent co-workers, you’ll see that mix shift more toward end-to-end agents and away from co-pilots,” she said, later adding that AI tools can do much more work, like outreach, faster than a human can.

Note-Taking Apps and Market Fragmentation

There were also a lot of note takers on the list, such as Otter.ai, Read AI, and HappyScribe — with no single option dominating. This is what Amble meant when she said there isn’t one product yet that has dominated the market; rather, startups are still picking “their own flavor” to see what tools they like best. This is also good for employees, who, with so many options, can pick what applications help them work best, rather than using a one-size-fits-all product “pushed down from the top,” Amble said.

Blurring Lines Between Consumer and Enterprise

The last big find in the report was the increasing intertwining of consumer and enterprise businesses. People are bringing the personal applications they use at home to work, and people who have started companies are using their favorite personal applications to help build their businesses. Before, there would have been a delineation between the two: a set stack for what to use while building a startup.

Amble and Moore cited Canva as an example: It’s a popular consumer app that also has a sizable enterprise audience. It took years for Canva to even add an enterprise plan. But as individual and enterprise use cases become harder to distinguish, companies are more willing to blend the two.

“Your TAM [total addressable market] is no longer one or the other, but you can sell into both,” she continued. She said companies building these products might also “professionalize” faster, meaning building out enterprise teams, like go-to-market, sales, and support, so they can start selling and accruing enterprise revenue sooner, rather than depending on individual consumers.

Rapid Evolution of the AI Landscape

Moore and Amble are expecting the list to change quickly in the coming years. Older companies are now launching AI features to stay relevant and accessible, while new entrants arrive with new ideas.

“Legacy players, legacy almost means, ‘what was 12 months ago,’” Amble said. “If we pull this again in 12 months, will the same note-taking apps even be on there? Or will there be a whole new set?”