Nearly one in three Fortune 500 companies and one in five Global 2000 enterprises are now live, paying customers of AI startups. That statistic comes not from a survey or a self-reported poll, but from proprietary data compiled by Andreessen Horowitz, drawing on internal data from portfolio companies, public records, and thousands of conversations the firm has with both startups and large enterprises. The numbers challenge the widely circulated MIT study claiming 95 percent of generative AI pilots fail to convert.
The data tells a clear story about where enterprise AI dollars are flowing, which industries are moving fastest, and where the next wave of opportunity lies for founders building for business customers.
Coding Is the Dominant Enterprise AI Use Case by an Order of Magnitude
Among all enterprise AI categories, coding tools account for the largest share of revenue by a wide margin. a16z describes coding as being an order-of-magnitude outlier even among successful categories like customer support and enterprise search. Companies like Cursor, Claude Code, and OpenAI Codex have reported explosive growth, with the majority of Fortune 500 and Global 2000 AI adoption concentrated in developer tooling.
There are structural reasons coding became the killer enterprise AI use case. Code is data-dense, text-based, precise, and verifiable. Developers can run code and immediately know whether it works, creating tight feedback loops that allow models to improve rapidly. From a business standpoint, the ROI is straightforward: companies report their best engineers are 10 to 20 times more productive with AI coding tools. Since hiring engineers is difficult and expensive, any tool that dramatically multiplies output has a clear business case.
Importantly, coding tools do not need to complete tasks end-to-end to deliver value. Finding bugs, generating boilerplate, and suggesting implementations all accelerate development even when a human remains in the loop. This makes adoption low-risk for enterprises. Engineers are also early adopters who demand best-in-class tools and can adopt them without the organizational bureaucracy that slows enterprise software rollouts in other departments.
Customer Support and Enterprise Search Are the Next Two Big Categories
Customer support AI is the second-largest category by revenue, and it operates at the opposite end of the enterprise spectrum from coding. Support teams are high-volume, high-turnover functions that rely on standardized operating procedures. AI agents can model themselves on these SOPs, handle constrained interactions like issuing refunds or resetting passwords, and escalate when they cannot resolve an issue.
Support offers some of the clearest ROI in enterprise AI. Quantifiable metrics like tickets resolved, customer satisfaction scores, and resolution rates make A/B testing straightforward. An AI agent typically answers more tickets, resolves them faster, and maintains higher satisfaction scores than human-only teams, all at lower cost. Most enterprises already outsource support to BPOs, so adopting AI agents involves limited organizational change management.
Companies like Decagon and Sierra have grown rapidly in this space, alongside vertical specialists like Salient in hospitality and HappyRobot in logistics. The category does not require 100 percent accuracy to be useful, since natural off-ramps to human agents exist, making sales cycles faster and pilots low-risk.
Enterprise search rounds out the top three categories. ChatGPT's own primary use case is search, so much of OpenAI's enterprise revenue is likely baked into this category. Standalone startups like Glean have thrived by helping employees locate information across disparate internal systems. Vertical search players have also emerged, such as Harvey in legal and OpenEvidence in medical search, each building a core offering around industry-specific information retrieval.
Legal and Healthcare Are Surprising Early Adopters
While tech companies unsurprisingly dominate AI adoption, two industries historically resistant to software have become eager buyers. Legal was a surprising first mover. Traditional enterprise software provided limited value to lawyers because static workflow tools did not accelerate unstructured, nuanced work. AI changes this by parsing dense legal text, reasoning over large document sets, and drafting responses. Harvey reached approximately $200 million in annualized recurring revenue within three years of founding, and Eve, which specializes in plaintiff law, crossed 450 customers and hit a $1 billion valuation.
Healthcare has responded to AI in a way it never did for traditional software. Companies like Abridge for medical scribing, OpenEvidence for clinical search, and Tennr for healthcare back-office automation have grown rapidly. AI bypasses the dominance of electronic health record systems like Epic by taking on discrete labor-intensive tasks rather than requiring a rip-and-replace. Medical scribing, clinical search, and revenue cycle automation have proven to be entry points that generate real revenue without requiring hospitals to overhaul their existing software stack.
What This Means for Founders
For founders deciding where to build, the a16z data provides a concrete map of proven enterprise demand. Coding tools, customer support automation, and enterprise search are categories with demonstrated willingness to pay. But the opportunity extends beyond these three. Healthcare and legal have shown that traditional slow-adopting industries will move quickly when AI addresses real workflow pain. Manufacturing, logistics, and retail remain largely untapped, with AI penetration still concentrated in tech-forward industries.
The data also reveals that enterprise AI adoption is still in its early innings. Nineteen percent of Global 2000 companies being active customers means more than 80 percent are not. For startups building for enterprise, the path is clear. Pick a use case with tight feedback loops, clear ROI metrics, and low organizational change friction. Code, support, and search fit that mold. The next wave will be industries where AI can replace or augment high-value human labor without requiring system overhauls.
