The enterprise AI sector is enjoying significantly rising investment, even as ROI from AI falls short of the hype. That’s according to new survey data from venture capital firm Andreessen Horowitz, which also portrays a market dominated by a small number of powerful model providers, with enterprises increasingly spreading their investment across multiple platforms rather than committing to a single winner.
Based on a survey of 100 CIOs and senior technology leaders from Global 2000 companies, OpenAI remains the most widely deployed model provider in large enterprises, with roughly four out of five respondents reporting active production use.
Anthropic has emerged as the fastest-growing rival, posting notable gains in enterprise penetration over the past year, while Google continues to build share across a broad range of workloads.
Avoiding Vendor Lock-In
Rather than a winner-takes-all market, enterprises are assembling portfolios. More than 80% of surveyed companies now use three or more AI model families in testing or production, up sharply from less than 70% a year ago. CIOs say the approach reduces vendor lock-in and allows teams to match models to specific tasks, such as customer support, document generation, software development, or data analysis.
Competitive lines are most clearly drawn among use cases. OpenAI continues to dominate early, horizontal deployments, including general-purpose chatbots, enterprise knowledge search, and customer service tools.
Anthropic, by contrast, has carved out a strong position in software development and analytical workloads, areas where CIOs cite rapid improvements in model capability since late 2024. Google’s models are viewed as solid all-rounders, though their adoption in enterprise coding workflows remains comparatively limited.
Yet OpenAI’s lead in enterprise AI spending is narrowing, as Anthropic and Google capture incremental wallet share. Executives surveyed by Andreessen Horowitz expect the balance to continue tilting through 2026, even as overall spending rises for all major providers.
Average enterprise spending on large language models has climbed from roughly $4.5 million two years ago to about $7 million today, with companies forecasting another 60-plus% increase over the next year.
Challenges of AI’s ROI
One of the more interesting findings is how these spending gains coexist with muted perceptions of return on investment. While most enterprises report positive ROI from AI initiatives, the benefits fall short of the dramatic productivity gains often touted in online discourse.
CIOs attribute the gap to two factors: the difficulty of translating raw model capability into real workflows, and the continued reliance on incumbent tools that may not fully exploit newer AI advances.
Moving Toward Packaged AI Applications
That tension is especially visible in the application layer. While speculation has grown about enterprises abandoning third-party AI software in favor of building directly on models, the survey suggests the opposite trend. Companies continue to migrate toward packaged AI applications, particularly when those tools offer deep integration and domain-specific workflows. In this environment, established vendors retain a decisive advantage.
Microsoft, not surprisingly, stands out in this regard. Its enterprise footprint gives it a dominant position in AI-powered productivity and development tools, including chat and coding assistants. Nearly two-thirds of surveyed enterprises said they prefer incumbent vendors when AI solutions are available, citing trust, system integration, and purchasing simplicity.
Startups still have opportunities, executives say, but the bar is high.
Closed Systems Gaining Ground
Another notable shift is growing confidence in proprietary, closed-source models. Preference for closed systems has steadily increased, driven by faster model improvements, limited in-house AI expertise, and (perhaps a contradiction) data security concerns. Trust in leading model developers has risen in parallel, with many enterprises now comfortable hosting models directly rather than relying exclusively on cloud intermediaries.
Taken together, the data point to a market that is arguably stabilizing and definitely intensifying in competition. A small group of AI providers is pulling ahead, enterprises are spending more than expected, and experimentation is giving way to pragmatic, multi-model strategies. Still though, with a technology that’s evolving as unpredictably as AI, the enterprise AI race remains in early innings.
