Enterprise LLM Spend Hits $8.4B; Anthropic Overtakes OpenAI in Market Share
Enterprise spending on large language models (LLMs) has seen a dramatic increase, more than doubling in just six months from $3.5 billion in late 2024 to $8.4 billion by mid-2025. This significant surge is detailed in the “2025 Mid-Year LLM Market Update” report, recently published by Menlo Ventures. The report, which surveyed 150 technical leaders across AI startups and large enterprises, highlights a rapidly maturing market where inference has become the primary workload, performance dictates vendor choice, and a notable shift in market leadership is underway.
A key finding from the Menlo Ventures data indicates a significant change in the competitive landscape among LLM providers. OpenAI, which dominated the enterprise LLM market through 2023 with 50% of usage, has seen its share fall to 25%. In contrast, Anthropic has emerged as the new market leader, capturing 32% of enterprise usage across production workloads. Google has also made substantial gains, securing 20% of the market, largely propelled by the strong adoption of its Gemini models. Meta’s Llama holds 9%, while DeepSeek accounts for 1% of LLM API usage.
Tim Tully, a partner at Menlo Ventures, commented on this shift: “Some might be surprised to see Anthropic overtake OpenAI, given its first-mover advantage. But our research puts real numbers behind what we’ve heard anecdotally from the market: Teams are prioritizing real performance in production. As enterprise LLM spend crosses $8 billion, Anthropic is capturing the majority share, and Google has quickly gained ground to claim the number three spot.”
The report also outlines several other critical trends shaping the enterprise LLM sector:
Rapid Spending Growth: The doubling of enterprise LLM expenditure from $3.5 billion in November 2024 to $8.4 billion by mid-2025 reflects a significant move of AI workloads into full production environments.
Vendor Loyalty and Upgrades: Despite the dynamic market, vendor switching remains relatively rare, with only 11% of teams reporting a change in model providers over the past year. However, a substantial 66% upgraded to newer models from their existing vendors, indicating a focus on continuous improvement within established relationships. 23% made no changes or upgrades.
Dominance of Closed-Source Models: Closed-source models now power the vast majority of enterprise workloads, accounting for 87% of usage. Open-source usage has declined from 19% to 13% over the past six months, a trend attributed to widening performance gaps compared to proprietary alternatives.
Inference Overtakes Training: Inference, the process of running an LLM to generate outputs, has surpassed training as the primary compute workload. This shift is evident as 74% of startups and 49% of enterprises reported that inference accounts for most of their compute usage, a significant increase from late last year.
Looking ahead, Menlo Ventures’ report predicts that “long-horizon agents” will drive the next major evolution of the enterprise AI stack. These advanced systems are designed to autonomously tackle complex, multi-step, and open-ended tasks, such as software development, research synthesis, and operational workflows, ultimately aiming for true auto-remediation.
Derek Xiao, an investor at Menlo Ventures, emphasized the transformative potential of these emerging technologies: “Long-horizon agents represent an operating model shift. The startups building agentic infrastructure today are laying the foundation for the next generation of $10B-plus platforms. With legacy vendors lagging behind, the opportunity is massive.” While still in early deployment stages, these agentic systems are anticipated to play a central role in the forthcoming wave of enterprise AI transformation.