Companies Are Spending 5x More on AI Now

Menlo’s estimate of annual enterprise AI spending rose from $1.7 billion in 2023 to $8.5 billion in 2024 and $33 billion in 2025, and is projected to reach almost $177 billion in 2026. That would make spending more than five times higher in 2026 than in 2025; the host said both total spending and its growth rate were accelerating. The estimates draw on a mid-September survey of about 400 US enterprise decision makers responsible for roughly $4 billion in AI spending. The host cautioned that Menlo invests in AI companies, including Anthropic, and is not an unbiased observer.

Average AI spending per enterprise that uses AI increased 4.7-fold from 2025 to 2026, while the share of enterprises reporting AI use rose from 55% to 65%. The remaining 35% said they were not using AI. Companies also spent substantially more than they had planned: average plans called for spending about 1.9 times the 2025 level, compared with actual average spending 4.7 times higher. Menlo sees this as evidence of future demand, but the host noted that spending concentrated among a few large customers could indicate vulnerability or precede broader adoption.

The shift toward AI agents is changing how businesses buy software: spending is moving from per-seat pricing toward per-token usage, which scales with work rather than headcount. The host argued that traditional per-seat budgets do not fit this model, and that cloud-compute budgets are only an imperfect comparison; AI may need its own budget category. Menlo found that 55% of AI spending came from new budgets and 21% from existing software budgets, but the host said these figures do not settle whether agents or headless software will disrupt SaaS. SaaS application market growth slowed from 13% in 2025 to 11% in 2026.

Frontier models’ share of AI spending rose from 35% to 63%, even as small-model prices fell by more than 80%. Surveyed buyers shifted toward more expensive models, suggesting that capability outweighed cost for many purchasers. Anthropic’s share of business API spending increased from 12% in 2023 to 55% in 2026; the speaker linked its rise to Opus becoming a default for long-horizon agentic coding. OpenAI’s share fell from 50% in 2023 to 27% in 2025, then recovered to 31% in 2026, which Menlo attributed to newer models and a stronger focus on business use cases. Google’s share fell from 21% in 2025 to 9% in 2026.

Open-source models represented 4% of AI spending but 19% of tokens in 2026, compared with 6% of spending and 11% of tokens in 2025. The speaker viewed token share as more meaningful over the medium to long term and expected open-source models to gain further share, while noting that spending could remain concentrated in expensive, state-of-the-art closed models. Kimi accounted for 61% of open-source spending. The gap between token use and spending highlights that open-source models can account for substantial usage without matching closed models’ share of dollars.

Agentic coding applications grew 7.7-fold. Combined revenue for Codex and Claude Code rose from $1 billion at the end of 2025 to $24 billion at the end of 2026, while other coding platforms grew from $2.6 billion to $7.8 billion. Financial-services applications grew eightfold from a much smaller base of about $800 million, compared with $34 billion for coding apps. The speaker suggested growth favored areas where results are verifiable and agentic coding capabilities can be applied.

A METER report using internal information about three unreleased Anthropic models found that the pace of model improvement had roughly doubled over the past year, primarily because of AI assistance rather than fully autonomous research. METER warned that autonomous contributions could become more significant soon and that rapid acceleration could produce risky capabilities. Anthropic’s Sam McCandlish called the report an independent external view of progress while acknowledging criticism of redacted evidence and saying Anthropic planned a supplemental report.