Enterprises are continuing to increase their investment in AI, but this does not mean that AI startups have already obtained stable and sustainable long-term revenue. The latest research shows that while corporate clients are expanding their AI budgets, they are also re-evaluating suppliers more frequently. The inertia of multi-year contracts, which was common in the past SaaS era, is weakening in current AI procurement practices.

Enterprise budgets are still expanding.

A survey of 150 IT practitioners by Madrona shows that 74% of the respondents plan to increase their AI budget in the next 12 months, while the rest intend to maintain their current spending levels. At the same time, less than half of the AI pilot projects will ultimately be able to enter the formal production environment.

Market research firm IDC predicts that corporate technology spending will reach $4.25 trillion in 2026, with the main increase coming from AI. For startups, this means that customers are still willing to try new products, and opportunities for piloting and purchasing have not disappeared.

Renewal stability is declining.

What is more noteworthy is that even after completing the deployment, companies may not choose to retain the same supplier in the long term. Research by Madrona shows that 77% of companies re-evaluate their AI suppliers every six months, or even on a rolling basis.

This procurement rhythm is significantly different from that of traditional enterprises SaaS. In the past, contracts and higher switching costs usually led to greater loyalty; however, in the AI software industry, companies find it easier to replace tools and are more willing to continuously compare the effects and prices of different products.

This means that even if AI startups advance their products from pilot phases to formal adoption, the related revenue may not be as certain as it used to be. Many companies disclose ARR growth to the public, and therefore face a higher risk of volatility.

Charging based on results is more popular.

Another venture capital firm, Andreessen Horowitz, conducted a survey among 50 technology-based AI purchasers and found that more than half of the respondents preferred for AI products to be charged based on work results, rather than on usage such as token.

What enterprises are more concerned about is how much work AI has actually completed, such as how many reports have been processed, how many tickets have been closed, and how many sales leads have been generated. If the billing method can directly correspond to these results, it will be easier for suppliers to demonstrate the value of their products, which is also more conducive to contract renewals and expansion.