US companies' spending on AI services is growing much more slowly than in previous years. The Ramp AI Index, which analyses company spending based on payment data, shows an increase of 1.65 per cent for October 2026. That is not a collapse, but a clear signal: the phase in which budgets for AI tools were revised upwards almost at will appears to be over for now.

What is notable here is less the figure itself than the structure behind it. According to the analysis, a large share of revenue comes from a very small group of customers: around one per cent of companies are said to be responsible for about 80 per cent of OpenAI's and Anthropic's revenue. The broad mass of companies therefore contributes comparatively little to the business.

Why the concentration is a risk

Such an uneven distribution makes providers vulnerable. When a few large customers carry the revenue, churn or renegotiations have a disproportionately strong effect. At the same time, the pressure increases to open up new revenue sources instead of waiting for organic growth in the mid-market.

Customer support – the use of AI systems in service and ticket processes – is currently regarded as the new growth driver. However, this field may bring lower margins than the previously dominant coding business, in which development teams integrate assistance systems directly into their working environment.

What this means for coding assistants

Coding assistants – tools that suggest, explain or refactor source code – were long the flagship product of the providers: clear benefits, measurable time savings, high willingness to pay. That is precisely why additional competitors are pushing into this market. When several providers compete for the same developer teams while budgets are barely growing at the same time, price pressure arises.

For companies, this is initially good news. More choice and falling prices improve their negotiating position. Anyone buying licences for development teams today can realistically expect better terms than two years ago.

The flip side: price pressure combined with stagnating demand usually leads to market consolidation. Not every provider advertising aggressive terms today will still exist in three years or hold the same prices.

Consequences for tool selection

The decisive question thus shifts from "Which tool is currently the cheapest?" to "How expensive will a switch be?". Some points that can be checked for this:

  • Interchangeability: Can the assistant be accessed via standardised interfaces, or is it deeply embedded in proprietary formats?
  • Data storage: Where are prompts, context data and analyses stored – and can they be taken along when switching providers?
  • Process dependency: Are review, test or deployment steps firmly tied to a specific product?
  • Contract terms: Short terms cost somewhat more but preserve room for manoeuvre in a market that is still sorting itself out.

Budget planning without assuming growth

Stagnating spending does not mean that AI tools are losing their usefulness. It suggests rather that many organisations are moving from the experimentation phase to the consolidation phase: instead of continually buying new licences, they are checking which of the existing tools are actually being used.

For internal planning, this is a good occasion to analyse usage data. The relevant metric is licences actively used, not licences purchased. Where tools are barely touched after the introduction phase, the problem is usually not a technology problem but a process problem – for example, a lack of integration into existing workflows or unclear responsibilities.

Assessment for web and software projects

For ongoing projects, this above all means: AI tools should be planned as interchangeable components, not as a foundation. Anyone who builds development and operations processes in such a way that switching providers remains manageable benefits from the current price pressure without bearing the risks of market consolidation. And in budget discussions, what will count in future is less the question of how much is invested in AI than which measurable share of daily work actually runs better as a result.

Sources