AI’s next big challenge? Making the money add up

After years of heavy investment in infrastructure, talent and model development, major AI providers are under pressure to show how those costs will be recovered.

As the industry shifts from scale to sustainability, consumers may face more upfront costs through subscriptions, usage-based pricing, bundled services or advertising-supported tools.

For founders, investors and technology leaders, the question is clear: who pays for the next phase of AI growth - and what does that mean for trust, access and innovation?

From investment boom to profitability pressure

The amount of investment into developing AI has been staggering. According to the 2026 Stanford AI Index Report, corporate investment between 2020 and 2025 was nearly $2trn, with AI capital expenditure in 2026 alone at $820bn. 

The various IPOs that have and are taking place this year - OpenAI and Anthropic included - suggest that capital is now needed to continue the development of AI. It also suggests profitability will become increasingly important after years of more research-focussed goals.

The future of AI pricing

Consumers currently enjoy access to AI models without necessarily having to pay for these services. Of course, this has been meant to drive adoption so that future payment models will be able to yield sustainable revenues.

A UK Government survey from earlier this year found that 73% of the public had used AI in their day-to-day life. But most users are not paying for those services. Although this may change over time (for example: if AI becomes even more embedded into daily activities) it’s likely to have implications for future pricing.

AI pricing models at a glance

Freemium: Free basic access with limits on speed, performance, and usage. Paid upgrades unlock advanced features and higher limits. Encourages broad adoption but may not generate enough paying users to be self-sustaining.

Subscription: No free tier; users pay for access. Different pricing levels offer varying features, performance, and usage. Creates predictable revenue but may limit adoption among non-paying users.

Utility (Pay-as-you-go): Users pay only for what they consume (e.g., tokens, API calls, compute power). Aligns costs with usage and supports vendor profitability. Can become expensive for heavy users and lead to unpredictable bills.

Bundled AI: Usually included within a broader service package (e.g., mobile or broadband plans). Simplifies payment through existing bills. May reduce pricing transparency and limit upgrade/downgrade flexibility.

Advertising-Supported: Free for usersrevenue comes from advertising. Follows a model similar to search engines and social media platforms. Raises unique concerns as AI could make advertising both more effective and potentially more harmful.

To add ads or not?

Monetising a platform without directly charging, or using advertising, is common enough. When it comes to AI platforms, it’s an approach that is increasingly being used.

Google’s AI Overview and Microsoft’s Copilot display ads with their query responses. OpenAI has been piloting ads in ChatGPT since February and recently OpenAI made its debut at the world’s top advertising event in Cannes, pitching its advertising business. 

Anthropic is a notable exception, making a commitment that Claude will remain completely ad-free (although Netflix had a longstanding commitment not to do so on their platform, until they did).

Perplexity, the AI-powered answer engine, is a particularly interesting case. It began using ads in 2024 but earlier this year abandoned them, citing a perception by users of a loss in credibility and a view that answers could lack integrity.

And therein lies the challenge for AI platforms. They can collect a lot of information about a user to enable the curation of deeply personalised advertisements. 

What is different is that consumers use AI to provide focused responses and to support and execute activities. Adverts, even if clearly marked as such, become part of the AI’s output, something that is unique compared to other platforms.

There are two key challenges: the first is whether users can easily distinguish between AI outputs and adverts, even if the latter are marked as such. Some early testing suggests this isn't the case.

The second is perception, as Perplexity discovered. The use of adverts can lead to users distrusting AI outputs, with a presumption of bias towards advertisers in responses.

The cost of intelligence

AI’s next challenge is not just building better models but working out who pays for them. As providers seek to recover investments, consumers are likely to face a mix of subscriptions, bundled services, advertising and new payment models.

The test will be whether companies can monetise AI without eroding user trust. Long-term success may depend as much on business model innovation as technological progress.

Supporting the innovators

At NatWest we’re passionate about providing the specific connections and expertise that founders and investors need to thrive.

Find out more: Venture Banking | NatWest Business 

The views expressed in this Sponsor article are the author's own and do not necessarily represent those of Cambridge Tech Week.