HSBC’s recent analysis of the financial challenge facing OpenAI shows how massive the scale of the company’s thinking is. It already claims revenues of $20 billion. It has committed to $1.4 trillion in building out the new data centers that will feed its ChatGPT interface. And even if it can generate $200 billion-plus in revenues by 2030, it will still need a further $207 billion in funding to survive.
Those are massive sums.
But a dozen or so AI insiders who talked to Fortunerecently at Web Summit in Lisbon described a different future for AI. That future, they say, is characterized by much smaller AI operations often revolving around AI “agents” that perform specialized, niche tasks, and thus do not need the gargantuan large-language models that underpin OpenAI, or Google’s Gemini, or Anthropic’s Claude.
“Their valuation is based on, you know, bigger is better, which is not necessarily the case,” Babak Hodjat, chief AI officer at Cognizant told Fortune.
“We do use large language models. We don’t need the biggest ones. There’s a threshold at which point a large language model is able to follow instructions in a limited domain, and is able to use tools and actually communicate with other agents,” he said. “If that threshold is passed, that’s sufficient.”
For example, when DeepSeek brought out a new model last January, it triggered a selloff in tech stocks because it reportedly cost only a few million to develop. It was also running on a model a lot smaller than OpenAI’s ChatGPT but was comparably capable.
“A 17 billion-parameter DeepSeek model was better than ChatGPT 3.5,” Hodjat said. “To put that into perspective, GPT 3.5 was more than 400 billion parameters and had to be run in a data center. A 17 billion parameter model can run on your MacBook. That’s the difference, and that’s the trend.”
A number of companies are orienting their services around AI agents or apps, on the assumption that users will want specific apps to do specific things. Superhuman—formerly Grammarly—runs an app store full of “AI agents that can sit in-browser or in any of the thousands of apps where Grammarly already has permission to run,” according to CEO Shishir Mehrotra.
At Mozilla, CEO Laura Chambers has a similar strategy for the Firefox browser. “We have a few AI features, like a ‘shake to summarize’ feature, mobile smart tab grouping, link previews, translations that all use AI. What we do with them is that we run them all locally, so the data never leaves your device. It isn’t shared with the models, it isn’t shared with the LLMs. We also have a little slideout where you can choose your own model that you want to work with and use AI in that way,” she said.
At chipmaker ARM, head of strategy/CMO Ami Badani told Fortune the company was model-agnostic. “What we do is we create custom extensions on top of the LLM for very specific use cases. Because, obviously, those use cases did vary quite dramatically from company to company,” she said.
This approach—highly focused AI agents run like separate businesses—stands in contrast to the massive, general-purpose AI platforms. In the future, one source asked Fortune, will you use ChatGPT to book a hotel room that fits your specific needs—perhaps you want a room with a bathtub instead of a shower, or a view facing west—or would you use a specialized agent that has a mile-deep database beneath it that only contains hotel data?
This approach is attracting serious investment money. IBM Ventures, a $500 million AI-focused venture fund, has invested in some decidedly unglamorous AI efforts that fill obscure enterprise niches. One of those investments is in a company named Not Diamond. This startup noticed that 85% of companies that use AI use more than one AI model. Some models are better than others at different tasks, so choosing the right model for the right task can become an important strategic choice for a company. Not Diamond makes a “model-router,” which automatically sends your task to the best model.
“You need someone to help you figure that out. We at IBM believe in a fit-for-purpose model strategy, meaning you need the right model for the right workload. When you have a model router that’s able to help you do that, it makes a huge difference,” Emily Fontaine, IBM’s venture chief, told Fortune.
