Even as big tech companies lead the cloud market, startups specialising in AI cloud services appear to keep entering the fray.

After ChatGPT emerged, GPU supply has increasingly failed to keep pace with demand. AI-focused cloud providers such as CoreWeave, Crusoe and Lambda, which secured GPU capacity from Nvidia, have challenged big cloud companies.

These companies quickly gained ground by highlighting that they do not develop their own AI that competes with their customers, which are AI companies.

Some AI-focused cloud companies are also shifting priorities. The Information recently reported that some firms, including Applied Digital and Crusoe, are leaving cloud operations to others and focusing on winning contracts to develop large-scale data centres. Lambda recently told investors that it is prioritising infrastructure buildouts for large AI companies such as OpenAI and Google.

As they grow, AI-focused cloud startups have also faced conditions that make it harder to move as fast as before. The Information reported that some have turned down deals, citing limited capacity to handle them, and that new players are emerging to fill the gap.

Newly emerging AI-focused cloud companies appear to be relatively less focused on supplying GPUs quickly than existing firms. Instead, they are focusing on developing tools and software that improve the experience of using chips.

Big cloud companies generate revenue by selling managed services or high-margin add-on features based on servers, storage and networks, and The Information reported that opportunities still remain for AI-focused cloud companies that do this well.

Against that backdrop, Accelsius, which specialises in liquid-cooling-based data centre cooling technology, has emerged as a startup drawing attention.

Accelsius recently raised $65 million in funding. The round was led by strategic investor Johnson Controls, a smart building solutions company, and was joined by electrical equipment group Legrand.

The Information reported that Accelsius uses a refrigerant that enables two-phase cooling, in which the coolant changes from liquid to gas as it absorbs heat from GPUs, unlike the current mainstream liquid-cooling method that circulates water over a cooling plate attached to servers.

The company said interest in liquid-cooling technology is growing as more data centre operators worry that a significant portion of their power budget goes into managing servers to prevent overheating and damage.

The Information also reported that Nscale is raising several billion dollars in investment to build data centres for OpenAI, while Fluidstack is increasing funding committed to supporting Anthropic infrastructure. AI startups such as Moonlite AI are also expected to raise investment within weeks, emphasising that they are better at running AI than major cloud providers.