By Jules Rimmer

Capex commitments are creating a bottleneck in funding for AI developers

Right now UBS Wealth Management identifies about a dozen frontier model developers with a growing cohort of Chinese competitors hot on their heels.

Downgrading U.S. technology stocks is not something an asset allocator does lightly – the sector has outperformed two years in every three for the last four decades – but UBS Wealth Management has taken that bold step.

Ulrike Hoffmann-Burchardi, global head of equities, said the firm has lowered its recommendation from overweight to neutral and tells investors that “the risk-reward profile is less favorable as economic realities sink in.”

In a podcast broadcast Tuesday, Hoffmann-Burchardi outlined her conviction that answering the question how AI capital expenditure will be turned into profits is becoming more difficult. AI capital expenditure has now hit 2% of GDP, she points out, and although this investment addresses a potential market of $30 trillion in global knowledge industries’ worker wages, the profits to be derived from that are not straightforward to calculate.

Hyperscaler capex spending for the “big five” is now widely forecast to exceed $600 bn in 2026, a 36% increase over 2025. Roughly 75%, or $450 bn, of that spend is directly tied to AI infrastructure (i.e., servers, GPUs, datacenters, equipment), rather than traditional cloud.

There is fierce competition among those companies operating at the frontier of artificial intelligence and Hoffmann-Burchardi is not sure who the winners will be, nor how many. She identifies about a dozen companies competing in the AI league tables, including OpenAI, Anthropic, Gemini (GOOG), xAI and Mistral, with lots of new entrants into the field from China.

This week alone, five Chinese AI companies – Zhipu (HK:2513), ByteDance, Alibaba (BABA), Moonshot and DeepSeek – are releasing or announcing new model launches. In a posting on X, former Fidelity fund manager George Noble observed their cost advantage, estimated by a RAND report, is that they charge one-sixth that of U.S. competitors.

Moreover, Chinese AI developers can power their AI datacenters with the cheap renewable energy in which they have invested so heavily in recent years, Noble adds.

To this burgeoning list of players, Hoffmann-Burchardi also adds Microsoft (MSFT). Last week, its head of artificial intelligence, Mustafa Suleyman, told investors its frontier model would be launched later this year. For UBS then, this landscape looks like “the textbook definition of perfect competition.”

At this stage, companies are yearning zero economic profits. When costs for training the models are factored in the companies no longer break even and rely on constant rounds of funding. Anthropic’s chief executive Dario Amodei admitted last week that revenues merely allow AI companies to raise funding to buy more compute (resources and capacity).

Access to funding then is likely to become a bottleneck, she says. And since the Chinese developers have a cost advantage, their competitive pressure, even with models that are less performant, is a big challenge to the dominant U.S. players.

Hoffmann-Burchardi points out that companies with cash flows from other businesses like Alibaba (BABA), Baidu (BIDU) and Tencent (HK:700) have an edge because they’re not obliged to choose between training and inference. For those laggard developers who struggle to differentiate themselves on the value chain, via applications or compute, funding will dry up.

The $600 billion of AI capex planned for 2026, Hoffmann-Burchardi concludes, raises the stakes in the battle for durable profit margins. The risk-reward profile is, therefore, less attractive.

-Jules Rimmer

This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.

(END) Dow Jones Newswires

02-17-26 0512ET

Copyright (c) 2026 Dow Jones & Company, Inc.