Signal over Noise with Ulrike Hoffmann-Burchardi

Hello and welcome to Signal Over Noise. I'm Ulrike Hofmann-Borchati, CIO for the Americas and Global Head of Equities for UBS Wealth Management. We downgraded U.S. tech last week from overweight to neutral.

Being neutral on tech is a bit like swimming upstream. It's a contrarian position and one not to take lightly. Over the last 40 years, tech has outperformed in roughly two out of three years.

So what drives our conviction that the risk reward in tech is getting less favorable? We believe that we are entering a new chapter in AI. One that I would call the face of young adulthood where economic realities will start to sink in.

We're now getting to a level of AI CapEx where the math is getting more challenging. Not only because we have reached a level of annual CapEx that is equivalent to 15 years of CapEx for cloud computing or that it now reached 2% of U.S. GDP, but because the key question is getting more difficult to answer.

How is AI spent going to turn into profits? Revenues are simple to reason through. AI addresses 30 trillion in global knowledge worker wages.

But profits? That's not so simple. The most important question is, how many winners will there be?

And with winners, I mean those at the frontier of artificial intelligence. Those companies that are driving this new wave of generative AI and are competing for the top spots on the AI league tables. The answer to this question determines the market structure and profit pools.

Right now we have about a dozen frontier model apps such as OpenAI, Anthropic, Gemini, XAI, Meta, Mistral, Alibaba, DeepSeek, Moonshot, Cohere, Baidu, and Tencent. And new ones that have been formed more recently. Safe Superintelligence, Thinking Machines, AMI Labs, Sakana AI, just to name a few.

And there will be more. Microsoft's head of AI, Mustafa Suleyman, said last week that Microsoft will be coming out with its own frontier model sometime this year. So at this stage, the frontier model market looks like the textbook definition of perfect competition.

A dozen competing labs, new entrants on the horizon, low switching costs, and performance transparency. In perfect competition, price equals marginal costs and companies earn zero economic profits. This describes what we're seeing right now.

The largest private frontier model apps do not seem to be profitable based on reports from recent funding rounds. There are two reasons frontier models need to spend on compute. For training, that is spend that is required to be at the forefront of intelligence.

And for inference, that is serving customers with AI queries. Frontier model companies are profitable when it comes to serving customers with inference. But once R&D costs for training are taken into account, companies no longer break even.

So they rely on more funding to keep going. Dario Amodai, the CEO of Anthropic, confirmed this in a podcast on Friday. He asks, why doesn't everyone spend 100% of their compute on training and not serve any customers?

Because if they didn't get any revenue, they couldn't raise money, they couldn't do compute deals, and they couldn't buy more compute next year. This means that access to funding will likely become an increasing bottleneck. So far, the frontier model companies have not yet distanced themselves from one another for an extended period of time.

Also, Chinese open-source models are close on the heels of closed-source US frontier labs, as recent releases have shown. While not quite as performant, they deliver intelligence at substantially lower cost, a challenge to pricing power for closed-source frontier models. Those closed-source frontier models that have access to cash flows from other businesses have a strategic advantage.

They don't have to balance the trade-off between training and inference, especially if they can use AI for their own applications. The alternative outcome is that over time, the market will morph into an oligopoly with only three to four players. This is possible if frontier model companies get to differentiate themselves along the AI value chain via applications or compute.

This strategic direction is already evident. OpenAI has launched OpenAI Health. Volanthropic has introduced financial and legal plugins for enterprise users.

And OpenAI's partnership with chip companies speaks to an attempt to differentiate with compute. If this differentiation is successful, the laggards will cease to be funded. If not, funding will likely dry up also because sustainable profit margins seem out of reach.

So to summarize, the 600 billion or more of CapEx this year raises the stakes in the intense competition to achieve durable profit margins. It will likely either lead to fewer private frontier model companies being funded, private companies encroaching on public company revenues, or both. This makes the risk-reward for the tech sector less attractive in our view.

With this, stay well, stay diversified, and stay ahead. UBS Chief Investment Office's investment views are prepared and published by the Global Wealth Management Business of UBS AG or its affiliate, UBS. This material has no regard to the specific investment objectives, financial situation, or particular needs of any specific recipient and is published for informational purposes only.

As a firm providing wealth management services to clients globally, UBS AG and its subsidiaries offer both investment advisory services and brokerage services. Investment advisory services and brokerage services are separate and distinct, differ in material ways, and are governed by different laws and separate arrangements. In the USA, UBS Financial Services, Inc. is a subsidiary of UBS AG and a member of FINRA SIPC.

For information, please visit our website at ubs.com forward slash working with us. For a full legal disclaimer applicable to the independent investment views produced by UBS, please visit our website at ubs.com forward slash CIO dash disclaimer.