As tech and AI valuations hit record highs in the United States, a growing share of asset managers is betting on infrastructure rather than end-user applications.
"During the gold rush, it wasn't the miners who got richest, but rather the people selling shovels and picks." Nearly two centuries later, this maxim still resonates with investors.
Anjali Bastianpillai, portfolio manager at Pictet Asset Management, agrees: "Providing the shovels and picks, i.e. the infrastructure, has been far more profitable than trying to find the most lucrative AI applications."
The infrastructure needed to roll out AI is built on a well-identified chain of players in semiconductors (NVIDIA, AMD, Broadcom), foundries (TSMC), memory (SK Hynix) and equipment makers (ASML, KLA, Lam Research)? The rollout of large-scale AI data centers is in full swing: the hyperscalers (Alphabet, Meta, Amazon, Microsoft) are investing heavily, with combined capex nearing $500bn in 2025, a rise of around 68% in twelve months.
Should we fear a bubble? Anjali Bastianpillai is reassuring: "What matters a great deal to us is that this capex is mainly financed by free cash flow. That's what makes the difference compared with the 2000 tech bubble." In other words, spending is massive, but it rests on solid balance sheets and already highly profitable businesses.
For Pictet AM, the valuation premium is therefore gradually shifting towards software and "physical" AI. "We're starting to see ROI, very compelling models," she observes.
Software and agentic AI enter monetization phase
Productivity gains are particularly visible with so-called "agentic" AI - systems capable not only of generating text or images, but above all of executing business tasks.
Five9 is pushing heavy use of its AI to automatically route calls to the right person in contact centers; Snowflake says that 80 % of customer service requests are now handled by AI (the equivalent of "400,000 working hours" automated), while CrowdStrike, with its Charlotte AI module, cuts data analysis time for cybersecurity teams "from several hours to a few seconds".
According to Pictet AM, the market is still largely underestimating this monetization, with "software valuations that remain at levels that are not very high compared with the Covid period". Anjali Bastianpillai sees "an inflection point". In her view, companies are beginning to capture the productivity gains they have been promising for the past two years.
The major platforms have already blazed a trail: with more than 3.5 billion users across its services, Meta has integrated generative AI into advertising and video content, contributing to growth of more than 30 % in Reels formats and to a run-rate of around $50bn for this vertical alone.
Alphabet, for its part, is posting more than 200% annual growth in revenue tied to its AI products, while Google Search remains extremely solid in terms of usage, traffic and revenue.
Physical AI, between robotaxis and humanoids
So-called "physical" AI (notably autonomous taxis and humanoids) is still in development, but it is already under close scrutiny from fund managers. Waymo and Baidu robotaxis are already completing more than 500,000 trips a week, Pictet AM notes. The main advantage of this technology would be safety, with "far fewer accidents". Between 2025 and 2034, this market is expected to grow at an annual rate of 53%, rising from $4bn in 2025 to $189bn in 2034.
On humanoids, the orders of magnitude are beginning to emerge - Goldman Sachs is pointing to a $38bn market by 2035 - but Pictet AM is favouring an indirect approach:
"We are not yet investing directly in humanoids, but rather in their value chain." In practice: NVIDIA for AI chips, NXP and Infineon for processing power, vision and perception players such as Cognex, and semiconductor equipment makers that will enable automation of the new factories.
In the face of recurring questions around AI - circularity of capex, longer depreciation periods, rising receivables at some suppliers - the portfolio manager remains steadfast and sticks to her compass: "fundamentals".
For Pictet AM, as long as demand holds up, capex is largely self-financed and business models deliver solid profitability, comparisons with 2000 are largely misplaced.
It is a discipline that appears to be paying off: after a change of manager in early 2023, Pictet's Digital fund has delivered roughly 48% in 2023, 34% in 2024 and more than 10% so far in 2025, with an annualized return close to 15% over ten years, around 4.5% above the MSCI ACWI. Robotics is doing even better, with nearly 18% annualized over ten years.
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.