Key points
AI infrastructure is a multi-year build-out. Real and growing revenue backs spending, with order books extending well beyond 2028.
The opportunity is wide, as networking, semiconductor equipment, and power/analogue chip companies should all be material beneficiaries alongside the obvious names.
Physical AI is the next wave. However, regulatory hurdles and trust-building mean a more gradual cycle than the current data centre sprint.
Recently, I had the opportunity to travel to Los Angeles and San Francisco to meet with the management teams of some of the world's most important technology companies. What I heard paints a picture that is at once more confident and more complex than the headlines suggest.
AI is not in a bubble
The most striking thing about spending time with these companies is how concrete and tangible the artificial intelligence (AI) investment cycle appears to have become. This is no longer a story about possibilities; it is a story about construction. Data centres are being built at a pace that was unimaginable five years ago. Power grids are being mapped and claimed. Entire regions – from the American mid-Atlantic to Malaysia to the Middle East – are being surveyed for where the next generation of computing infrastructure will sit.
At the heart of this build-out sits Nvidia, the maker of the specialised chips that power AI. Meetings with Nvidia left little doubt about the scale of demand they are navigating. Its customers – the world's largest technology companies – are placing purchase orders that extend well beyond 2028. The total planned investment in AI infrastructure across the industry has been estimated at several trillion dollars by the end of this decade, a figure that would have seemed fanciful not long ago but which now appears entirely plausible.
Figure 1: AI takes the lion's share
AI-linked companies are driving a disproportionate share of S&P 500 revenue, earnings, and capital expenditure growth.
Source: Empirical Research Partners, Q1 2026 estimates. Excludes financials and REITs; measured as aggregates.
Why are companies willing to spend so much? Because AI is no longer just a research project; rather, it is generating real revenue. The companies building and deploying AI models are now making money from every piece of analysis, every customer interaction, every automated task their systems handle. That shift from experiment to profit engine is the single most important development of the past year, and it is what gives credibility to the enormous spending plans we are seeing.
One of the most useful ways to think about this is through the lens of the cost of electricity used to power AI. When AI processes a request, it produces what engineers call "tokens", the basic units of information. Every token costs money to produce, and companies are in a fierce competition to drive that cost as low as possible while keeping quality high. The value of faster, more efficient chips is measured directly in the economics of every AI product their customers sell. This is why spending on computing infrastructure behaves less like discretionary technology spending and more like investment in essential industrial machinery.
Beyond chips: the networks that connect everything
When most people think about AI infrastructure, they think about the chips doing the processing. What is often overlooked is the networking equipment that allows those chips to work together. Imagine tens of thousands of chips in a data centre that need to share information with each other thousands of times per second – the speed and reliability of the connections between them has a direct bearing on how efficiently the whole system operates.
We met with Arista Networks, Lumentum and Astera Labs to learn about the latest development in AI networking. What struck us was the strong demand and the need for technological improvement at every part of the networking driven by AI inference and agentic AI. We see higher demand on scale-up (connecting the AI chips together within a server rack), and even the demand at scale-across (connecting datacentres together) has surprised a number of companies. Companies tell us they are seeing their customer base broadening out and higher networking requirements at each generation of AI chip.
Quiet beneficiaries
Away from the spotlight of the AI chip wars, another group of companies is benefiting enormously and with considerably less fanfare: the makers of the extraordinarily complex machines used to manufacture semiconductors.
To understand why this matters, consider that building a modern chip requires hundreds of precisely sequenced steps, each performed by specialised equipment that can cost tens of millions of dollars per machine. As the world demands more and more chips, the manufacturers of these tools – companies like Applied Materials, ASML, Lam Research, and KLA – find themselves with order books extending further out than ever before.
What I heard from these companies' management teams was a level of confidence in forward demand that I have rarely encountered. Visibility into demand stretching to 2028 and beyond, driven both by AI computing and by the resurgence of memory chip manufacturing, is giving these companies the confidence to invest in capacity, expand their margins, and plan for a sustained period of growth.
From digital to physical AI
One theme that surfaced repeatedly in my conversations was the question of what follows the current wave of AI development. The systems most people are familiar with today – the chatbots, the image generators, the coding assistants – are digital AI. They exist in software and produce digital outputs.
The next frontier, which several executives described as already arriving, is physical AI: systems that perceive and act in the real world. Autonomous vehicles are the most visible example, with self-driving taxi services now operating commercially in multiple cities. Industrial robots capable of adapting to new tasks with far less programming than before are beginning to appear in factories.
The investment implications of this shift are significant. Physical AI requires not only the training infrastructure that currently drives most spending, but also edge computing – processing power installed close to where decisions are being made, whether that is in a vehicle, a factory, or a building. This represents an additional layer of demand on top of an already stretched supply chain.
The primary constraint on how quickly physical AI can expand is not, as might be expected, the technology itself – executives were largely confident that the engineering challenges are being solved – but rather regulation and the time required to build consumer and business trust. These are slower-moving forces, which suggests that while physical AI's eventual impact may be profound, the investment cycle will be more gradual than the rapid build-out currently underway in data centres.
Investment takeaways
Spending two weeks in intensive conversation with the people that are running these businesses left me with several broad convictions.
First, the AI investment cycle has more duration than many sceptics assume. The companies spending enormous sums on computing infrastructure are doing so because the returns – measured in AI-driven revenue and cost savings – are visible and growing. Customer commitments extend years into the future, and the companies supplying the infrastructure have multi-year order books to match.
Second, the opportunity is broader than the headlines suggest. The names dominating AI coverage, notably a handful of chip companies and technology giants, represent only part of the picture. The networking companies connecting AI infrastructure, the equipment makers enabling chip manufacturing, and the analogue chip companies managing power delivery across every piece of the system are all benefiting substantially.
Finally, it is worth acknowledging what we do not know. The pace at which AI-generated revenue will grow, the ultimate size of the physical AI opportunity, and the timing of any eventual slowdown in capital spending are all genuinely uncertain. What is less uncertain is that we are in the middle of a technological transition of considerable importance – one that is creating real and durable value for the companies best positioned within it.
We believe our portfolios are well positioned to benefit from these trends, and we will continue to update you as our thinking evolves.
—
This article reflects meetings held at the Jefferies LA Technology Conference (May 2026) and the BofA Securities San Francisco TMT Conference (June 2026).
Important information
This document is intended for retail investors and/or private clients. You should not act or rely on any information contained in this document without seeking advice from a professional adviser.
This is a marketing communication. Issued by Sarasin & Partners LLP, 50 George Street, London, W1U 7DY. Registered in England and Wales, No. OC329859. Authorised and regulated by the Financial Conduct Authority (FRN: 475111). Website: www.sarasinandpartners.com. Tel: +44 (0)207038 7000. Telephone calls may be recorded or monitored in accordance with applicable laws.
This document has been produced for marketing and informational purposes only. It is not a solicitation or an offer to buy or sell any security. The information on which the material is based has been obtained in good faith, from sources that we believe to be reliable, but we have not independently verified such information and we make no representation or warranty, express or implied, as to its accuracy. All expressions of opinion are subject to change without notice. This document should not be relied on for accounting, legal or tax advice, or investment recommendations. Reliance should not be placed on the views and information in this material when taking individual investment and/or strategic decisions.
Capital at risk. The value of investments and any income derived from them can fall as well as rise and investors may not get back the amount originally invested. If investing in foreign currencies, the return in the investor’s reference currency may increase or decrease as a result of currency fluctuations. Past performance is not a reliable indicator of future results and may not be repeated. Forecasts are not a reliable indicator of future performance.
Neither Sarasin & Partners LLP nor any other member of the J. Safra Sarasin Holding Ltd group accepts any liability or responsibility whatsoever for any consequential loss of any kind arising out of the use of this document or any part of its contents. The use of this document should not be regarded as a substitute for the exercise by the recipient of their own judgement. Sarasin & Partners LLP and/or any person connected with it may act upon or make use of the material referred to herein and/or any of the information upon which it is based, prior to publication of this document.
Where the data in this document comes partially from third-party sources the accuracy, completeness or correctness of the information contained in this publication is not guaranteed, and third-party data is provided without any warranties of any kind. Sarasin & Partners LLP shall have no liability in connection with third-party data.
© 2026 Sarasin & Partners LLP. All rights reserved. This document is subject to copyright and can only be reproduced or distributed with permission from Sarasin & Partners LLP. Any unauthorised use is strictly prohibited.
