I recently read Tyler Hamilton’s book The Secret Race – an explosive exposé of performance-enhancing drug use in professional cycling, much of it centred on seven-time Tour de France winner Lance Armstrong.
The Artificial Intelligence (“AI”) arms race is, of course, perfectly legal (unlike the cyclists’ EPO!), but it shares an important similarity with a cycling peloton: large teams working together at huge speed, pulling each other along – but only temporarily. Eventually the alliances break, the cooperation fades, and the real race begins.
Calling the AI winner today would be like predicting the yellow jersey before the first climb. Our approach is to gain diversified exposure across the AI ecosystem – the chip designers, infrastructure builders, hyperscalers, and the companies using AI to boost productivity – rather than trying to back a single champion at this early stage.
Source: Bloomberg
For now, they are cooperating out of necessity: the pace of innovation is too fast for any one of them to go solo. But they are not friends. They are frenemies – training together today, but each waiting for the moment to break away.
Nvidia: The Bike Designer Everyone Depends On
If the hyperscalers are the big teams, Nvidia designs the ultra-high-performance bikes they all want to ride. With around 90% market share in AI-specific GPUs and 70%+ gross margins, Nvidia is the critical supplier of the AI era.
It also recognises its vulnerability: roughly 60% of revenue comes from just four customers. That concentration makes it risky should a hyperscaler succeed in designing its own chips – something already underway at Alphabet (TPUs) and Amazon (Trainium/Inferentia).
To broaden its base and keep demand strong, Nvidia is spreading its bets across the ecosystem. In September this year, it signed a letter of intent with OpenAI (the research company behind ChatGPT) stating it intends to invest up to $100bn as OpenAI deploys large-scale infrastructure in the coming years. This is not a lump-sum payment; it is a ceiling on a multiyear partnership and is the largest AI infrastructure commitment ever announced. It is part of a wave of multi-hundred-billion-dollar AI infrastructure commitments across the industry – deals that collectively exceed $1 trillion in long-term planned investment.[1]
TSMC, Broadcom and the Hidden Enablers
If Nvidia is the bike designer, then TSMC is the bike manufacturer in this AI peloton.
Nvidia does not manufacture its chips – that falls to TSMC, whose two biggest customers (Nvidia and Apple) account for around 35% of revenue. TSMC is the bottleneck in cutting-edge chip production.
Meanwhile, Broadcom and others provide the high-speed networking inside data centres – the equivalent of the radio systems coordinating riders in a cycling team.
These parts of the ecosystem are less glamorous, but essential.
The Research Labs: The Legal, Performance-Enhancing Drug
If GPUs are the bikes, research labs such as OpenAI, Google DeepMind and Anthropic are the legal performance-enhancing drug powering this race.
OpenAI’s valuation has surged from $1bn in 2019 to around $500bn today, despite generating c.$13bn in revenue and with only 5% of its 800 million monthly active users paying for ChatGPT. This is astonishing scale and adoption – but clearly not sustainable without continued funding.
For now, incentives are circular:
- Hyperscalers fund research labs
- Labs create models that drive cloud demand
- Model training requires Nvidia chips
- Nvidia supports both labs and cloud providers
- Talent flows to whoever pays the most
This system works beautifully while the peloton is in sync. But it cannot last forever.
What This Means for Investors
The AI trade has surged this year, leading some to question whether it resembles the dot-com era. We see three key differences:
- Profits – today’s leaders are cash-generative and strategically essential. US technology companies now generate a return on equity more than twice the level seen in 1999.
Source: Bloomberg
- Valuations – far below the extremes of 1999. US tech is running on a forward price/earnings multiple of 27x : roughly half the level in 1999
- Supportive policy backdrop – the Federal Reserve is cutting interest rates. A marked contrast to ‘99/00, when the Fed hiked rates 6 times in 12 months!
But, we also believe in learning from history.
Today’s US equity market is extremely concentrated: the top 10 companies represent roughly 40% of the index. High concentration is one of the few clear similarities to 1999 – and the lesson from that period remains relevant.
When the dot-com bubble burst, many of the era’s “must-own” names dramatically underperformed for years, and several disappeared entirely. Microsoft, an undoubted winner of the internet boom, underperformed the US share market for nearly 19 years. For context, Microsoft is the best performing stock of those that made up the top 10 companies in March 2000. Others such as Intel have only just moved back into profit in price terms over the 25 years that have followed, whilst £1,000 invested in Cisco (the biggest stock in the US index in March 2000) would today be worth just £2,250 today: some £6,000 less than had it been invested in the broader US stock market.
The takeaway is not that today’s AI giants are destined for the same fate – they are far stronger businesses. Rather, it is that concentration magnifies both opportunity and risk.
Our View
Rather than trying to crown a single winner, we prefer to own the ecosystem – hyperscalers, chipmakers, infrastructure enablers and the businesses deploying AI productively. This captures the structural opportunity while avoiding the risks of excessive concentration.
AI is not 1999. But 1999 still offers a lesson worth remembering.
In a peloton travelling this fast, breakaways will come – and leadership will change hands. Our job is to ensure clients benefit from the whole race, not just the early front-runners.
[1] Source Financial Times
