The concerted sell-off in companies related to Artificial Intelligence (AI) over the last few weeks comes as investors question the capital spend of players such as Open AI, which as at this moment will have to raise additional capital if it is to meet its aggressive US$1+ trillion datacentre build targets.
The fall in values was sharper than the run-up in valuations for the year to date as at 31 October, which brought to double digits the increase in share market indices over the period.
Part of the problem was the US Government shutdown and a US Federal Reserve trying to make decisions without clear data. It’s understandable that some investors are worried about the macro and are taking profits.
However, concern about an AI bubble which is set to pop imminently seems premature.
One obvious data point: the continuing strength in AI was demonstrated by Nvidia’s recent earnings report. The company surpassed analyst expectations for both earnings and revenue, with sales growth re-accelerating compared to the previous quarter. Nvidia is now guiding for fourth quarter revenue that will exceed its entire annual revenue from just two years ago, and by the end of 2027, the company forecasts its annual revenue could hit USD350 billion or more.
Nvidia’s outlook is supported by cloud CapEx, projected to grow at +34% in 2026
Source: Company Data, Evercore ISI Research, FactSet, Exclude OpenAI
It’s true that Nvidia-funded neo-cloud companies like CoreWeave use that investment to make purchases with Nvidia, but this is a small fraction of Nvidia’s current and projected revenue. The vast majority of this investment is coming from financially robust giants like Amazon, Microsoft, Meta and Alphabet. What’s more, relative to 12 months ago it is evident just how much visibility has improved. Twelve months ago, Evercore forecast 2025 cloud CapEx to be +24% and 2026 to be +6%. Compare that with the +74% and +34% seen above.
Nvidia dominates the AI hardware market
Yet market moves suggest that AI spend is too high, with an unfunded component because the revenue model is not yet entirely clear. By way of example, some pundits believe there is no “killer” app, which in the case of the internet was email. But this misses the point – AI is millions of different things to different users, with the large language model (which is a probabilistic forecasting tool) the source of all these applications. To a teacher, it is lesson plans generated from an image of a page in book, while for doctors it is a means to update medical records instantaneously rather that relying on memory as that last job of the day. Businesses are now deploying AI agents to automate workflows, whether it’s through customer service, coding or reporting, effectively augmenting their workforces. AI agents become digital co-workers that can work 24/7. Then comes physical AI, robotics, extending productivity into factories, logistics and healthcare. In aging societies with shrinking workforces, that’s a powerful new growth driver.
To coders it is something different again… the list stretches as far as the eye can see. AI is being woven into how we work and how we live. It may feel like a solution in search of a problem to some, but that is just “the shock of the new” to borrow a phrase from Robert Hughes, one of the world’s pre-eminent art historians. Within a year or two, AI will be viewed as a necessity.
The Scale of the AI Economy
Let’s frame the scale.
- Around 1 billion fixed broadband households already pay close to US$1 trillion a year for connectivity.
- Nearly 6 billion mobile subscribers pay another US$1.5 trillion for mobile and data. This is the foundation upon which the AI economy will build.
- There are around 1.4 billion knowledge workers globally. If each spends $20 a month on AI tools like ChatGPT Pro, that’s $350 billion per year in annual revenue.
A slice of any of these would fund the chip depreciation of US$200b on a US$1 trillion data centre capital build, assuming a five year depreciation schedule.
AI is already in the workforce. Businesses are now deploying AI agents to automate workflows, whether it’s through customer service, coding or reporting, effectively augmenting their workforces. AI agents become digital co-workers that can work 24/7. Then comes physical AI, robotics, extending productivity into factories, logistics and healthcare. In aging societies with shrinking workforces, that’s a powerful new growth driver.
Investment and Utilisation
AI is likely to reach utililty scale in the coming years, requiring providers to quickly adopt new levels of sophistication to stay competitive. That’s why AI uptake is happening faster and more broadly than any previous technology cycle. Microsoft, Amazon, Google, Nvidia, Broadcom, and model builders like OpenAI and Anthropic, are providing and building the compute and AI infrastructure everyone is relying on.
In our view, this cycle is very different from the 2000 tech bubble. Back then: huge amounts of capital went into “dark fibre” that sat unused for years. Investment ran far ahead of demand. Today: Nvidia GPUs that are deployed are getting fully utilised straight away. There’s virtually no idle capacity. Usage and investment are moving together. AI demand is real, immediate and monetised.
While overbuild is a credible risk, given the financial heft of the large companies driving the AI spend, the downside risk is limited (at least until these customers start financing their AI CapEx with heavy debt). As Mark Zuckerberg said on Meta’s Q3 earnings release, “In the worst-case scenario, we would have just pre-built for a couple of years, in which case there are multiple ways for us to digest that.”
Investment Strategy
At Loftus Peak, with our 3 to 5-year investment horizon, we have allocated more capital to the essential tools and enablers of AI where most of the value has accrued so far. We have also invested into the application layer, but it is too early to conclude the best of these opportunities.
At this stage, our main goal has been to participate in the growth in volume across the entire AI ecosystem, no matter which application or software layer ultimately dominates. Simply put, it’s a safer way to achieve solid returns.
Share this Post