Let’s start with the easy one. I’m A Celeb 2025 is almost finished. The more tricky version of this headline question might relate to the Epstein files or even the filing of war crimes charges in The Hague against US Secretary of War, Pete Hegseth. Not any time soon me thinks. We could ask Sleepy Don but he might become angry – about the sleepy bit, not the war crimes or paedophilia. Actually, the question most asked in recent weeks is about the end of the AI boom. I asked my own excellent AI ‘friend’ Claude (courtesy of Anthropic) about ‘bubble’ mentions in the media and even he agreed in his remarkably comprehensive market summaries of public and private markets that the AI bubble question is occupying investors’ minds. Mine, not so much. More on Claude later, but first a historical perspective. The last technology boom in 2000 did indeed end in a bust but generalisations on technology can be misleading.
Back in 2000 we should remind ourselves of the telecoms companies racking up massive debt obligations to acquire mobile spectrum licences and build out fibre/internet networks. Then there were the infrastructure suppliers like Ericsson, Nortel and Cisco dependent on those telecoms, internet and wireless expansion projects. Then the projects stopped. A possible over-simplification by this writer, but a combination of over-build and debt pressures slowed activity and cratered the valuations (growth expectations) of the leading infrastructure players. For illustration, Cisco was trading on a price/earnings multiple of 200x in late 1999. Twenty five years later the Cisco share price has finally recovered to within touching distance of its $80 high in 2000. However, one must make a distinction between the infrastructure plays and the tools/applications which were built on those over-priced networks….
The Nokia phone in my year 2000 pocket didn’t end up ruling the world but Apple and the mobile internet did. Similarly, Google was just 2 years old at the time and wouldn’t IPO until 4 years later, the same year as TheFacebook Inc was born. Mobile networks enabled commerce (Amazon) and communities (social media platforms) to flourish and generate enormous wealth. Readers might be now detecting a similar pattern with AI. The race for computing power (in 2000 it was networks) is an infrastructure story but investors must not lose sight of the applications of AI and the business models possible (Amazon was an online book store once). The tools like Claude, ChatGPT and Sora are really only in their infancy. The infrastructure story is driven by GPU/TPU chips (Nvidia), cloud computing, hyper-scale data centres and energy. And it’s possibly infrastructure again where risks are building. The CEO of IBM, Arvind Krishna, in recent days put some numbers on those risks.
Krishna cited a data centre power consumption estimate of 100 GW which at current costs would mean an $8 trillion capital expenditure in the next few years. Now, for the wet blanket of capital reality. That ginormous $8 trillion spend would need to earn profits of $800 billion just to pay the interest/cost of that capital. Yep, that’s stretchy but get ready for the other reality. This infrastructure isn’t piping, fibre, railways or copper which lasts for decades and is depreciated gently over time. The chips which currently power Nvidia’s $5 trillion valuation and sit inside all these data centres could become technologically obsolete within 5 years. Arguably, at current innovation/evolution rates that timeline is too optimistic. Imagine having to replace all your chips every 3 years… ? That should make creditors to these huge data centre projects a little queasy. The International Financing Review summarized the massive acceleration in borrowing as follows:
“An unprecedented splurge from companies at the forefront of the AI boom that has left banks and investors potentially on the hook for billions. Alphabet, Amazon, Blue Owl Capital, Broadcom, Oracle and Meta have between them issued US$120bn of corporate bonds since September – and are raising another US$38bn in the loan market. The debt binge shows no sign of abating, with JP Morgan predicting US$300bn of bond issuance next year – and US$1.5trn by the end of 2030. Another US$2.3trn could be raised in equity, structured finance and private capital markets over the next five years, as hyper-scalers tap every available pocket of capital to finance the US$5.3trn of investments into AI they are expected to make”
Before everyone runs for the hills, we need to be mindful of some very positive starting points. These technology giants tapping the debt markets in most cases are swimming in cash, have dominant market positions and are generating prodigious annual cash flows of almost $700 billion. These are not the fragile telecom balance sheets of the TMT bust in 2000. Of course, OpenAI, sits in the middle of that famous Financial Times graphic showing $1.2 trillion of data centre projects. In my personal view, OpenAI is the weakest link but that could take years to play out. The harsh truth for all investors is that we don’t really know who will win the foundational large-language-model (LLM) race. Google’s Gemini 3.0 seems to be winning this month and did anyone notice Google share price is up 67% year-to-date? Yep, and my Claude’s parent company, Anthropic, is looking to IPO at a $350 billion valuation. These are very early days. Just ask Nvidia. Actually, don’t. They are saying nice things about almost everyone because all are prospective customers. But….. as always watch what a company does, not what it says.
Nvidia made a $2 billion investment in chip designer, Synopsys, this week. This is just the latest move by Nvidia in what can only be described as a deal spree. In 2025 alone the company has backed 77 equity investments in start-ups, as well as making 5 outright acquisitions (Source: CB Insights). Let’s just say it looks like Nvidia is hedging its AI ‘winner’ bets. Indeed, the ‘AI infrastructure’ bubble fears run the risk of missing the true lessons of the TMT bubble bust of 2000. ChatGPT might be today’s Nokia but the monthly user statistics tell another ‘mobile’ story. ChatGPT is used by 800 million people each month. Gemini is fast catching up with 650 million devotees and Microsoft’s Co-pilot has 200 million monthly users. The market, business or individual, is already converted. That’s the true investor opportunity.
Meanwhile, there’s a bigger story brewing at the other side of the world. Arguably, we really do need to see that story end very soon. More next week on why troubles in Japan’s bond market REALLY scare me……
