The JSE has delivered one of the best dollar returns in global equity markets this year, yet headlines stay focused on US mega-cap AI stocks. Do South African investors stick with a local market, finally rewarding patience, or pay up for the global AI leaders and risk buying into a bubble? That’s the wrong question.
AI is a technological revolution, but for investors it remains a classic capital cycle: the real question is which parts of the AI value chain you back, and at what price. The capital cycle is straightforward. Strong returns attract capital.
Capital funds new capacity. Excess capacity crushes returns. Capacity is written off or consolidated.
Read Full Article on Business Day
[paywall]
Survivors eventually earn decent returns again. Bubbles are that cycle on fast-forward. Narratives outrun information.
Capital chases stories faster than data can catch up. Three questions position AI in this cycle. Some cycles are driven by real technological change: steam, electricity, rail, semiconductors, the internet — and, plausibly, AI.
Others are driven mainly by re-rating existing assets: think Japanese land in the 1980s or US housing in the 2000s. Plenty of leverage; not much new productivity. AI sits closer to genuine technological change than to a credit boom — but we still have to ask how much of today’s price reflects future cash flows, and how much reflects a story about those cash flows.
Funding structure matters. When banks and households lever up to chase a boom, the hangover is long and ugly. Equity-funded bubbles are brutal for shareholders but usually leave the financial system intact.
Today’s AI build-out is still driven mainly by large, profitable tech firms deploying their own balance sheets and by equity investors. On the surface that makes this cycle less systemically dangerous than 2008. But the picture is evolving.
Big tech has shifted from sitting on oversized cash piles to issuing significant debt to fund capex. The hyperscalers are now spending between 21% and 35% of revenue on capital expenditure — higher than utilities and exceeding what AT&T spent at the height of the telecom bubble. More concerning, late-cycle warning signs are appearing.
Meta secured $27bn in financing with Blue Owl Capital for its Hyperion AI data centre — the largest private-credit deal ever. And circular deals are emerging, where infrastructure providers invest in customers who then spend heavily on their chips and cloud. These are clear late-cycle signs.
The final question: if this all goes wrong, what remains? The telecom capex boom of the late 1990s left fibre networks that outlived the companies that built them and underpinned future innovative business models. The pattern is clear: infrastructure-heavy bubbles funded by debt may leave reusable assets.
Narrative-driven equity bubbles often leave only painful lessons. AI today looks more like dot.com equity than telecom debt: equity-funded, story-driven, with an asset base that is overwhelmingly intangible and hard to measure. Data centres age quickly; chips will be obsolete within a few years.
What will endure is harder to value: organisational learning, regulatory frameworks, a workforce comfortable using AI tools, and new business models born of experimentation. Participants cannot know in real time whether they are funding the future or funding a mirage.
[/paywall]
