The best way to play scarcity

So when it emerges that Altman is courting Middle Eastern investors and semiconductor fabricators to develop an AI chip venture, people pay attention. Even then, ambition has boundaries. If reports are accurate, the project is looking to raise as much as $7tn (£5.5tn). Not a typo.

For context – if context can be applied to this sort of scale – $7tn is the combined annual gross domestic product of Japan and the UK. It’s $3tn more than the ‘dry powder’ on the balance sheets of the entire private capital industry, per BlackRock estimates. Nothing in the history of project finance comes remotely close to this sort of fundraising.

According to some commentators, however, there is some logic to the ballpark figure. Currently, ChatGPT is in its fourth generation. Today, thanks to its $10bn investment from Microsoft (US:MSFT), OpenAI is developing its fifth full version of the large language model. If each new model in the series requires 30 times the level of computer power, electricity and training data as the previous one, as it has done thus far, then the costs could well be in the trillions in just two iterations time.

The energy portion of this will be challenging enough. But there are plenty outside the tech industry with the know-how, capital and vision to come up with big answers. The greater issue may be developing the hardware necessary to not only push out the frontier of AI, but to leapfrog the incumbent player. And in case you’ve been living under a rock, you’ll probably know that this hardware is currently monopolised by one company, Nvidia (US:NVDA).

Indeed, a by no means outlandish reading of Nvidia’s feverishly-anticipated (and received) fourth-quarter earnings results is that, for now, the Santa Clara company is the insurmountable top-dog in a white-hot industry. Not only is demand for its products growing at a colossal clip, but by reinvesting billions of dollars of its cash into those products, it is further cementing its industry dominance. While estimates vary, it is likely to take years for competition to draw level.

While Nvidia is currently the greatest beneficiary of the AI sector’s mammoth capital spending, the scale of the sector’s capital requirements is hard to measure. Trillions? Eventually, maybe – although over what time scale clearly matters. However, there is even less certainty around the demand for whatever it is a ChatGPT-7 might provide. As such, prospective investors in Altman’s project can only hope to sketch the vaguest of back-of-fag-packet net present value estimates.

Of course, Altman is unlikely to raise $7tn for his venture. But this doesn’t necessarily mean AI’s rollout is capital-constrained. Moreover, judging by both Nvidia’s top line and the levels of AI-related capital and R&D spending in the tech sector, this isn’t a field that lacks fresh cash. After all, if Nvidia wanted to drop $20bn on a product upgrade, at its current rate of earnings it would only have to wait about four months to fund it entirely from free cash flow.

If anything, the unquantifiable scale of the AI prize is likely to mean a fair slug of capital spending is already inefficiently allocated. Whenever too much capital chases too little profit, returns are likely to suffer. What’s more, if the pace of AI development is as fast as some claim, then plenty of capital risks funding products that prove undifferentiated or even commodified. The tech sector may be entering its true golden age, or a turbocharged phase of creative destruction.

Looking for the capital-starved

If today’s bleeding edge of technology feels closer to science fiction than the parochial concerns of ordinary investors, it might be useful to ground it all in economics. Fortunately, this also provides some alternative ideas to the tech-tech-tech narrative.

We don’t know how much OpenAI spent developing ChatGPT, although losses reportedly exceeded $500mn in the year before its platform exploded into public consciousness. And although we can only guess at margins and profits, commercialisation appears to have been a success. Annualised sales reportedly climbed from $1.3bn to $2bn in the last quarter of 2023. Although it is hard to disentangle growth from the financial and logistical support of Microsoft, OpenAI’s return on capital may well be positive. Obviously, we can say the same of Nvidia.

While it’s possible that in 2024, every penny spent on AI is capital well-allocated, the normal distribution of business failure rates suggests this is unlikely. At the same time, AI investing is based on the idea that the technology represents the highest marginal use to the consumer. Like all use cases, this involves estimates, and is only true until it isn’t. In many supposed cases, we’re not yet sure the highest marginal use test applies. This means that at some point over the current AI capital cycle, the marginal utility to consumers may well diminish, along with returns on capital. Then again, it might not. As noted, lots of very clever people are involved.

Another effect of the magnetism of new technology is that it gets very crowded, which ends up raising expectations and bidding up the present value of future expected profits. It also serves to distract investors from the sectors in which capital scarcity is real.

One investor who thinks about this dynamic a lot is Tian Yang, chief executive at the research group Variant Perception (VP). Among many other lead indicators, Yang and his team are advocates for the idea that capital cycles serve as powerful predictors of long-term equity returns. Capital scarcity is a key component of this theory, which VP applies on a sectoral (rather than individual stock) basis, and calculates as a composite of the three following ratios:

After screening out more crowded sectors – which VP defines as those subject to more speculative investor flows, and therefore prone to higher volatility – this capital scarcity indicator currently favours gold mining, Latin American (specifically Mexico and Brazil), and oil and gas equities.

If each of these sectors feels a world away from the glamour and drama of the AI story, that’s sort of the point. So have investors failed to identify their risk-reward profile?

Let’s start with gold mining. Despite a poor recent record of returns (from gold equities, if not the yellow metal itself) current projections are for global output to climb over the medium term. Despite falling grades and the growing challenge of replacing reserves, a 2023 report by Fitch Solutions forecasts that “high prices by historical standards” will continue to incentivise capital spending and production in the medium term. The report reckons output will climb by an average of 1.5 per cent a year until 2023, almost double the rate between 2016 and 2020.

Because VP excludes valuations from its composite score, it’s tricky to separate the physical market outlook from equity market assumptions. That’s important because when gold miners aren’t making a hash of their operations, they often face elevated geopolitical risk. Both issues explain high capital costs. Nor does a capital scarcity framework tackle the thorny issue of the gold price, which determines earnings, but is impacted by all manner of non-industrial and market factors.

As with gold miners, investor returns from oil and gas stocks have a lot to do with hard-to-predict commodity prices. Unlike gold, however, energy’s essential function within the global economy arguably makes it a more investable proposition – even if fossil fuel demand is nearing a peak. But does the listed oil and gas sector tell the full picture of the industry’s access to capital? Given the weight of state actors in the energy market, it’s hard to say. What appears to be a lack of funding could simply represent the appetites of a certain kind of public investor.    

Thematically, Latin America feels like a safer story. If there was one takeaway from full-year results for Bearbull Income Portfolio holding Anglo American (AAL), in which the miner booked large impairments to its diamond and platinum group metal assets, it is that its future lies in South America, rather than South Africa.

Beyond resources, there’s the US near-shoring angle, and the multiplier effect this might bring economies south of the Rio Grande. As to why capital is scarce, familiar ‘emerging markets’ reservations around geopolitical and foreign exchange risks are likely to apply. Set against the tech-powered lights further north, the oversight is understandable.

Of course, private investors’ ability to literally plug the capital gap is limited. Unless you participate in a gold miner’s equity fundraising, buying shares in the open market isn’t capital formation. That means speculative investments in under-supplied commodities are sometimes the best way to ‘play’ scarcity. For investments predicated on capital scarcity to work, we eventually need the big money to flow in, too. Maybe that’s not so different to Altman, after all.