Listen here.Veteran investor James Gubb joins BizNews Edge to cut through the AI hype, weighing today's data centre spending against history's biggest capital booms and explaining why he's backing energy and nuclear stocks over AI names themselves. Also on today's show: Sasol's operational turnaround continues even as its share price cools after a strong run, Nedbank locks in its East Africa expansion via Kenya's NCBA Group, and Kumba Iron Ore's ugly trading update lays bare the toll of a strong rand and Transnet's rail failures. Plus, why Thursday's Reserve Bank rate decision just got more hawkish..Transcript.Host open — Alec HoggGood day and welcome to BizNews Edge. It's Tuesday, the 21st of July 2026. I'm Alec Hogg, good to have you with me.Later in the show, James Gubb takes a scalpel to the AI boom and finds some uncomfortable truths underneath all the hype.But first, Pretoria: Thursday is decision day for the South African Reserve Bank's Monetary Policy Committee, the body that sets interest rates for South Africa. The renewed war between the US and Iran, which keeps escalating, has kiboshed some of the more optimistic thinking from Reserve Bank Governor Lesetja Kganyago. He'd been hoping a decline in the oil price might allow the Reserve Bank to hold off on another hike. That hope has now faded, and the odds favour another interest rate increase on Thursday, a quarter point, 25 basis points, taking the repo rate to 7.25%.Markets round-upLet's get you up to speed on what's been moving markets this morning, starting with a name that's had a stellar few weeks, and one that irritates me because we sold it out of the portfolio a little while ago.Sasol. Sasol's latest business update, covering the year to end June, was released on SENS this morning, and it tells a genuinely encouraging operational story. Forget the oil price, this is about operations, which is exactly where CEO Simon Baloyi said management would focus, and they're making excellent progress. Simon, incidentally, is one of a star-studded cast heading to the Drakensberg for BNC9, the ninth BizNews Conference, running 11–14 August. There are still tickets left, so come along and join us.The share price didn't quite reward the update the way the numbers deserved, but that's largely because Sasol has already been on a tear. Secunda has delivered its best production year in five, and the international chemicals business, the one that went through such a rocky patch at the Lake Charles plant in Louisiana, has settled down and actually beat its own guidance. All in all, Sasol looks to have found its operating rhythm again after fighting for its life not long ago.The market's reaction was more about timing than substance. Sasol shares are up 16% in the past month, so this morning's update was largely already priced in. We saw an early pop, then profit-takers pulled the stock back a few percentage points by midday. Don't mistake this breather for a change in direction. The real test comes on 1 September, when Sasol releases its actual results for the year to end June, and the big story there will be how far it has progressed on deleveraging, paying down the debt that had investors so concerned.Nedbank. Nedbank gave formal confirmation this morning of a deal we've been tracking since May: its move into East Africa via a controlling stake in Kenya's NCBA Group. NCBA shareholders have given it the thumbs-up, with close to 80% accepting Nedbank's offer on SENS today. That puts Nedbank's stake comfortably at around two-thirds of the enlarged bank. It'll be settled mostly in cash, which limits dilution, expanding Nedbank's equity by under 9% for a business that should contribute meaningfully more than that to future profits.The market shrugged, and in this case that's a compliment: it was priced in months ago, and Nedbank shares barely moved while the rest of the banking sector had a soft day. What stands out here is the discipline. Nedbank has bought its way into a faster-growing part of the African continent without stretching its own balance sheet. NCBA could become a real jewel in the crown, the way African operations have performed for Standard Bank and, more recently, Absa. It's the kind of expansion I'd like to see more South African corporates, beyond the banking sector, pursuing.Kumba Iron Ore. Kumba's trading statement this morning was ugly: earnings down by around a third, headline earnings per share down closer to 40%. A genuinely painful set of numbers, and the market sold the stock down hard through the morning. The causes are worth understanding rather than just wringing our hands. The rand's 11% strengthening over the half year has hurt any exporter selling dollar-priced commodities, and Transnet's rail network, which has let the country down badly in recent years, forced another shutdown in May, hitting both production and sales. You can't sell what you can't get to market.There is a silver lining: Kumba should benefit from the changes underway at Transnet, including the partial privatisation of some rail lines, which ought to be a boon for the company going forward. This is a decent business held back by infrastructure that can't keep pace and a currency nobody at Kumba can control. Kumba's financial results land on 28 July, and I suspect they'll disappoint, particularly on the dividend, which had been a juicy one.Anchor story: James Gubb on the AI boomOnto our anchor story. There's a mania playing out in markets around artificial intelligence, and we've found someone who can give us real insight into it. James Gubb has spent a career separating genuine value from a good story, and right now AI is a very good story. The question is whether there's genuine value underneath it.James is a chartered financial analyst and a former top-rated South African money manager, now retired to Plettenberg Bay, where he keeps his mind active on exactly this kind of analysis. He's drawn heavily on the book Devil Take the Hindmost [transcribed as "Engines That Move Markets" — please verify exact title], which examines centuries of speculative booms, from railways and telegraphs to the dot-com era, and applies those lessons to AI. James has compiled a full report, published on BizNews.com.In our conversation we covered Oracle's credit downgrade (which landed yesterday), why today's data centres are, relatively speaking, bigger than the pyramids, and who the likely winners and losers of the AI boom will be, including where James is putting his own money. Here are the highlights; the full interview is available in Director's Cut on BizNews.com and on YouTube.On scale: AI capex vs the pyramidsAlec Hogg: It really is a reminder, for those of us who've been around a while, of the dot-com boom and bust, and then the Global Financial Crisis in 2008. You draw a comparison between those two. Do you see today's AI boom as more of a Global Financial Crisis-style event, or a straightforward crash?James Gubb: We're just trying to look at what the capex spend has been over history. I was actually trying to work out how big this spend is relative to the pyramids, because I genuinely think it goes back that far in commercial history. I'd assumed the dot-com boom was the biggest until now. Have you seen the size of some of these data centres being built? They're utterly extraordinary, and it's not just one, there are plans for a hundred. There's a lot of red tape in the way, but it's pyramid level.Alec Hogg: Explain the scale for those who haven't seen these data centres.James Gubb: I can't give an exact figure, go and Google it, but the base of one is roughly the size of the Great Pyramid of Giza.Alec Hogg: So a genuinely huge undertaking. And you've got Elon Musk on the other side saying he'll build them in space because it's more efficient. When you compare it to the pyramids, at that stage in human development they were off the charts, bigger than anything else built before. Is that the parallel?James Gubb: Yes. There's debate about exactly when the pyramids were built, but I'm assuming around 6,000 to 7,000 years ago, and I'm using that just to give scale and context to what's happening now. It's massive, it dwarfs the dot-com capex spend on fibre, which, GDP-adjusted, this is around four times bigger. We all remember how the dot-com boom ended: they overbuilt, just as the railways were overbuilt before them. This kind of capex build doesn't have a history of rewarding the people who fund it. It rewards the people who buy the assets cheaply from the originators after they go bankrupt.On where the money will actually be madeAlec Hogg: A property analogy people might understand better is golf resorts. When there's a boom in building them, the first developer often loses money, the second breaks even, the third makes a fortune. You mentioned Apple sitting on a lot of cash. Who's positioned well for this period?James Gubb: There are lots of areas within AI you can invest in, but I'll go slightly tangential: in the end, I'd put my money in energy. Through massive competition, the cost of compute is going to fall to roughly the cost of energy plus the depreciation of silicon. So the real bottleneck ends up sitting at the energy supply level, and that's where I'm putting my money, because the valuations on those stocks are, frankly, cheap. As for the current AI froth itself, you need to be a real specialist to pick winners there. Something like open routing might be worth a look, CoreWeave is an example.Alec Hogg: When you say energy, from a South African context we'd think of Sasol as an energy company. Would that qualify?James Gubb: I'm thinking more nuclear, but yes, plain old energy suppliers generally. If this exponential trend continues over the next 10 to 15 years, demand from data centres will start to matter. Right now it's immaterial, and it'll stay close to immaterial for the next three to five years. But by year seven, eight, ten, twelve, you're looking at a material increase in demand just from compute power requirements. And building a new power plant, with nuclear still the greenest option to my mind, takes 10 to 13 years. So there's going to be a massive supply gap precisely when demand can no longer be met. That's the end result of this AI wave, and one of the smartest ways to play it is through that energy bottleneck.On the competitive landscape: US vs ChinaAlec Hogg: On the US side we've got Claude from Anthropic, which seems superior to ChatGPT, the original pioneer that the retail market first knew, and Gemini from Google. But the Chinese haven't been sleeping either, we've got DeepSeek and now Moonshot, reportedly better still. Does that make things harder for Anthropic and OpenAI as they look to come to market?James Gubb: Correct, the Chinese are certainly not sleeping. They've been somewhat stymied on top-end chip access, but they've simply got smarter. This whole market gets eroded from the bottom, and I go into this in detail in my notes. Top-end models get undercut by their own previous iterations, which don't disappear, they still do very good work making companies more efficient, often at a tenth or even a hundredth of the price. I don't think that's sustainable, especially with Chinese competition that also has an energy cost advantage over the US, though that's a longer-term issue. The Chinese are throwing money at AI and, quite perversely, have become champions of open, free access, releasing open-weight models to the world, while the Americans are trying to restrict all of this. There were rumours this week of the US government banning open-weight Chinese models. If that happens, it's one of the more short-sighted moves I can think of, exactly the wrong direction for America to take.One more thing: the rate at which frontier models are being released is already close to one every week or so. The question was put to me: at what point do we see frontier-level model releases at a rate of roughly one a day? The answer was January, seven months from now. That's what exponential growth does, it's hard for us to process with a linear mindset, but it genuinely boggles the mind.Alec Hogg: If I'm a banker lending to Anthropic, who own Claude and are currently top of the pile, that competitive pressure might make me more cautious about how much I lend. What about OpenAI and ChatGPT? We've already seen some restructuring from Microsoft there.James Gubb: They are. They're conserving cash as best they can, sidestepping hyperscaler obligations much as Apple has done, in effect stepping back. I'm not entirely sure what they'll do next, but they're clearly trying to reduce their exposure to spending on new large language models.Alec Hogg: That's a significant signal. If anyone has insight, it's Microsoft, who backed OpenAI from day one and are now stepping away.James Gubb: Agreed.Alec Hogg: What about the postponed OpenAI listing? It was expected this year and has apparently been pushed to next year. What does that tell us?James Gubb: It's fairly obvious: they're not able to raise capital at a valuation that makes commercial sense to them. It's that simple.Host close — Alec HoggThat was James Gubb, and I don't think you'll look at the AI story quite the same way again. The idea that real money will be made not in the flashiest AI names but in unglamorous energy suppliers powering the data centres, that's the one that stuck with me, and it's exactly why I'm irritated we didn't hold onto Sasol when we moved across to the [portfolio name unclear in audio — please verify] portfolios. It's done well for us before. Let's wait for a possible pullback after the financial results, and perhaps a softer oil price, before adding Sasol back to the portfolio. But long-term, I like the thesis.And as mentioned, come along to the BizNews Conference, you can rub shoulders with Sasol's chief executive and hear the inside story firsthand.That's BizNews Edge for today. Thanks for spending your time with me. Head to BizNews.com for the full James Gubb interview in Director's Cut, plus today's full market coverage. I'm Alec Hogg, until tomorrow, cheerio.