On Monday morning President Cyril Ramaphosa wrote to tell us that South Africa is building a “secure and inclusive digital future”. The evidence, apparently, is that Google has held a summit in Joburg, Amazon is spending billions on cloud infrastructure and Microsoft is building more hyperscaler capacity. There will be hubs, accelerators, training programmes and, inevitably, a R3m innovation centre. Sorted, then.
But I couldn’t get past the uncomfortable contradiction. Having spent several paragraphs extolling our growing dependence on foreign cloud and AI companies, Ramaphosa then warned that sovereignty increasingly depends on a country’s ability to “secure its data, develop its own digital capabilities and exercise meaningful control over the technologies on which its economy depends”. That problem I’ll come to later.
The part of the debate that interests me is whether the vast amount of capital now being committed to the sector will earn an acceptable return and, more importantly for South Africa, who will own the value created?
I put this to Hendrik du Toit recently because, as he is co-founder and CEO of Ninety One, I consider him one of the smartest capital allocators around, and because he has spent several decades watching investors confuse technological change with the prices they are prepared to pay for it.
His answer is more nuanced than the usual Silicon Valley liturgy.
“The influence of AI, and the pace at which AI is evolving in our society … is going to be profound,” he says. “I think we are in a moment that is much more profound than when the internet was first founded and commercialised. But that doesn’t mean that all AI assets should trade at very high multiples.”
On Tuesday Samsung reported that its second-quarter operating profit rose 19-fold to 89.4-trillion won ($58.5bn), more than it earned in the previous three years combined, as the AI data centre buildout drove memory chip demand and prices sharply higher. Revenue more than doubled.
And yet investors promptly erased more than $100bn from its market cap.
There, in one trading session, was the entire AI tension of astonishing profits today and a market already worrying about how long they can last.
Will the hyperscalers get a return on their huge investment — or will this all end in tears?
— Hendrik du Toit
The scale of the wager is becoming difficult to comprehend. Amazon, Microsoft, Google and Meta alone are expected to pour nearly $600bn into data centres this year as global spending on the infrastructure needed to feed the AI boom pushes through the $1-trillion mark. Investors have mostly applauded this, because there is no easier way to look old-fashioned than to ask for evidence of a return.
Du Toit asks the question more directly: “Will the hyperscalers get a return on their investment — their huge investment — or will this all end in tears?”
He is careful not to declare a bubble. He has been around too long for that sort of radio-and-television punditry. Manias, he notes, can continue for a very long time.
But he adds: “There will probably inevitably be periods of disappointment.”
That’s asset manager language for somebody eventually discovering that free cash flows and profit still matter.
Then Palantir CEO Alex Karp wandered into the argument on various US media platforms last week.
Karp’s complaint about the frontier AI companies is essentially that businesses are paying with the data and institutional knowledge they feed into the systems. Businesses are essentially feeding their IP into OpenAI and Anthropic and hoping that some version of that accumulated knowledge doesn’t return to the market as a product available to everyone else.
Now, I am not under the illusion that Karp has joined a monastery. Palantir happens to sell precisely the sort of controlled deployment he thinks companies should buy instead. His attack on the frontier labs is therefore about as disinterested as a BMW salesman warning you about Jetour depreciation.
Still, it’s an important and uncomfortable point he raises. Token pricing measures usage. It doesn’t measure value. And corporate South Africa has become rather too comfortable confusing the two.
Du Toit is refreshingly candid about Ninety One’s own experience.
He visited OpenAI in 2023, returned convinced the technology would fundamentally change the business, and encouraged employees to experiment. Ninety One was liberal with access to OpenAI and Claude. Most of the firm, he says, is now “AI literate”.
But when I ask where the hard benefits were, he spares me the usual consultant speak.
“I haven’t seen any tangible projects — I mean, here and there, yes,” he says. “Hopefully in about a year’s time I think we will show very tangible benefits and very tangible productivity gains.”
That is the most honest AI strategy I have heard from a large company.
Du Toit also makes the observation that most companies would prefer to ignore: “Your data has to be really well organised internally.”
And this is where I become more concerned about South Africa than about whether Samsung is trading on the right multiple. Let the chipmakers fall where they may.
But for South Africa Du Toit sides with the president.
“The Americans are monetising the data of the developing world through our participation in Google, OpenAI, Anthropic and others. They take it and they don’t pay us for it, and they monetise it and sell it back at us.”
He believes South Africa needs a data strategy and a “control framework for our own data”.
I think both Du Toit and the president are right, though I can already imagine what we are likely to do with the idea. We will appoint a task team with representatives from six departments, three regulators and someone with “fourth industrial revolution” still in their LinkedIn biography.
A strategy document will appear after 18 months. By then we will have achieved data sovereignty by renting it back from a US hyperscaler.
Would you like to comment on this article?
Sign up (it's quick and free) or sign in now.
Please read our Comment Policy before commenting.