In response to the “sex-mad tech CEO’s antics” of recently resigned Nvidia boss Jensen Huang, much of the Australian news media is taking the lead from the American (and News Corp) commentariat and getting all From China with Love. As Greg Sheridan at The Australian put it, “It was the honey trap that changed history.”
But Huang’s affair with Chinese entrepreneur Alice Huang (and two Nvidia employees) is the amuse-bouche to a much more substantial story about an AI bubble. With the sneeze that is Huang’s resignation, speculation abounds that the world’s second-most valuable company (valued at around US$4 trillion) has caught a serious cold — one it could share with the rest of the tech sector.
Nvidia is a key organiser of the globalised network of monopolies churning out the trillion-odd microchips and related technologies underpinning the AI boom. It dictates the world’s production schedule as it filters through the different stages of the chip-making process, one spread across Taiwan, South Korea, China, the Netherlands, Japan and the US. Like most of the Silicon Valley tech giants, Nvidia doesn’t actually “make” anything itself, at least not in the everyday sense. It’s oxymoronically “fabless”: a non-fabricating manufacturer.
Nvidia’s clout thus comes from its orchestration of the overall supply chain. But an overarching problem has emerged recently: it has been orchestrating a bit too vigorously, pumping up the market for its products in two eyebrow-raising ways.
First, there’s China. Silicon Valley critic Ed Zitron — who’s been passing out Nvidia-bubble-bursting pins for over a year — wrote this weekend about Huang’s affair scandal: “Everybody wants this to be a localised story about a horny CEO and a dastardly Chinese spy, but what it actually is is a coordinated, multi-layered conspiracy to dodgy US export controls that may very well have been totally ignored if it wasn’t for the espionage angle.”
He’s pointing to the work of Christian Lamarco of Culper Research, who, in his May paper on Nvidia’s “China Problem”, noted that more than 20% of Nvidia’s compute revenue for its must-have proprietary microchips was coming from China. Much of this, Lamarco writes, is funnelled through South-East Asian intermediaries, including Singapore-based cloud computing company Megaspeed, formerly fronted by a Chinese citizen Huang Le — better known now as Jensen’s girlfriend, Alice.
Good business. Except America, under both Donald Trump and Joe Biden, has restricted the sale or other access to China of the money-making, top-grade chips needed for frontier AI. The Trump administration placed further chip restrictions on China today in response to the reporting on Nvidia.
Then there’s Nvidia’s extensive use of creative financial backstops to companies who need to buy its products. Another sceptic, Michael Burry of The Big Short fame, has been pointing to the flaws in Nvidia’s financing plan, which involve providing guarantees to companies that can’t afford its chips to the tune of about US$500 billion.
Why should you care? Because one of the first casualties of Nvidia slumping and the AI bubble popping — or even just slowly leaking air — could well be Australia’s grand plans to use renewable-powered hyperscaled data centres to elbow our way into an already crowded supply chain. In September, Nvidia announced it was working with eight named partners to more than double Australia’s data centre capacity (with full-stack Nvidia products, of course), adding up to 2 GigaWatts by the end of 2027 on top of the nation’s current computing capacity of 1.6 GW.
That’s a lot of jobs, particularly in construction, in an industry already struggling with a declining housing sector. It’s also before Treasurer Jim Chalmers has to start worrying (or begins worrying more) about a slowdown (or even, gulp, a recession) as a result of a broader global AI market crash.
And yet, apart from a few stories from finance reporters, it’s clear that parts of the US news media (and the Australian media that follows them) haven’t really known what to make of Nvidia. They like investigations with the high drama of whistleblowers or leaked documents. Instead, Nvidia critics have been relying on dry, publicly available documents hidden in plain sight.
The Wall Street Journal followed up the China angle on the weekend, with its local reporters digging into just why Malaysia (rather than, say, Australia) has become South-East Asia’s data centre hub: it’s a convenient backdoor for Chinese institutions to access compute power running on the high-grade Nvidia chips that China can’t legally import. It’s seemingly close enough for smuggling those same chips into China, and far enough from the US Commerce Department charged with enforcing the export ban.
In the US telling, Huang Le is a mystery, with her role as “intelligence asset” in the original New York Times reporting upgraded to the far more glamorous “spy” in the more frothy commentariat. But she has her own backstory, with Beijing-based Caixin reporting on her and Megaspeed’s background in Chinese data centre development and her ties to a “shady” Shanghai fund manager.
What commentators think about the China angle is often shorthand for a more profound story about a future with AI. It’s a way of asking: who gets the power in AI and who makes the money? The answers turn on the open weights vs closed models debate. Both China and Huang are team “open”, where you host the AI on your laptop like any other software and build your agents off the back of it. The big US labs OpenAI and Anthropic (and America itself) are team “closed”, wanting models only accessible online through AI tokens sold as a high-priced service.
The company’s financial deals with potential data centre customers have apparently been causing tension inside Nvidia, too, particularly the deals brokered by Huang’s son-in-law Nico Caprez. According to The Wall Street Journal, the risks to Nvidia’s stability triggered a tense senior management meet-up in August where Huang clashed with then chief financial officer (and now acting CEO) Colette Kress.
But while its clout gives it control over supply, buyers can be harder to corral. Now, local resistance to data centres, reflected in this month’s US midterms, coupled with the closing of the China backdoor, threatens to knock all that finely calibrated pre-buying and pre-selling out of financial kilter.
In the shakeout, new players see an opportunity. Elon Musk, who likes to vertically integrate his supply chains under his control, launched Texas-based TeraFab in March to make his own chips. Google designs its own chips, but has them manufactured through the same global network (and a lot of the same companies) as Nvidia. He’s even offered Huang a job to run it.
Now, it’s all up in the air. Expect the chatter about honeypots and priapic billionaires to quickly give way to a more chin-scratching consideration of what the sudden lurch in the AI rollout means for all the rest of us.