Tech Tonic | OpenAI, Nvidia and a $100 billion desperate hype pivot

The OpenAI and Nvidia deal is pristine financial engineering, one designed to buy time and mask the plateauing reality.

Sam Altman believes the artificial intelligence (AI) industry is bottlenecked by compute, and that the entire industry as well as OpenAI is in no position to widen the scope of services they offer. Altman estimates the need for 10 gigawatts of compute over the next couple of years if the world wishes to have AI research a cure for cancer research or offer free education to everyone on earth. The self aggrandising tone, also simply known as boasting, means an attempt to prove one’s superiority by recounting accomplishments so that others will feel admiration or envy. The AI technology that is there currently, and the limitations of which have been highlighted in this summer of discontent, clearly tell us AI companies are in no position to find a cure for cancer. Or provide free education to everyone on earth.

Analysis shows that at current scaling trends, training a large model by 2032 would require resources equivalent to 2.2% of US’ GDP. (Lennart Heim)

Cue, Nvidia announcing they will invest $100 billion into OpenAI to help finance what will be “at least 10 gigawatts of AI data centers with NVIDIA systems representing millions of GPUs for OpenAI’s next-generation AI infrastructure.” Basically, Nvidia is investing in OpenAI, money that OpenAI will use to pay Nvidia for its hardware. If you are wondering that this must be some sort of circular economy between the AI bros, you’re absolutely right. This is Nvidia taking care to ensure there is ‘demand’ for their AI chips.

Last time on Tech Tonic:Time will tell if Nothing’s AI pivot is revolutionary or delusional

I’d suggest you strip away the superlatives and grand pronouncements, and what emerges is a more troubling picture — two industry entities engaged in an elaborate financial theatre designed to mask an uncomfortable reality that artificial intelligence has hit a wall.

This also ensures OpenAI has funds to be seen paying for new AI hardware. After all, everything boils down to optics and valuation in the world of AI these days. Nvidia didn’t pay this out of its pocket. It is estimated this OpenAI announcement added more than the $100 billion they’re willing to invest, to Nvidia’s market value. Circular economy, where money is printing money. While the rest of us foolishly think AI will research a cancer cure.

The timing of this deal is no coincidence. Ilya Sutskever, co-founder of both Safe Superintelligence and OpenAI, recently said that results from scaling up pre-training have plateaued — a damning admission from one of AI’s most prominent architects. For years, the industry’s gospel has been simple — throw more compute, more data, and more money at large language models, and they’ll get exponentially better. That era appears to be ending.

He isn’t the only one. Google DeepMind’s Sir Demis Hassabis said that any claims that current AI systems have PhD-level intelligence, is “nonsense”. No words minced, and Sir Demis believes we are looking at a window of at least five to ten years before we have the first hints of an artificial general intelligence (AGI) system that won’t make the simple mistakes with high school maths that current models make.

AI researcher Lennart Heim’s analysis (from as far back as 2023 — some folks could see the reality, while AI bros saw a chance to make lots of money) shows that if current scaling trends continue, training a single large AI model by the year 2032 would require resources equivalent to 2.2% of the GDP of the US — an undeniably staggering figure, and an unsustainable trajectory. When your business model requires consuming a significant portion of the world’s largest economy just to train one model, you’re not scaling — you’re hitting physics.

The OpenAI-Nvidia deal isn’t really about breakthrough technology — it’s about moving money in circles to create the illusion of progress and justify impossible valuations. Here’s how the game works. Nvidia “invests” $100 billion in OpenAI, which then turns around and spends that money on... Nvidia chips. Meanwhile, OpenAI gets to point to a $100 billion investment as validation of its astronomical valuation, even though the money will flow right back to its “investor.”

It’s a masterclass in financial engineering that makes Nvidia look like a visionary investor while guaranteeing massive hardware sales to itself. Both companies claim they’re at the centre of the AI revolution and hope to avoid harder questions about whether that revolution has stalled.

I might be completely wrong here, but here’s something to ponder over. Consider the timeline — the first gigawatt of Nvidia systems won’t be deployed until the second half of 2026, expected to be on Nvidia’s Vera Rubin platform. That’s nearly two years away. If OpenAI truly had breakthrough models ready to scale, would they risk waiting that long keeping them in the freezer, to deploy infrastructure needed to train? Is this delayed timeline suggesting this is more about maintaining momentum and valuation, rather than any immediate technological progress.

The focus on infrastructure — data centres, chips, power consumption, is itself telling. When you can’t make meaningful progress on fundamental technology, you pivot to talking about the impressive scale of your infrastructure. It’s easier to wow investors with 10 gigawatts of compute power than to explain why your latest model isn’t meaningfully better than its predecessor. And OpenAI has seen it all, with these less than impressive new models, released after much hype this summer.

By creating the appearance of massive, long-term investment demand, both companies can justify their current market positions to increasingly skeptical investors who are beginning to ask hard questions about AI’s actual return on investment. Kick the can down the road. Delay the uncomfortable questions. Hope for a miracle in the meantime. Cash in at some point and laugh all the way to the bank.

The sooner we stop mistaking elaborate financial engineering for actual engineering, the sooner we can get back to realising what AI is and isn’t capable of. The fact is, the emperor has no clothes. Perhaps, just visions of a very impressive data centre.

Vishal Mathur is the Technology Editor at HT. Tech Tonic is a weekly column that looks at the impact of personal technology on the way we live, and vice versa. The views expressed are personal.