Don’t trust Nvidia

I have bought Nvidia products for 30+ years. I personally benchmarked every generation of their products, trained my first small neural networks 25 years ago when nobody was talking about “AI” (they were just discovering what the internet is), last year I published a novel research paper describing a benchmark framework to evaluate models on truthfulness, and today I fine-tune the most complex AI models as a software engineer. I know this industry much more deeply than most financial analysts.

Nvidia has uncontested leadership in their field. But I lost trust in Jensen Huang, co-founder and now ex-CEO, around 2020, during the crypto rush, and subsequently AI, for his lack of integrity during these years, for misleading investors, and for completely abandoning his historical customers by organizing unfair and unsustainable scarcity as well as lying about his real strategy.

For more than 30 years, since the 1990s, people like me bought its products every year, and we financed Nvidia’s development. Then, during Covid and the subsequent crypto hype, followed by AI hype, Jensen Huang decided to betray his historical customers by tripling its prices, to jump on the hype wagon to make more money in the short term, at the expense of his historical customer base. Worse, he illegally lied to its investors by explaining Nvidia revenue came from very stable customers just like me, while most of the profit actually came from unstable crypto miners and subsequent AI hyperscalers. He was fined by the Securities and Exchange Commission for that lie.

But he didn’t stop lying, as he lied again to investors by using circular financing at massive scale, giving money to its clients to buy their products to artificially create fake revenue as well as maintain artificial scarcity. We are talking about 25% of Nvidia that is completely fake, even if circular financing is legal in itself when a manufacturer helps a retailer boost its operations. He abused the law as much as he could, way beyond the original spirit.

Nvidia manipulated the markets

Connected to that, he uses Nvidia cash flows to buy all the possible capacity in the few factories in the world capable of building the latest silicon generations, to be able to artificially maintain extremely high silicon prices, while 20% of its GPUs still sit idle in warehouses as we speak, so the demand is not as high as Nvidia wants to pretend it is. Again, in a free market, it’s okay to buy capacity, or even overproduce to anticipate future demand, but when you do it to impede all competition when there is not enough real demand to cover it, it becomes illegal market manipulation. This tactic was made illegal in the first place because everyone buying a computer, laptop, smartphone, or using online services pays the price indirectly, as tech product inflation is close to 100% per year in some categories due to his personal choices. Most humans on earth pay him a tax well beyond the value he created, because of artificial manipulation of the physical supply constraint. It’s the story of a guy who initially digged a water well to sell water (good value creation), then decided to buy all the water of all neighboring wells to triple water prices (bad value creation).

Of course there is a real jump in demand due to AI, but that jump wasn’t enough to max out production capacity, since 25% is self-financed, and 20% of GPUs are idle as we speak. We had a margin of safety in production, more than capable of absorbing all AI demand. But maxing out the world’s production capacity artificially allowed Nvidia to triple prices.

Of course, every business owner should look to maximize profit, but some methods are short-termist while others are plainly illegal. Eventually, you lose your well when you decide to abuse your customers.

So I really feel like Jensen Huang’s wife: loyal for 30 years, then not only betrayed by being lied to and dumped, but also defrauded through illegal market manipulation, to add insult to injury. Just like if Jensen was asking for an alimony pension to his ex-wife after cheating her.

So, today we learn that he has not just one but 3 mistresses, including one potential Chinese spy? I didn’t know about that until recently, but that’s completely illustrative of the person. What should we expect from someone betraying his supposedly most important person on earth? In long-term investing, pristine integrity is not a nice-to-have, it’s mandatory, as everything else relies on it. Just like in a restaurant kitchen, if the place is dirty, you can expect much more than one cockroach.

The world is over-building AI capacity

But that’s not all. No. Nvidia can make great products and still have a dishonest CEO; the company would actually fare better without him. But there is another structural problem impeding long-term prospects: the world is currently overbuilding AI capacity. Just like the railroad and internet revolutions, we always build 10 times more rail lines and fiber lines than people can actually pay for at the moment, overestimating the real demand people can pay for, and then the bubble pops when there is a shortage of new money, and investors are eventually forced to sell for less than 10% of what they originally paid to build it.

AI is no different. There is no shortage of developers capable of burning billions of tokens per month, myself included, but none actually pay what they truly consume, as they all pay minuscule prices for unlimited packages ($200/month really paid for an actual $30,000/month of tokens burned). Most of what devs consume in AI (and devs are the biggest token consumers) is essentially paid for by naïve investors. That won’t last indefinitely. I know that naïve investors will one day stop paying for my free lunches. As soon as AI companies are forced to charge what their users really burn, most power users will switch to local open-wieght models, and then there is no more need for AI data centers at all. The only reason why devs currently pay $200/month is because GPUs are artificially expensive at $3000/GPU (for what a dev needs). If true token cost ($30,000/month) is compared to the normal price of a GPU ($1000), the math instantly reverses. We are far away from the myth of “AI companies will eventually replace everyone, so, in theory, they can ask for the same amount of money as a software engineer’s salary.” AI won’t replace many people, but it will definitely change the workflow of many people.

What’s left for Nvidia? The regular office workers for whom their current laptop is already more than enough to run small open-weight LLMs to do internet research, draft their emails, search their inbox, organize their invoices, and write clear explanations on any topic. For that, even smartphones are enough. These people will definitely not pay $200/month for what they need, and may not even pay anything anymore as soon as prices rise and they switch to local solutions. Today you can run an older version of ChatGPT locally on your phone or laptop for free with just one-click install. Few people bother doing it at the moment as AI is either completely free or super cheap, but they will immediately switch if they are asked for more money. And no, very few people actually need frontier models, as most people are not top-level scientists or engineers, or even dealing with very hard problems in the first place. How many of you more often ask AI to draft emails than to solve Navier-Stokes and Black-Scholes equations? Clap your hands if you asked AI what Navier-Stokes and Black-Scholes equations are haha

So there can be multiple true statements at once:

  • AI is a true revolution just like the internet.
  • AI really affects virtually everyone, and will for many years/decades ahead.
  • Nvidia manipulated the market with unsustainable prices and artificially created demand and scarcity.
  • Jensen Huang is a pathological liar who betrayed even his wife, on top of the whole world.
  • The world doesn’t need Nvidia to run most free open-weight AI; there are many alternatives, if only Nvidia wasn’t manipulating the world’s manufacturing lines.
  • Nvidia will eventually come back to its pre-Covid revenue, and so will its stock price. I expect the company to eventually trade at 10% of it’s today value, just like Cisco back in the days.
  • But since the world hasn’t yet realized the full extent of the situation, I would say the stock price may just “dip.” Maybe 20-30%, maybe it will recover, until the inevitable fall later.
  • As economist John Maynard Keynes said, “The market can stay irrational longer than you can stay solvent,” which means the bubble can continue to grow for years, or pop now, especially with Anthropic (Claude AI) IPO.

But I fear the AI bubble may continue to grow

Larry Fink, CEO of BlackRock, the biggest fund in the world with 15,500 billion dollars under management, announced that he was working closely with Jensen Huang to create “Data Center Asset-Backed Securities,” on the same model as the 2008 financial crisis subprime products, of which he was one of the creators.

Larry Fink declared that he believes there is no AI bubble, and also never wrote a single line of code in his life, let alone on AI. It’s like telling the world “I know nothing about space rockets or physics, but I can assure you that we will travel faster than light next year. Trust me because I was the man behind one of the biggest financial crises of modern times”. It’s as cynical as if Adolf Hitler were explaining that he loves Jews so much that he created holiday centers for them.

Everyone who has savings in ETFs, pensions, or various investment products (like most French “assurance vie”), all around the world, will pay the price. Why are our fund managers stupid enough to buy that crap? Because they have a mandate to buy products with good financial notation and short-term performance, exactly what these debts are at the moment. These fund managers are mostly paid for how much money they manage, never taking the loss with you, even continuing to charge for your money management as things go down. So of course they do what their mandate tells them: buy shit that shines before being flushed down the toilet and protect themselves behind “there was no way to know the market would collapse, all ratings and consensus were good.”. Just like after every crisis, following years of egregious exuberance, being willingfully blind to all red flags to keep profiting from it.

So, as tragic as it may sound, I unfortunately believe that there is still room for Nvidia and the AI bubble to grow, to accumulate more debt from innocent everyday people, who have no idea that their savings or pensions are being used to gamble against their will.

Nvidia, along with Tesla/SpaceX, are the two companies that led me to distrust passive investing in index funds because the vast majority of them are infected by these hype-driven scams. That’s why I invest only in individual companies, after a very deep industry analysis of each of them, staying away from index-investing and not letting other people manage money for me, even for pension.

Special mention to China, adding gasoline on the fire

And China? What happen with China? China massively invested to develop open-weight LLMs first because they didn’t want to lose the tech race as they did with microchips, but also because they are putting gasoline on the US bubble, soaping the slide to help America fall and hurt itself even more badly. Indeed, free models are exactly what make Anthropic and OpenAI essentially worthless, and free models can also run on potatoes, the proverbial Pareto 80% of the job done with only 20% of the ressources, making Nvidia expensive GPUs way less valuable, just like China killed the clothing industry, the steel industry, and many others before. Free Chinese models can also slightly reframe history in their favor, so they can brainwash millions of people in the West at scale for free. Triple bonus for China. Fortunately, we also have European free LLMs like Mistral, otherwise we would be screwed in the EU.

Disclaimer: I do not own any Nvidia share, short, or option on them, in any direction, and do not plan to have any in the future, at any price. I just stay away from them, and I sincerely wish they would quit for good such practises. Once trust is lost, it takes years of corrective measures to be built back, if ever.