The company raked in $13.5 billion in revenue since May, it revealed in its Q2 2024 earnings, with the unprecedented demand for its generative AI chips blowing past any difficulty it might have had selling desktop and laptop GPUs into a shrinking PC industry. Data center accounted for a record $10.32 billion of that revenue, more than doubling in just one quarter, and Nvidia made $6.188 billion in profit as a result — up 843 percent year over year. And while gaming is more than a billion dollars short of pandemic highs, it was actually up 22 percent year over year to $2.48 billion in revenue, too.
I don’t really post about financial results anymore – the amounts of money “earned” by tech companies are obscene and utterly destructive – but I do want to highlight NVIDIA here, if only to be able to link back this a few years from now after the “AI” bubble has popped.
Thom Holwerda,
I think you are too optimistic (or is it pessimistic?). AI is likely not only to have staying power, but to grow in influence. It might take a while for AI to gain traction everywhere, but the financial incentives behind replacing employees aren’t going away. Once businesses see how increasingly proficient AI can help cut down employee costs, my gut feeling is that it’s going to be prohibitively expensive to hire humans to replace the AI. Some companies will prefer human employees, but capitalism is cut throat, and it’s those that are most profitable that grow while those that are less profitable get squeezed out of the market and are bought out.
So the question of AI’s ultimate viability doesn’t really come down to how we feel about AI so much as whether the companies who decided to replace humans with AI are more profitable. I think it’s safe to say that employee costs keep going up while AI is becoming more accessible and there’s no indication this trend will change.
I disagree, i think AI is indeed a bubble. but like all bubbles, what’s inflated doesn’t just go away.
Lets compare this to the Dotcom bubble, In the late 90’s thousands of new dotcom startups were popping up, and garnering millions in capital from investors, with very little revenue to show for it. Eventually, the investors realised dotcom startups weren’t all they were cracked up to be, with many running out of capital and going bankrupt.
The same can be seen in AI companies. Runway (literally the first result searching for evidence) was recently valued at £1.5billion from investor funding (https://www.reuters.com/technology/ai-company-runway-valued-15-bln-latest-funding-source-2023-06-29/), however, as of december 2022, it had $1million in revenue (https://research.contrary.com/reports/runway?head=;;traction). The crazy valuation vs revenue (revenue, not profit, BTW) is a clear sign of a “bubble” IMHO.
The problem is, there’s a lot of people out there claiming AI can do a lot of things that it can’t. It’s not a sentient, free thinking being, it’s just a very well trained model, and people are claiming that it can “write Shakespeare” or “paint Picasso”, or “run a multimillion dollar company”. It might make a good fascimile of Shakespeare or Picasso, but it’s not the same as what a real human can make. And as for running a multimillion dollar company, maybe we should get AI to run Twitter.
AI won’t go away, but there will be a large amount of losers when the bubble pops, and AI settles down into a more established field.
The123king,
Well, in the context of AI though it’s a bit different though. The more milestones that AI reaches, the more money that companies can actually *save* using AI over hiring humans. The normal economic pressures that bring about other collapses are when companies can’t spend the money to prop up the bubble any more. But with AI displacing human labor, the exact opposite could happen: *humans* may get the boot because AI is actually saving money.
I think many of us are in for a rude awakening that human employees may be the bubble today. When corporations hit hard times and start running out of money, it may be the employees that are seen as most expendable instead of AI. After all it would make much less financial sense to replace the AI saving corporations tons of money.
I’d say AI is still only niche today, but as it continues to evolve, most office jobs could become redundant. AI will do the job faster, 24×7, have fewer mistakes, all while being cheaper than an office of employees. I believe that training is currently a bottleneck for AI. It is very expensive to train AI models, but it scales very well compared to humans. Once that training is achieved it’s going to be the humans that are much more expensive to maintain in the office.
Take IBM’s project to use AI to port mainframe code.
https://www.osnews.com/story/136752/ibms-generative-ai-tool-aims-to-refactor-ancient-cobol-code-for-its-mainframes/
There’s mountains of porting work to do, but with human developers you’d be looking at trillions of dollars to do it. It makes sense to get the AI to do it, and the results can be tested against running production systems to whatever arbitrary degree of certainty is needed. This is an excellent use of AI.
Some people see where AI is at today and conclude that AI is a joke. But realistically the quality of AI is only going to improve and IMHO we all need to factor this in when contemplating the future of AI.
“James May reacts to AI versions of himself”
https://www.youtube.com/watch?v=Dhou-yXBNYA
This claim was made in the 50’s and 60’s, as computers started making their mark on society. Sure, the hundreds of typists and “computers” (actual job title of people doing maths) found themselves out of a job, but were often rehired or replaced by maintenance crew, programmers, and operators.
Sure, AI probably *will* take some poeple’s jobs, but i don’t think it will be as much of a shift as people claim. Development teams might get smaller, design staff might shrink, but they won’t just “go away”
I think this is a complete misnomer. Sure, AI can help with porting efforts and migrations (like the COBOL example in another topic), but you’re still going to need trained competent programmers, to refactor, comment, and generally “make maintainable” the resultant AI code. This is, of course, if the resultant code is actually of any use.
The123king,
I agree with what you’re saying. Job titles changed, but at least in the past new jobs always opened up. Computers made human workers much more efficient.
I will note that, at least in the US, employee wages and benefits have not tracked with their efficiency gains. Employers have been disproportionately rewarded for the advances in automation and computing.
https://www.weforum.org/agenda/2020/11/productivity-workforce-america-united-states-wages-stagnate/
AI will probably not take more prestigious jobs, like project managers and salesmen who are well connected. But for heads down back office workers….AI could pose a huge threat for those kinds of workers. People want to believe they are irreplaceable but regardless of how they feel it’s going to be simple economic calculations in the future that ultimately determines the fate of the workforce.
Unlike with previous technological revolutions, becoming more educated doesn’t provide the same hedge against AI as it used to be against general automation. AI will be competing for educated jobs. Fortunately we’re not there yet, but I don’t think people are truly ready for the repercussions.
How so? Nobody forces you to buy those datacenter GPUs, and we are not even talking about a consumer product here, it’s about companies making an investment as they see fit.
Sorry Thom, this line reeks of “someone is making lots of money and that someone isn’t me”, disguised as moral indignation.
Even though I have reservations against nvidia (locked down drivers, feature differentiation on the same chip by customer segment, severe lack of RAM on new GPUs, etc), I don’t think calling the earning obscene is correct.
They are legitimately the leader in datacenter AI chip technology, surpassed by Google’s own (internal) TPU architecture, but those are not for sale:
https://www.semianalysis.com/p/nvidiaopenaitritonpytorch
If you are going to train large models you need at least something like an A100 (which is not only “hot” but is practically useful). And they go from $5,000 a piece: https://www.ebay.com/itm/394740799900 (You need multiple in practice with also a specialized interconnect between them).
And H100, the new version, is almost impossible to acquire at the moment.
Don’t mix this up with the “crypto” craze. Of course “AI Boom” will also subside from the peaks. But the continued baseline demand will be there.
Exactly the point i’m trying to make. Sure, AI is a bit of a bubble, and it will pop, but AI will exist in a more feasible capacity. The dotcom bubble popped, and Ebay, Amazon, social media and search engines are as big as ever.
The123king,
Sure, while AI has lots of hype like crypto did, it is substantially nothing like crypto. AI could play out very differently from the “bubbles” we are used to. I agree that eventually we’re going to hit a recession, but companies looking to cut expenses may not cut AI spending like you expect. Assuming the AI is actually capable of replacing other employees, then AI could actually benefit and grow in a recession. Companies looking for ways to cut costs might find that AI enables them to replace their most expensive assets,. I predict it’s going to hard for companies to move in the opposite direction since that will cost them more money.
As the old saying goes “if you want to make money during a gold rush, sell shovels”
SUN had made lot of money during dot-com boom, then it had huge issues with surviving after that. We will see how it will go for Nvidia.