Nvidia exec says AI is more expensive than actual workers — yet some companies don't see the extra costs as a negative
It's easy to point and laugh, but the picture might be more nuanced than it seems.
The viability of the circular economics of the AI world has long been debated, but most discussions are about the movement of cash within the industry. Slowly but steadily, though, reports are coming in from executives, such as an Nvidia executive and Uber's CTO, that they are starting to realize that the price of tokens may well outweigh that of plain ol' human brain cells. In the face of mass layoffs to make room for AI agents, that fact is seemingly so ironic that Alanis Morrisette is probably writing a song about it.
There are all sorts of enterprise pricing arrangements for LLMs, but for most standard users, the price of a standard AI assistant is $20 a month for a standard plan, and $200 for the pricier, fully-featured version. Token-based pricing is where the real spend is, usually in the form of coding assistants like Claude Code or GitHub Copilot, as well as automation agents with planned tasks of varying complexity that usually run repeatedly on a schedule.
The continuous nature of those sessions requires a constant trickle of money, as many firms' bean counters have come to realize. Bryan Catanzaro, Nvidia's VP of applied deep learning, recently told Axios that "For my team, the cost of compute is far beyond the costs of the employees", quite an interesting statement from the company selling the shovels for the gold rush.
That perspective is shared by Uber's CTO Praveen Naga, who "[went] back to the drawing board because the budget [he] thought [he] would need is blown away already" as of two weeks ago. Likewise, Swan AI's Amos Bar-Joseph posted a while back on LinkedIn about how proud he was about a $113k bill from Anthropic (makers of Claude) for a four-person team.
Oversimplified math pins that amount that at $28k per person per month, which is likely more than each person's monthly wages. Jokes abound right now that "companies have discovered jobs again," and the humor is backed up by a 2024 MIT study stating that 77% of the time, it was preferable to have humans do the work.
And yet, the popular sentiment of "I told you so" may be partially misguided. Many CEOs see these bills as a good thing, as it means that their employees are making progress on large-scale automation — in short, driving innovation, at least supposedly.
Uber's Naga said that 11% of its live updates to code are written by AI agents, and reportedly envisions said agents taking on the roles of software engineers. To wit, "the vision for [him] as a CTO is to transform from software engineering to [AI] agent software engineering." Nvidia's own Jensen-Huang seemingly believes that his engineers' productivity is measured by their spending in AI tokens, wanting a $500k salaried engineer to spend at least $250k worth of tokens per year.
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
It is true that many companies are probably finding out the hard way that tokens are more expensive than the workers they're supposedly replacing. However, a business spending additional millions in tokens in order to permanently automate the majority of its workflows might net a long-term win, one that likely leads to job cuts as said automation nears stability. "Just hire more people" would be an easy reply to that, but people don't work tirelessly 24/7.
The third scenario is a failed investment in AI automation, either due to a lack of a business structure, unsuitability of tools for the tasks, or simply an inability of the business to properly instruct the clankers. Recent studies have shown that the vast majority of companies rushing to implement AI without a strong plan end up experiencing massive losses on those initiatives.
After all, any developer will tell you that it's really easy to build a product if the customer can describe it accurately and the specifications don't change, a fact put succinctly by Edward Berard a long while back: "walking on water and developing software from a specification are easy if both are frozen."
Whether all this extra spending on tokens in addition to workers is a temporary tandem expense next to salaries as AI learns the ropes and takes over, or whether it's a complementary expense as AI becomes a force multiplier for those employees, will remain to be seen, and it's likely contextual. But it's almost certain that layoffs will continue as companies feel out, and finance, this new era of tech.
Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.
Bruno Ferreira is a contributing writer for Tom's Hardware. He has decades of experience with PC hardware and assorted sundries, alongside a career as a developer. He's obsessed with detail and has a tendency to ramble on the topics he loves. When not doing that, he's usually playing games, or at live music shows and festivals.
-
Marlin1975 File this away under DUH. None of the AI numbers make from a cost to provide to what you get from it.Reply
OpenAI not meetings numbers so much their fall up lacky CFO even is getting worried.
But they need to pump till their shares vest/go public and then can dump their stocks. -
PEnns An expected outcome that was totally unexpected to some!!Reply
As if the over-the- top optimism totally overlooked the facts that expensive AI machinery (and their upkeep and maintenance) and AI centers were....not expensive at all!
What now?? Fire the AI and re-hire the "cheaper" people??
But don't worry, the Tech Bros will still make money. -
Marlin1975 ReplyPEnns said:An expected outcome that was totally unexpected to some!!
As if the over-the- top optimism totally overlooked the facts that expensive AI machinery (and their upkeep and maintenance) and AI centers were....not expensive at all!
What now?? Fire the AI and re-hire the "cheaper" people??
But don't worry, the Tech Bros will still make money.
Can't do that. That would mean the stock options are worthless. Got to keep pumping the AI bubble up till....?
Its why most AI "purchases" are either from other AI companies and/or AI Hardware suppliers. Got keep that money moving so you have no idea where its at or where its from. -
watzupken I think the people at the top of these companies clearly are aware the cost of AI data center investment and maintenance is going to be more than hiring "warm bodies" to do the work. But hey, they don't own the company, so their aim is not about caring for the company. They are more concerned about their pockets. Since investors love AI, they all just triple down on unsustainable investments so that they can inflate share prices, which also contributes to a fatter wallet right? Ultimately when the company goes belly up, shareholders will lose money or they will ask government to bail them out with other peoples' money.Reply -
DS426 ReplyIn the face of mass layoffs to make room for AI agents, that fact is seemingly so ironic that Alanis Morrisette is probably writing a song about it
Lol, that was random. -
usertests If the "AI", however you're measuring a single agent, is doing more work than a human, it could make sense.Reply
Also, the cost of running it will be decreasing, likely rapidly from hardware and model improvements. The cost of the human worker is probably stable if wages are stagnant against inflation/productivity. -
Marlin1975 Replyusertests said:If the "AI", however you're measuring a single agent, is doing more work than a human, it could make sense.
Also, the cost of running it will be decreasing, likely rapidly from hardware and model improvements. The cost of the human worker is probably stable if wages are stagnant against inflation/productivity.
Even if the cost was cut in half, no sign they are anywhere near that, it would still cost more than they are bringing in. If anything cost are going up, not down. So the sales pitch of "cost will come down and more will buy..." is about as solid as a fart in the wind.
And what little they have brought in, from those not in the circle of AI, has been mostly negative. So growth is flat at best when trying to sell those not in the AI bubble. OpenAIs own numbers show that. -
usertests Reply
At this point, I think we'll see the bubble pop, and unprofitable companies swept out of the market and bought up by the tech giants. They will continue their investments over the next few years, launching AI products/services whenever it makes sense or using it internally. The hype cycle will happen again, but tied to real results.Marlin1975 said:Even if the cost was cut in half, no sign they are anywhere near that, it would still cost more than they are bringing in. If anything cost are going up, not down. So the sales pitch of "cost will come down and more will buy..." is about as solid as a fart in the wind.
And what little they have brought in, from those not in the circle of AI, has been mostly negative. So growth is flat at best when trying to sell those not in the AI bubble. OpenAIs own numbers show that.
About the cost, models are becoming more capable at the same sizes over time, and new hardware generations improve perf/Watt and perf/$, or they won't be bought. Memory will naturally become cheaper as new densities are developed. Samsung may introduce a 4F cell structure, and in the longer run (decade), we'll see 3D DRAM.
Those things WILL happen. If your position is that the LLMs and agentic AI are doing nothing useful and are a liability for many companies, I'm more sympathetic to that view. Some companies have been forced to rehire people that they were eager to lay off. -
nimbulan So they're basically spending a bunch of extra money in the vain hope that they can almost completely replace employees in the future and that will somehow translate to saving money even though it's more expensive. I already knew most businesspeople were shortsighted idiots but this really takes the cake.Reply -
Marlin1975 Replyusertests said:At this point, I think we'll see the bubble pop, and unprofitable companies swept out of the market and bought up by the tech giants. They will continue their investments over the next few years, launching AI products/services whenever it makes sense or using it internally. The hype cycle will happen again, but tied to real results.
About the cost, models are becoming more capable at the same sizes over time, and new hardware generations improve perf/Watt and perf/$, or they won't be bought. Memory will naturally become cheaper as new densities are developed. Samsung may introduce a 4F cell structure, and in the longer run (decade), we'll see 3D DRAM.
Those things WILL happen. If your position is that the LLMs and agentic AI are doing nothing useful and are a liability for many companies, I'm more sympathetic to that view. Some companies have been forced to rehire people that they were eager to lay off.
Problem is there are no profitable AI companies, at least those selling just AI. Hardware yes due to AI "free" money being thrown at them.
Google, MS, Meta, etc... are using other funds to prop their AI divisions up while laying off others. OpenAI and other AI providers are not even close to breaking even, let alone profit.
I agree cost may come down some day, but that day is years if not decade+ away.