Oh boy. Over the past few days, an article has been doing the rounds as a testimonial for the ludicrousness of AI versus human costs. From that piece, one specific quote (originally from another article) has gone viral and is now under scrutiny all over the internet. Fortune, Tom’s Hardware, TechSpot, Futurism, Yahoo Finance. Reddit. Twitter. LinkedIn. The whole AI-and-jobs cycle this week is resting on this one comment.
“For my team, the cost of compute is far beyond the costs of the employees,” Bryan Catanzaro, Nvidia’s vice president of applied deep learning, told Axios.
Catanzaro runs the team doing advanced research to make advanced models that work on advanced chips. So of course his compute bill is going to be bigger than his payroll. It’s the most obvious place where compute costs more. Anyone surprised by that hasn’t read the org chart.
Most companies are not building frontier AI. They are running customer support, payroll, dashboards, marketing, and code. For them the bill looks nothing like Catanzaro’s, and pretending it does is just lazy. The trouble is that in our velocity-driven information economy, nuance gets lost.
I have been watching companies make announcement after announcement, and all I see are the sins of their past. Meta is laying off another 8,000 people. So many of them are coming from ill-fated projects like Reality Labs. It is a company that got bloated, over-hired, and, believe it or not, didn’t have a clear idea of its own sprawl.
For context, Meta has cut close to 36,000 jobs since 2022. The first rounds were honest about pandemic over-hiring. The 8,000 being slashed in May, plus 6,000 open roles closed, are now being sold as an AI story. The company still has 78,000 employees, give or take a few hundred. Same problem, new script.
Meta lost $4 billion in Reality Labs in the first quarter of this year alone. CNBC reports the unit has accumulated over $80 billion in operating losses since late 2020. That tells you how bad things are from a management and strategy standpoint. I don’t even need to call an insider to confirm the mess.
As I have said before, the greatest myth in this business is that the technology industry is some kind of mystical master of efficiency and modern management. Everyone running the Block & Tackle play is using the AI cost story for cover. It lets a CEO point at the future instead of admitting the sins of over-hiring during the pandemic, corporate bloat that should have been cut two years ago, and management strategy nobody can explain in one sentence.
Catanzaro’s quote is true. It is also specific. One specialized team at the most important chip company of the moment is not how AI will weave its way through the economy. For that, look at the recent MIT research study.
We estimate that, in 2024-Q2, AI models successfully complete tasks that take humans approximately 3-4 hours with about a 50% success rate, increasing to about 65% by 2025-Q3. If recent trends in AI capability growth persist, this pace of AI improvement implies that LLMs will be able to complete most text-related tasks with success rates of, on average, 80%–95% by 2029 at a minimally sufficient quality level.
When I read that, my own thinking is that this is all part of a continuum. Whatever you think of it, we are moving down this path. Using software to train itself, to streamline operations, as both Block and Meta are doing, is what would have been called “productivity gains” in the non-tech white-collar world of the 1990s and 2000s. As I have written, for now, software is eating its own.
I often remind myself of the wise words of Bill Gates, who once said, “Most people overestimate what they can achieve in a year and underestimate what they can achieve in ten years.” The same is true of technologies. I will close with the words of Roy Amara, of the Institute for the Future:
“We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.”
As for Bryan Catanzaro: enjoy the meme, get back to work, and ignore the press. Some of us understood exactly what you were saying.
April 30, 2026. San Francisco
Log in
