The $600 Billion AI Compute Bill is Ultimately Paid by Everyday Workers

The clouds outside are quite thick. At 15 degrees, Shanghai is gloomy, making one’s mood feel a bit heavy as well.

I just messaged a frontend developer friend in Silicon Valley. He hasn’t replied for a long time; I guess he’s also waiting for an internal HR notification email.

A Cruel Game on the Ledger

To be honest, when I was helping my team test a new internal automated workflow yesterday, watching hundreds of data entries—which previously required manual comparison—process instantly in the background, I felt a vague sense of unease.

Mark Zuckerberg actually casually mentioned at the end of January this year, “Projects that previously required large teams can now be done by a single top talent.” Many thought this was just the CEO painting a rosy picture as usual during earnings season, until Reuters leaked the specific numbers yesterday: Meta is brewing a plan to lay off nearly 20% of its workforce, which is about 16,000 living, breathing people.

The hardest-hit areas are content moderation, recruiting, and non-core product lines. Meanwhile, the core AI team’s headcount is fully preserved and is even expanding aggressively.

Simply put, this is to cover the massive $600 billion AI infrastructure investment over the next two years.

People have been talking about AI stealing jobs for a while, but most felt it was still somewhat distant. This time is different; it’s an extremely brutal asset class conversion. Wall Street loves this kind of ROI optimization—directly turning your monthly salary into the depreciation costs of rows of GPUs in data centers.

Thinking About This Deal from Another Angle

If you look closely at Meta’s recent moves, you’ll find that their top management is acting very decisively, even with a hint of a last-ditch effort.

The testing of the Llama 4 model faced quite a bit of resistance last year. The highly anticipated super-sized version, “Behemoth”, was shelved directly due to various reasons. Logically, when facing a technical bottleneck, standard management common sense dictates hiring more smart people to collaborate and tackle the problem. But Meta’s solution is: throw more money at buying compute power, and use internal AI tools to eliminate all existing redundant processes and middle-layer communications.

Just a couple of days ago, they spent nearly $2 billion acquiring an Asian AI startup, along with Moltbook, which focuses on AI agent social networking. Where does this money come from? It can only be squeezed from the salary pool of ordinary, non-hardcore employees.

This brings us back to the replacement logic I mentioned at the beginning. At this juncture, when the marginal output of large models begins to stably cover a portion of human daily work, the irreplaceability of non-hardcore technical roles is being completely leveled. In the past, your project experience was an asset. Now, if your experience can be distilled into dozens of high-quality Prompts or an automated execution package, you are purely a cost center in the eyes of corporate finance.

To Put It Bluntly, This Time Is Different

Many people might compare this workforce upheaval to the wave of Silicon Valley layoffs back in 2022 and 2023.

At that time, Zuckerberg called 2023 the “Year of Efficiency,” and over those two years, they laid off more than 20,000 people. But if you horizontally compare the motives back then, it was actually just the peak of the internet pandemic dividend. Top tech giants realized they had expanded blindly and hired too many people for “exploratory businesses,” so it was purely about cutting expenses and squeezing out the excess fat.

But there’s a problem you need to understand: these 16,000 people this time are not just part of a simple “trimming the fat” efficiency drive.

Amazon actually also just cut about 10% of its workforce—also 16,000 jobs—early this year. The synchronization of these two leading companies in the spring of 2026 is terrifying. They are not removing idle personnel this time; they are removing the gears that once kept the company running normally. These gears are now being replaced by systematic AI services. Examples include basic code reviews, repetitive UI component adjustments, and data cleaning and extraction for daily operations.

In other words, the last round of layoffs was because the company temporarily didn’t need that many people; this round of layoffs is because the company thinks compute power is more useful than humans.

A Few Things I Sometimes Wonder About

Seeing the anxiety in foreign online communities these past few days, I actually feel a bit dazed.

If this extreme trend of shifting from human capital to compute capital continues, will there be a day when a multi-trillion-dollar tech giant actually consists of only a few dozen elite, top-tier architects, plus millions of data center servers running day and night?

I sometimes wonder: when a large company is entirely made up of various AI Agents sending emails to each other, reviewing each other’s code, and confirming requirements documents with each other, does this company still count as a human organization in a sociological sense?

Or, to be more realistic, after we workers sitting in front of computers are optimized away in batches, who will consume the content and information streams efficiently produced by these AIs? We can’t expect AI agents to click on those recommended ads themselves, right? ╮(╯▽╰)╭

Maybe I’m overthinking it. Technological shifts have always been like this: old professions are wiped out, and perhaps in some corner I can’t see, new niche markets are springing up. It’s just that during this transitional period of replacement, the growing pains experienced by those caught in the middle cannot be borne by anyone else.

Anyway, Enough Rambling

As I finish typing these words, the rain outside seems to be falling harder, pattering noisily against the glass. The half cup of iced Americano on my desk has long turned to room temperature water. It really feels like today is not a suitable day to discuss grand narratives.

I’m going to shut down my computer now. I still need to check if my friend has replied to my message.

References:

—— Lyra Celest @ Turbulence τ.