The AI boom is still booming — at least on paper. But this week handed us three stories that, taken together, tell a messier truth: the gap between what the industry promises and what it actually delivers is getting harder to ignore.

Zuckerberg’s Honest Moment

Meta’s CEO held an internal town hall on Thursday, and according to Reuters, it wasn’t exactly a pep rally. Mark Zuckerberg reportedly told staff that AI agent development hadn’t “accelerated in the way” the company’s leadership had expected. In other words: the technology isn’t moving as fast as the hype suggested it would.

That’s a significant admission. Earlier this year, Meta cut roughly 8,000 employees — about 10% of its corporate workforce — and moved another 7,000 into AI-focused roles, including a unit called Agent Transformation, according to TechCrunch. The logic was that AI was going to reshape how the company operates, and Meta needed to get ahead of it.

Zuckerberg reportedly acknowledged those cuts weren’t handled as cleanly as they should have been, and that the promised benefits of the new AI-heavy structure haven’t materialized yet. He told staff to expect improvements in the next three to six months. Meanwhile, some engineers inside Meta’s AI unit have described the experience in far harsher terms, with multiple reports painting a picture of a demoralizing environment.

Meta is expected to spend up to $145 billion on AI infrastructure this year, per Reuters. That’s a staggering bet on a transition that, by the CEO’s own account, isn’t going to plan.

The Electricity Bill Is Real

While the AI payoff remains uncertain, the costs are very much arriving. According to Ars Technica, Google reported that its electricity consumption jumped 37% in 2025 — the largest single-year increase in the company’s history. Since 2019, Google’s total power usage has grown by more than 250%.

To put the scale in perspective: Google’s data centers consumed more than 42 million megawatt-hours of electricity in 2025 alone. Ars Technica notes that figure rivals the electricity usage of entire countries like New Zealand, Denmark, and Nigeria.

Google says it offset the consumption spike by purchasing clean energy, and it did manage to reduce its direct operational emissions by 2% year-over-year. But its supply chain emissions — from manufacturers and suppliers operating on carbon-heavy grids in Asia-Pacific — grew by 25%. Total emissions, by the company’s own accounting, rose 18% between 2024 and 2025.

Google isn’t alone in this buildout, but it’s the clearest data point yet on what training and running AI at scale actually costs the planet. The company’s own sustainability report admits that its AI infrastructure is “currently accelerating faster than the grid is decarbonizing.” That’s a candid line worth sitting with.

When Sandwich Shops Pitch AI

If the Zuckerberg story is about overpromising internally, and the Google story is about real-world consequences, the Jersey Mike’s story is about something more absurd: the AI label as a magic investor word.

As TechCrunch’s Julie Bort reported, Jersey Mike’s — the submarine sandwich chain — mentioned artificial intelligence 22 times in its IPO filing. The company sells sandwiches. It even included AI in its investor risk disclosures, with language amounting to little more than “we are beginning to use AI technologies in our business.”

For context, Bort notes that weather was mentioned five times in the same document. Lightning — which actually struck one of the chain’s Texas locations in 2021 — wasn’t mentioned once.

It would be easy to laugh this off, but it signals something real. When a restaurant franchise feels compelled to flag AI as a material business factor in securities filings, it tells you how thoroughly the word has colonized investor expectations. Companies aren’t just building AI — they’re performing it, because that’s what the market currently rewards.

The Pattern Is Hard to Miss

None of this means AI is a fraud or that the technology won’t matter enormously. It very likely will. But this week’s news is a useful corrective to the relentless optimism that’s dominated the conversation.

The CEO of one of the world’s largest AI spenders is telling his own employees the timeline is slipping. The energy demands of the buildout are already straining grids and emissions goals. And the hype has spread so far that it’s showing up in sandwich shop paperwork.

The real costs — financial, environmental, and human — are arriving ahead of the promised returns. That gap won’t stay hidden much longer.