Author(s): Zoom in on AI
Originally published on Towards AI.
Your electricity bill is helping control Bezos’ compute buildout. Elon wants to take the whole thing to orbit. Neither plan has been proven yet. That’s the scary part.
Zoom in by AI 9 February 2026
I write about how AI really works (not the marketing version), where security and reality don’t match, and the economics and infrastructure behind the hype.
Stop me if this sounds sarcastic.
Four companies, Amazon, Google, Microsoft and Meta, are indicating broadly $655 billion In capital expenditure this year to build AI data centres. This is infrastructure spending on a scale that most governments would struggle to implement.
Their stocks fell anyway.
Then Elon Musk did what Elon Musk does: He moved on. They merged SpaceX with XAI in a deal widely described as historic in size and began selling a different answer to the same hurdle. Put a data center in orbit, power it with constant sunlight, cool it without water, skip the grid, skip the permits, skip the protests.
Here’s the thing no one is telling us clearly:
Both strategies are bets against odds you can’t force your way around.
One can bet that a 70-year-old grid can absorb AI-era load spikes without political revolt. The second is betting that you can run industrial computers in an environment that punishes hardware and does not forgive collisions.
And somewhere in between, your monthly bill goes up so the chatbot can respond faster.
Let’s zoom in.
I. The $655B Question: What Are You Really Buying?
Start with the scoreboard.
- Amazon: ~$200B
- Alphabet/Google: $175-185B
- Microsoft: ~$150B (Based on guidance and reporting)
- Meta: $115-135B
it is called ~$655B On the conservative end.
The market reaction tells you that the mood has changed. Investors rewarded spending because spending meant “we’re winning.” Now they want a reliable line from capital expenditures to revenues within a time frame that doesn’t feel like trust-based accounting.
Amazon CEO Andy Jassy argued that the spending was not a wasted move and that demand for AWS was outpacing supply. CNBC Coverage:
Wall Street looked at similar numbers, then saw capital spending exceed consensus estimates and hit sales.
One quote captured the atmosphere: We went from “capital spending sprees” to “show me the revenues.”
And here’s the detail that matters more than the headline number: how much cash was gobbled up by the production.
A breakdown claims that Amazon is reinvesting the vast majority of operating cash flow into infrastructure and that free cash flow has narrowed sharply relative to spending.
This is the new reality of AI infrastructure: Construction business plan.
II. The part they don’t want you to read: Your electricity bill is part of the subsidy
All this computing requires electricity. Not “very”. Industrial, stationary, mixed power.
The problem is that the US grid is not yet built. A widely cited statistic: Much of it was designed decades agoAnd it’s already stressful.
PJM Interconnection, the largest grid operator in the US, is the clearest case study because it serves major data center corridors and millions of people. PJM has warned about reduced reliability margins due to increased demand.
Here’s the mechanism that turns “Big Tech capital spending” into “Your bill goes up”:
When grid operators and utilities purchase capacity and reliability resources to meet increasing loads, Those costs don’t stay within the hyperscaler budget. They are allocated through regulated structures and eventually land on the ground. rate payer.
Bloomberg quantified how data center demand is showing up in purchase costs.
Then it becomes increasingly political.
CNBC says residential electricity prices had already increased in 2025 and are projected to rise further in 2026, while local communities and national politicians begin to consider data center growth as a cost-of-living issue.
That’s why Musk’s classroom pitch suddenly stops sounding like pure cosplay. That’s pointing to a real bottleneck.
Third. Musk’s pitch: “It’s always sunny in space”
Musk’s story is simple: Earth is slow and expensive. Space is fast, scalable, and solar.
Merging is a platforming trick that makes the pitch consistent. Put rockets, satellites, and an AI company under one roof, then sell into a future market where computation isn’t tied to the terrestrial grid.
The coverage is full of details that make the pitch seem real: the FCC filings, the orbital “data-center system” and the idea of solar-powered compute nodes.
And yes, part of what seems like a fever dream is also in mainstream reporting: an IPO story timed to Musk’s birthday.
He sells orbit as a way to avoid everything that collides with Earth-based calculations:
- grid queues
- local opposition
- water usage
- permission to delay
- rising rates
That’s the sales deck.
Now comes engineering.
IV. Reality check: There’s physics in the classroom too
Musk’s space thesis is not impossible. It is not free.
1) deadline
Some analysts view orbital computation as a long-horizon project, which will likely occur well before the convergence of economics in the 2030s.
2) Latency: training vs inference
Quick Framing:
- Training Has bulk computation and can tolerate higher latency.
- Estimate This is real-time product behavior and is latency-sensitive.
Even in low Earth orbit, round-trip paths can be a real hurdle for the fastest-growing product category: interactive AI.
3) Radiation and reliability
Cosmic radiation destroys electronics. A chip that performs well in the lab is not the same as a fleet that survives years in orbit.
4) Debris and collision risk
Space traffic is not imaginary. The collision cascade is a planetary exoplanet.
5) Cool Motive: Cash
CNBC also framed the merger through a simple lens: XAI’s problem and need for funding while markets are still excited about AI.
There may be class vision. this can also happen strategic insurance And a funding story.
V. Reality check: Earth-based capital spending is a treadmill
Now look at the topological plan, and you’ll see its own mesh.
treadmill effect
AI hardware is depreciating mercilessly. A state-of-the-art fleet increasingly becomes “last generation”. This pushes hyperscalers into constant reinvestment cycles that don’t slow down.
Enterprise AI ROI is still uneven
A cool point that’s easy to overlook in the hype: too many enterprise AI projects still fail to deliver meaningful returns.
debt is rising
This buildout increasingly depends on bullish debt markets as well as operating cash flows.
When spending becomes a treadmill, the main question is not “can they afford it.” It’s “can they continue building if the market declines.”
VI. Scorecard: Who’s Winning Right Now?
Apple is the quiet winner.
It spends much less capital than competitors and still ships AI features by partnering for compute where it makes sense.
Amazon is playing the scale game with the highest sensitivity.
AWS growth is strong, but if demand softens, there’s not much of a buffer when you’re promising ~$200B per year.
Google is hedging better than most.
Large terrestrial construction and experiments that keep options open, including orbital concepts.
Musk is a wild card.
If orbital calculations become feasible earlier than expected, it would compress the shelf-life of Earth-based capital expenditure assumptions. If not, the merger still serves as a funding move wrapped in moonshot.
Oracle looks like collateral damage.
It became vulnerable to changing narratives around mega-deals and infrastructure expectations.
Seventh. what happens next
next 90 days
- nvidia earnings: Do capital expenditure announcements translate into orders on an implicit scale?
- Amazon Q1: If AWS growth slows down, the story becomes increasingly weak.
- spacex ipo filing:S-1XAI will reveal the financial reality and orbital plans in detail.
- Space News: IPO preparations and investor meetings.
next 12 months
- 2026 elections: Data center energy costs could become a local ballot issue in grid-stressed areas.
- orbital prototype:Does any serious verification take place in the real world?
2-5 years
- Does political reaction limit construction work?
- Does the “Capex Treadmill” Economics Break Down?
- Does the classroom move from story to system?
zoom in
I would say what is avoided is the earnings call.
We’re watching one of the biggest infrastructure bets in modern history unfold in real time, and the business model underlying it is still being perfected.
Amazon is spending ~$200B per year to build capability in an enterprise market where most companies still struggle to achieve real returns. Musk is selling an orbital option that turns grid failure into a growth narrative. Google is on the defensive. Apple is charging the reverse without swallowing the entire bill.
Meanwhile, ratepayers feel the strain first, and politics come after the bills.
For years CEOs have repeated the same line: The risk of under-investing is greater than the risk of over-investing.
This has been true from previous booms up until the moment it wasn’t.
The real question is not Earth versus space.
The real question is what happens when $655B per year meets a grid that can’t scale fast enough, an enterprise market that can’t mature on schedule, and a public that starts to notice who pays for the transition.
We’re about to find out.
Follow Zoom into AI to delve more deeply into the economics, infrastructure, and reality behind the AI hype.
The marketing version is free. One has to pay attention to the truth.
Published via Towards AI
