Big Tech’s AI buildout still looks huge, but the economics just got uglier. Reuters reported that Microsoft, Amazon, Alphabet, and Meta are expected to spend about $635 billion on AI infrastructure in 2026, up from $383 billion in 2025 and just $80 billion in 2019. That spending was already aggressive. Now it is colliding with a war-driven energy shock that makes data centers more expensive to build and run.
The problem is simple: AI data centers do not just consume chips. They consume electricity at scale, and a jump in oil and power costs changes the math fast. Reuters, citing S&P Global, said persistently high oil prices could force capital-spending cuts and even trigger a wider equity correction if companies and investors start doubting whether the returns justify the cost. That is why this is more serious than a normal “AI is expensive” complaint.
What changed in 2026
The AI spending plan itself has not collapsed. Meta just boosted its West Texas AI data center investment to $10 billion, and Reuters said the four hyperscalers are still on track for more than $630 billion in combined AI-related capex this year. The real change is that the financing and operating backdrop got worse at the same time.
Reuters also reported that higher bond yields are hitting Big Tech at the worst possible moment. Over the next two years, analyst estimates cited by Reuters suggest hyperscaler capex could reach around $1.4 trillion, consuming nearly 90% of operating cash flow. On top of that, these companies are expected to issue about $175 billion in debt in 2026, up from $121 billion in 2025. That means the AI boom is becoming more dependent on external financing just as financing gets pricier.
Why data-center economics look more fragile now
This is where the hype crowd keeps fooling itself. They act like AI capex is just a one-way growth story. It is not. Data centers are power-hungry assets, and power costs are now a bigger variable than they looked a few months ago. Reuters’ S&P Global report said the market is starting to test whether rising energy prices will squeeze returns enough to make companies slow or rethink parts of their rollout.
The pressure is not only theoretical. Reuters reported that Meta’s Texas AI facility is planned at 1 gigawatt of capacity. That tells you the scale of electricity demand involved. When projects get this large, power availability, grid reliability, cooling costs, and financing costs all become central to profitability. That is exactly why the energy shock matters.
| Indicator | Verified number | Why it matters |
|---|---|---|
| Big Tech AI capex in 2026 | $635 billion | Shows how large the spending wave has become. |
| Big Tech AI capex in 2025 | $383 billion | Confirms 2026 is a major acceleration. |
| Big Tech AI capex in 2019 | $80 billion | Shows how extreme the expansion has been. |
| Expected hyperscaler debt issuance in 2026 | $175 billion | Higher rates now hurt more because funding needs are rising. |
| Meta Texas AI data center scale | 1 gigawatt | Power demand is massive, not marginal. |
Why investors are getting less comfortable
Investors were fine with giant AI spending when money was easier and energy costs were calmer. That changed in March. Reuters said nearly 70% of hyperscalers’ operating cash flow is already going into capex, and the Roundhill Magnificent Seven ETF has fallen about 20% from its October high. That does not prove the AI story is broken, but it does show investors are beginning to punish companies when cost and financing risks rise faster than expected.
There is also a broader risk here. If AI infrastructure returns get questioned, it does not just hurt one stock. Reuters’ S&P Global report said a persistent energy shock could trigger a much larger correction because so much of the recent market optimism has been tied to AI spending continuing almost without restraint. That is the uncomfortable truth most people prefer to ignore.
What this means now
A few things matter more than the hype:
- AI spending is still enormous, but it is no longer operating in a friendly cost environment.
- Data-center power demand makes AI infrastructure unusually sensitive to energy shocks.
- Rising yields matter because Big Tech is leaning more on debt as capex expands.
- The market is no longer assuming every dollar of AI capex automatically creates value. That last point is an inference from Reuters’ reporting on valuations, debt, and investor skepticism.
Conclusion
Big Tech’s AI spending binge just ran into a very expensive problem because the buildout now depends on massive power use and increasingly costly financing at the same time. The spending plans are still there, but the margin for error is getting thinner. The real issue is not whether AI investment continues. It is whether investors keep rewarding it once the power bill and debt bill both get uglier.
FAQs
How much are major tech companies expected to spend on AI infrastructure in 2026?
Reuters reported that Microsoft, Amazon, Alphabet, and Meta are expected to spend about $635 billion on AI infrastructure in 2026.
Why are energy costs such a big issue for AI?
Because AI data centers consume huge amounts of electricity, so higher energy prices directly raise both operating costs and return-on-investment pressure.
Are these companies relying more on debt now?
Yes. Reuters reported that hyperscalers are expected to issue about $175 billion in debt in 2026, up from $121 billion in 2025.
Is the AI spending boom slowing down already?
Not clearly yet. Spending plans remain large, but Reuters and S&P Global said the energy shock is now testing how sustainable that pace really is.