The AI Boom Is Strangling Tech Stocks (Oct 12, 2026)

High rates and expensive energy might appear to be enough to push technology stocks sharply lower. Yet the equity market has remained resilient. LSEG says strong earnings and software stocks supported equities through renewed energy disruption in the third quarter, even as higher yields hurt bonds.

There is no contradiction. A stock reflects both its future earnings and the rate used to discount those earnings. If AI lifts expected profit quickly enough, equities can rise even while discount rates move higher. The question is which force moves faster.

The first column has been stronger so far. But concentration makes the balance more fragile. S&P Global estimates that Alphabet, Microsoft, Amazon, and Meta together represented roughly 18% of the S&P 500’s market value in early August 2026. The more a small group supports the index, the more the whole market depends on how quickly their investment converts into cash.

A strong index is therefore not the same as a low-risk market. Equities can remain firm while AI earnings outrun the increase in discount rates. If either side of that balance changes, market direction can change quickly as well.

AI looks like software on a screen, but the current investment cycle is physical. Data centres require land, buildings, substations, transmission, cooling, and large amounts of electricity in addition to semiconductors.

The IEA expects U.S. electricity use to increase by more than 420 TWh over the next five years, with data centres accounting for about half of that growth. The EIA expects average U.S. wholesale power prices across the hubs it tracks to reach $52/MWh in 2026, up 11% from 2025. Brent crude averaged $114 per barrel in September, $23 above August.

Those costs return to technology valuations through a simple chain.

1. Data-centre investment raises demand for power, equipment, construction, and labour.
2. Prices and wages rise when supply cannot respond quickly.
3. Persistent inflation leaves the Federal Reserve less room to lower rates.
4. Higher long-term yields reduce the present value of distant cash flows.
5. The market assigns less value to the same dollar of future AI revenue.

The Treasury market is already showing this pressure.

The two-year yield barely changed, while the 10- and 30-year yields rose. Investors are pricing more than the next policy meeting. They are also demanding compensation for long-term inflation, fiscal supply, and capital demand.

The Federal Reserve’s September minutes describe the same two-sided effect. AI investment can lift earnings and productivity, but it can also push total demand ahead of supply and raise energy and infrastructure costs. AI can improve the growth rate while increasing the discount rate applied to that growth.

Large technology companies hold substantial cash, but they are not funding the AI buildout with cash alone. Bonds, leases, joint ventures, and private credit all play a role.

The BIS says hyperscaler gross bond issuance topped $100B in 2025. Companies used long maturities to lock in funding for multi-year data-centre projects. They also placed assets in joint ventures and special-purpose vehicles financed by private-capital firms and insurers.

These structures can make reported capital expenditure and balance-sheet debt look smaller. The economic burden does not disappear if the technology company signs a long lease or guarantees usage, payments, or residual value. The BIS describes such structures as a form of shadow borrowing.

OECD figures show the scale.

Default risk is not the only concern. If lenders demand a higher return, every new data centre has a higher hurdle rate. AI revenue, utilization, or pricing must rise faster for the project to create the same shareholder value. The buildout can continue while the return available to shareholders falls.

The idea that lower AI capital spending automatically means higher FCF and higher tech stocks is only half right. The reason for the slowdown matters.

A good slowdown

Customer demand remains healthy, utilization rises at existing data centres, and less new investment is required for each additional dollar of revenue. AI revenue keeps growing while capex growth moderates. This is the most favourable setup when operating cash flow and FCF improve faster than capital spending.

A bad slowdown

Enterprise customers reduce AI usage or resist pricing, and weaker cloud demand and semiconductor orders force companies to cut investment. Rates may fall, but revenue and profit forecasts fall as well. A lower discount rate may not offset weaker earnings.

The next earnings season should therefore be judged on more than the size of each capex budget. The important question is how much revenue and cash each dollar of installed capital is beginning to produce.

Treating all AI-related stocks as one group hides the important differences. A company funding the buildout, a company selling equipment, and a company adding AI to software receive and spend cash at different times.

For NVIDIA, customer AI capital spending is revenue. For Microsoft, Amazon, Alphabet, and Meta, much of the same spending is an upfront cost intended to create future revenue. Suppliers must prove that orders are durable. Hyperscalers must prove that the capital earns an adequate return.

This is why a slowdown has different effects across the market. Hyperscaler FCF may benefit from lower investment, while an equipment supplier may see weaker orders first. The company’s place in the cash-flow chain matters more than the broad AI label.

The levels below are illustrative monitoring markers, not targets or forecasts.

A one-day move through any threshold is not enough to change the thesis. The useful signal is whether several indicators move in the same direction for at least two quarters.

• The 10- and 30-year Treasury yields and the 2s10s slope
• Brent crude, U.S. wholesale power prices, and data-centre power contracts
• AI revenue, capital spending, and operating cash flow at Microsoft, Amazon, Alphabet, and Meta
• The gap between FCF growth and capital-spending growth
• Hyperscaler bond issuance, credit spreads, and maturity profiles
• Long-term obligations in joint ventures, leases, and purchase guarantees
• Whether semiconductor backlogs and cloud growth weaken at the same time

The positive signal is not simply a lower capex number. It is evidence that installed capital has started to produce more revenue and cash.

Saying that AI is strangling tech stocks does not mean the AI industry is failing. It means demand is so strong that the buildout is consuming enormous amounts of power, energy, and long-term funding, raising its own costs and discount rate.

AI-driven earnings growth has outweighed that burden so far. That is why equities have remained resilient despite high rates and expensive energy. From here, however, the speed of capital recovery matters more than the size of the spending plan.

The best market is not one in which AI investment rises without limit. It is one in which AI revenue and FCF keep growing while the need for new capital and the cost of funding decline. If revenue slows while power costs, interest expense, and lease obligations remain, the AI boom can become the largest source of pressure on technology valuations.

The next question is therefore not whether AI spending continues. It is whether the money already spent produces cash returns above the cost of capital.