AI infrastructure risks mount as NVDA, MSFT and GOOG diverge

Investors aren’t turning cautious on AI because of headlines. They’re reacting to something deeper: the capital and energy demands behind the AI boom now look difficult to sustain. The concern isn’t about innovation. It’s about the infrastructure needed to power it.

AI labs are committing trillions to data centres, chip supply, and power capacity. Utilities are pushing back. Chip depreciation cycles are tightening. Financing structures behind the boom now look stretched.

This has created the first real wobble in the AI rally in months. Even so, the technical picture isn’t uniformly bearish. Some leaders are still healthy. Others are flashing early warnings.

NVIDIA: Oversold enough for a bounce, but not the strongest name anymore

NVDA is at a tactically important level. Price sits beneath the EMA-20 Bollinger band with an oversold Stoch RSI signal, the same combination that triggered every major NVDA rebound during the 2025 AI rally. Historically, NVDA reclaimed the EMA-20 quickly and rode the upper half of the band higher.

Today, that reclaim is missing.

This is the difference between a normal pullback and the start of a deeper unwind. January showed the pattern clearly:

• break below the band.
• fail to reclaim.
• grind lower for weeks.

We are back at the same setup. A bounce is likely, but NVDA only recovers its trend if it breaks and holds above the top of the orange band. If it fails at the band again, the structure weakens. NVDA is no longer the cleanest of the leaders.

Microsoft: Oversold, near VAL, and the weakest of the big three

Microsoft mirrors NVDA’s weakness, but with even less strength behind it. MSFT is sitting near the Rally VAL and testing the lower boundary of the entire AI rally structure.

Price is still suppressed by the EMA-20. Historically, when MSFT breaks below the band, it tends to revert back into the band before choosing direction. That is the likely path now: a relief move into the orange bands, then a decision.

MSFT is currently the weakest of the big three:

• flatter trend.
• more consistent rejections at the EMA-20.
• heavier structure.

A short-term bounce is still likely, but it lacks the momentum that GOOG currently has.

Google: The best-performing of the three, but stretched and ready to cool off

GOOG is the strongest AI leader for now. Fundamentally, Google is funding its AI expansion through real cashflow: TPU v6 rollout, Gemini Ultra integrations, and significant internal capex. Google is not relying on circular financing or trillion-dollar promises.

Technically, GOOG remains in a strong trend:

• price is far above the EMA-20 band.
• Stoch RSI is nearly overbought.
• two unfilled gaps below act as natural magnets.

Every time GOOG reaches this level of stretch, it returns into the EMA-20 band. This is normal for strong trends. It is not bearish by itself. As long as GOOG holds the band on any retracement, the broader AI trade, and the SPX by extension, remains in good shape.

Why investors are nervous: The AI boom is colliding with physical limits

This is the real issue. It isn’t the charts. It’s the demands behind the technology.

1. Power limits

Amazon has already clashed with utilities over power delivery. Nvidia warned in filings that future demand is now limited by access to electricity and capital.

2. Trillion-Dollar infrastructure

OpenAI’s CFO stated that a single gigawatt-scale data centre costs around $50B:

• $15B for land and power shell.
• $35B for GPUs.

OpenAI’s long-term plan involves up to 23 such centres. The math quickly becomes extreme.

3. Financing challenges

GPUs depreciate in 12–18 months. Lenders do not want them as collateral.
This is why OpenAI floated the idea of a government “backstop”.

4. Circular funding loops

• hyperscalers fund AI labs.
• labs buy chips from NVDA and AMD.
• chipmakers promise reciprocal investments.
• hyperscalers rely on deliveries to build the next models.

It is a loop. Investors understand that loops break under stress.

5. OpenAI’s financial strain

• $11.5B lost in one quarter.
• Sora2 likely burning $4–5B per year.
• $1.4T in commitments.
• even a $1T IPO covers only a very small fraction of the cost.

6. Grid capacity

The AI buildout requires power equivalent to multiple nuclear reactors. The USA built only one new nuclear plant in thirty years.

This mix of capital pressure, energy limits, and financing complexity is what is making investors uneasy.

Bottom Line

We are in a moment where AI hype is still strong, technical setups are mixed, and the bottlenecks behind the scenes are becoming hard to ignore.

• NVDA: bounce likely, trend unconfirmed.
• MSFT: weakest structure, oversold.
• GOOG: strongest trend, stretched.

If NVDA and MSFT reclaim their EMA-20 bands, the AI rally stabilises. If GOOG holds the band on retracement, SPX remains supported. If the leaders fail at their bands, the market may be pricing in something bigger: the AI expansion may not scale as fast as promised.