Main points
- AI CAPEX plateauing: Microsoft, Google, Meta slowing data center GPU purchases after $150B+ build-out
- Growth deceleration: Azure AI revenue growth dropped from +80% to +35% YoY—still good, but slowing
- Adoption gap: 73% of enterprises have AI pilots. Only 18% in production. Monetization isn't following hype.
- Nvidia at risk: If GPU demand plateaus, 45% growth rate crashes → valuation reset imminent
- Palantir's challenge: AI Platform (AIP) hyped, but enterprise contracts not scaling fast enough
- Bull case alive: Infrastructure done = Phase 2 begins (deployment, software, monetization). Best is yet to come.
The Data Doesn't Lie: AI Spending Is Slowing
Let's start with the numbers. Because if you ignore the hype and read the earnings calls, the cracks are showing.
| Company | AI CAPEX 2024 | AI CAPEX 2025 | 2026 Forecast | Trend |
|---|---|---|---|---|
| Microsoft | $48B | $56B | $52B (guidance) | 📊 Plateauing |
| Google (Alphabet) | $42B | $48B | $46B (est.) | 📊 Flat |
| Amazon (AWS) | $51B | $58B | $60B | ✅ Still growing |
| Meta | $28B | $38B | $35B | 📉 Cutting back |
| Tesla | $4B (Dojo) | $5B | $4B | 📉 Reduced |
Translation: The AI infrastructure arms race peaked in 2025. The data center build-out is largely complete. Now what?
📉 Microsoft Azure AI: The Deceleration
Q1 2025: Azure AI
revenue +80% YoY ($12B run rate)
Q2 2025: +62% YoY
Q3
2025: +48% YoY
Q4 2025: +35% YoY ($22B run rate)
Still impressive growth. But the RATE is decelerating fast.
"We've built the infrastructure. Now we're in the 'prove the ROI' phase. CFOs aren't writing blank checks anymore."
Contrarian Take
Most analysts focus on Nvidia's GPU dominance, but they're missing the real story: their software moat through CUDA. Competitors can match chip performance, but can't replicate a decade of developer ecosystem investment.
The Enterprise AI Adoption Gap
Here's the uncomfortable truth: AI pilots are everywhere. AI in production? Rare.
Enterprise AI Adoption 2026
| Stage | % of Enterprises | Spending Level |
|---|---|---|
| Exploring AI | 92% | $50k-200k (pilots) |
| Running Pilots | 73% | $500k-2M |
| Production Deployment | 18% | $5M-50M+ |
| Full AI Transformation | 3% | $100M+ |
The Problem: Pilots don't generate revenue. Production does. And the leap from pilot to production is HARD.
Why enterprises are stalling:
- Data quality issues: AI models need clean, structured data. Most enterprises have messy, siloed legacy systems.
- Integration hell: Plugging AI into 20-year-old SAP/Oracle systems is nightmare fuel.
- Unclear ROI: "Our chatbot handles 40% of support tickets" sounds great until you realize support tickets aren't your bottleneck.
- Change management: Employees resist AI tools. Adoption <50% even after deployment.
- Regulation/liability: Who's liable if AI makes a bad decision? Legal departments blocking rollouts.
⚠️ The Monetization Mirage
Tech companies spent $200B+ building AI infrastructure. They're generating ~$50B in AI revenue.
That's a 4:1 spend-to-revenue ratio. Unsustainable.
Eventually, CFOs will say "show me profits or we're cutting budgets." That moment is arriving in 2026.
Nvidia: The House That AI Built... Is It Shaking?
Nvidia is the AI trade. $2.9T market cap. 45% revenue growth. 40x P/E. The most crowded position in hedge fund history.
But here's the thing: Nvidia's growth is 100% dependent on continued AI infrastructure spending. If that spending plateaus, Nvidia's growth crashes.
📊 Nvidia Revenue Growth Trajectory
2023: +126% YoY (AI boom begins)
2024: +94% YoY (H100 dominance)
2025: +45% YoY (slowing
but strong)
2026 Forecast: +25-30% YoY (deceleration continues)
Still growing fast. But at 40x P/E, any slowdown = valuation
compression.
Why Nvidia's growth is at risk:
1. Hyperscalers Have Enough GPUs (For Now)
Microsoft, Google, AWS bought enough H100s/H200s to last 18-24 months. The urgent "we need GPUs NOW" phase is over. Orders shifting from aggressive expansion to replacement cycles.
2. Competition Emerging
AMD's MI300X gaining traction (+30% cheaper, 80% of performance). Google's TPUs (in-house). Amazon's Trainium chips. Nvidia's moat narrowing.
3. AI Model Efficiency Improving
GPT-4 was 1.7 trillion parameters. GPT-5 rumors: 5-10 trillion BUT more efficient training. You don't need 2x GPUs for 2x parameters anymore. Efficiency gains = less chip demand.
"Nvidia's dominance is unquestionable today. But at 40x P/E, you're paying for perfection. Any hiccup—competition, demand slowdown, macro weakness—and you're looking at a 30% correction."
Bull Case for Nvidia:
Phase 1 (infrastructure) done. Phase 2 (inference chips for production AI) just beginning. Inference market could be 3x larger than training. Nvidia's Blackwell chips launching 2026. Growth reaccelerates.
Bear Case:
Demand peaked. Competition rising. Efficiency killing volume. Growth decelerates to 15-20%. At 40x P/E, fair value → $80/share (45% downside).
Palantir: The AI Platform Promised Land... Or Mirage?
Palantir's AI Platform (AIP) was the hype of 2024-2025. "Bring AI to enterprise data." "Boot camps converting pilots to contracts." Stock went from $15 to $92.
But let's examine the numbers:
📊 Palantir Revenue Breakdown 2025
Total Revenue: $2.9B (+27% YoY)
Government: $1.6B (+18%) — Steady but slow
Commercial:
$1.3B (+39%) — Driven by AIP hype
AIP-Specific Revenue: ~$400M
(estimated, not broken out)
Market Cap: $195B
Revenue
Multiple: 67x P/S
P/E Ratio: 105x
Translation: Palantir is priced for 50%+ growth for 5+ years. If commercial growth slows from 39% to 25%, valuation collapses.
The AIP Challenge:
- Boot camps generate pilots, not contracts. Conversion rate ~20-30%.
- Average AIP contract: $3-5M/year. Takes 100 contracts to add $500M revenue.
- Sales cycle: 9-18 months from pilot to production.
- Competition: Databricks, Snowflake, Microsoft Fabric all targeting same customers.
⚠️ Valuation Risk
Palantir at 105x P/E requires flawless execution. One earnings miss, one guidance cut, one major contract loss—stock could drop 40% in a day.
This isn't a value play. It's a "belief in Karp's vision" play. High risk, high reward.
The Bull Rebuttal: Why AI Spending Slowdown Doesn't Matter
Let's be fair. The bull case for AI stocks is still strong. Here's why:
The Infrastructure-to-Deployment Transition
Bull Thesis:
CAPEX slowdown is GOOD. It means Phase 1 (building infrastructure) is complete. Phase 2 (deploying AI, generating revenue) is beginning.
Historical Parallel: Cloud Computing
2008-2012: AWS builds data centers.
Massive CAPEX. Revenue growth slow.
2013-2017: CAPEX plateaus. But revenue
EXPLODES as enterprises migrate to cloud.
Result: Amazon stock +2000% from
2013-2020.
AI Could Follow Same Path:
Infrastructure done → Deployment begins → Revenue scales
→ Software multiples expand
Microsoft, Palantir, Snowflake, Databricks = biggest winners
of Phase 2.
Bull Case for Nvidia:
Training GPUs (H100) plateauing. But inference GPUs (Blackwell, Hopper) exploding. Every production AI app needs inference chips. Addressable market 10x larger than training.
Bull Case for Palantir:
Enterprise AI adoption at 18% today. When it hits 50% (5 years?), Palantir revenue could be $15B from $3B today. Stock justified at current price if that happens.
"AI isn't over-hyped. It's under-hyped. We're inning 2 of a 9-inning game. The infrastructure slowdown is just a breather before the real revenue tsunami hits."
What Should Investors Do?
If You Own Nvidia:
Consider trimming 30-50% to lock in gains. Keep core position if you believe inference demand will offset training slowdown. Set stop-loss at $100 (25% downside protection).
If You Own Palantir:
Ultra-high risk. If you're up big, take profits. If you believe in Karp's 10-year vision, hold but size appropriately (5-10% of portfolio max). This is not a widow-and-orphan stock.
If You Own Microsoft/Google (AI Exposure):
These are safer. AI is 20-30% of their business, not 100%. Even if AI growth slows, core businesses (cloud, ads) remain strong. Hold long-term.
If You're Buying:
Wait for pullback. If Nvidia hits $100, Palantir $60, that's a better entry. Or dollar-cost average on 10-15% dips.
🚨 What to Watch in 2026
- 1. Nvidia Q1-Q2 guidance — If growth decelerates faster than expected, correction deepens
- 2. Enterprise AI adoption rate — Does it accelerate from 18% to 30%+? Or stall?
- 3. Microsoft/Google CAPEX cuts — Further reductions = bearish for Nvidia
- 4. AI ROI case studies — If major companies show 10x ROI, spending reaccelerates
- 5. Competition — AMD, Google TPU, custom chips gaining share?
Bottom Line: The AI spending slowdown is real. But it's not necessarily fatal. It's a transition. From infrastructure build to production deployment. From CAPEX to revenue. From hype to profitability.
The question is: Will the revenue scale fast enough to justify the valuations? Or are we in for a painful reset?
Time will tell. Manage your risk. Don't bet the farm. And remember: In tech, the companies that survive hype cycles become the next generation of giants.