a16z released their inaugural “State of Markets” deck this week. Here are the 10 top takeaways for B2B founders and execs:
#1. $5.5 Trillion in Private Unicorns — 10x Growth in 10 Years
Private tech companies valued above $1B now represent ~$5.5T in aggregate value. That’s:
- 15% of the entire Nasdaq market cap
- ~40% of Nasdaq excluding the Mag 7
- 10x growth from where it was a decade ago
There are now 6x more private unicorns (~1,300) than public companies with a $1B+ market cap. The growth story has moved behind the private market curtain.
#2. Fewer Than 5 Public Software Companies Will Grow 30%+ in 2026
Across all of public software, internet, and fintech combined, a16z counts fewer than 5 companies expected to grow over 30% next year.
That’s why the IPO window is effectively closed for most companies. The public markets only want outliers now.
#3. Top Decile Revenue Growth Exploded Upward in 2025
a16z’s data shows revenue growth diverged sharply in 2025:
- Public software median: 12% NTM growth
- Top quartile: 21%+ NTM growth (trading at 20x+ multiples)
- Top decile: 30%+ growth—and almost exclusively private companies now
The companies trading at premium multiples (20x+) are growing at nearly 2x the median rate. And the truly elite growers? They’re not going public anymore. The median company IPO’d in 2024-2025 was 14 years post-founding, versus 5 years 20 years ago.
The power law got more powerful. The gap between winners and everyone else is wider than it’s ever been.
#4. 480 Trillion Tokens/Month — 50x YoY Increase
Google reported monthly token consumption hit 480 trillion tokens. That’s a 50x increase from where consumption was just one year prior.
OpenRouter’s data tells the same story:
- 2024: ~10 trillion tokens/year
- Mid-2025: 100+ trillion tokens/year
- Recent weeks: 1 trillion tokens processed per day
#5. Token Prices Down 3x, Consumption Up 5x
From a16z’s OpenRouter analysis:
- $/million tokens dropped to ~1/3 of February 2025 levels
- Tokens consumed quintupled over the same period
- Every price drop correlated with a consumption spike
Jevons Paradox is real. Cheaper tokens = more usage, not less spend. Almost anyone in B2B planning for AI to get net cheaper in the short run … is planing wrong.
#6. 7-8 Year Old TPUs Still at 100% Utilization
At a16z Runtime, Google shared that TPUs from 2017-2018 are still running at 100% utilization.
That’s not speculative demand. That’s not dark fiber. That’s real, sustained usage on hardware that should be obsolete by now.
#7. $300B+ BigTech AI Capex in 2025
The infrastructure buildout numbers:
- Microsoft: $80B on data centers in 2025
- Google: $75B capex in 2025
- Combined BigTech: $300B+ in AI infrastructure spend
a16z’s math: To pay this back requires ~$1 trillion in AI revenue to materialize. We’re not there yet.
#8. GPU Depreciation Curves Mirror Shale Oil
a16z drew the comparison explicitly: GPU depreciation curves look like shale oil depreciation curves from the mid-2000s.
Shale pattern:
- Year 1-2: Strong cash flows
- Year 3+: Each well generates less
- Eventually: Cumulative returns stagnate, investors lose money
The bull case: AI demand is real and growing. The bear case: We’ve seen this movie before. The honest answer: The payback math is still unproven.
#9. Earnings Beat = Multiple Expansion (Non-Linear)
a16z’s public market data shows the relationship isn’t linear anymore:
- Beat earnings → Get a premium multiple
- Miss earnings → Get hammered
- The biggest multiples go to the biggest earnings surprises
This isn’t a “1x EV to TAM” market. Valuation is increasingly binary: you’re either a winner or you’re not.
#10. Tech Market Cap Has Pulled Away From Everything Else — By $25-30 Trillion
The historical gap between tech and non-tech market performance is now the widest it’s ever been. The numbers:
Market cap growth over 10 years:
- Tech added $25-30T in market cap over the last decade
- Excluding Mag 7, tech still added ~$10T
- Non-tech? Nowhere close
Mag 7 concentration:
- 2015: Mag 7 = 12% of S&P 500
- 2025: Mag 7 = 34-37% of S&P 500 (3x increase)
- Mag 7 market cap doubled from $10T to $20T in just the last 2 years
Top 10 dominance:
- Top 10 stocks now = 40% of S&P 500 market cap (up from 18% in 2015)
- Top 10 generate 32.5% of index earnings (up from 17% in 2015)
- Market concentration is the highest in 50+ years
Individual standouts:
- Nvidia alone added $4.4T to its market cap since 2015
- Nvidia went from 0.8% of Mag 7 value to 21.4% (smallest to largest)
Tech isn’t a subset of great companies anymore. Tech increasingly is the great companies. And within tech, it’s increasingly just the top 10.
The Math
Here’s where we are:
- $4.7T in private unicorn value
- $300B+ in annual AI infrastructure spend
- 480T tokens/month consumed (50x YoY)
- 100% utilization on 7-8 year old hardware
- <5 companies in public software growing 30%+
- ~$1T in AI revenue needed to justify the capex
The demand is real. The utilization is real. Whether the revenue materializes at scale is the trillion-dollar question.
