Thesis Pieces
State of AI: Investing
Following the money from software to the real world, and what it means for you.
Who here…
- …invests? Your returns
- …builds a product or service? Your funding and customers
- …is building a career? Your next role
That’s everyone. Where the money moves decides all three.
AI starts physical, and ends physical.
- Apps Next · Bits drive atoms Robotaxis, robots, AI-designed drugs
- Models Now · Bits Frontier labs and software
- Infrastructure Now · Atoms Data centers: the AI factories
- Chips First · Atoms GPUs: whoever had them won
- Energy Always · Atoms Power: the constraint under it all
Framework: Jensen Huang, NVIDIA blog, “AI is a five-layer cake” (Mar 2026)
The biggest buildout in history.
| FY23 | 15 |
|---|---|
| FY24 | 48 |
| FY25 | 115 |
| FY26 | 194 |
| 2022 | 162 |
|---|---|
| 2025 | 448 |
| 2026 plan | ~835 (estimate) |
| 2023 | 18% |
|---|---|
| 2024 | 34% |
| 2025 | ~50% (estimate) |
| Q2 '26 | 70%+ |
Most of this money buys atoms: chips, buildings, power.
Sources: NVIDIA annual results (fiscal year ends late January); Epoch AI via Visual Capitalist; TMT Finance (2026 plan, incl. Oracle); Crunchbase. Hatched bars are plans or estimates.
Software’s shelf life is shrinking.
- $285B wiped off software stocks in 48 hours after AI agents launched (Feb 2026)
- 30% faster churn: AI apps lose paying subscribers quicker than other apps
- 2 months for ChatGPT to reach 100M users. Uber took 70 months.
Illustrative
Sources: SaaS Sentinel / Forbes (Feb 2026 selloff); RevenueCat State of Subscription Apps 2026; company data via Business Today.
The money is moving into the real world.
3× a normal year, in 6 months.
| 2022 to 2024, avg per year | $14B |
|---|---|
| 2025 | $38B |
| H1 2026 only | $47B |
Biggest bets on atoms, 2026
- SpaceX + xAI IPO at $1.77T, raised $75B
- Waymo $16B raise at a $126B valuation
- Anduril $5B raise at a $61B valuation
- Tesla Cybercab in production; Optimus line next
Source: Crunchbase; physical AI = robotics, autonomous vehicles, aerospace, drones, industrial automation, sensors. Tesla Q2 2026 call.
Where bits meet atoms.
-
Was · 2010s to 2024
Bits
Software eats the world: SaaS, apps, per-seat pricing.
VCs want Growth and seats
-
Is · 2025 to 2026
Bits buy atoms
AI money pours into chips, data centers and power.
VCs want Revenue and retention, not demos
-
Will be · 2027 and beyond
Bits drive atoms
AI goes IRL: robotaxis, robots, devices, AI-designed drugs.
VCs will want A physical moat: hardware, real-world data, regulation
Software alone gets copied in months. Software plus the real world is hard to copy.
Get on the rising curve.
Investors
Rising Power, infrastructure, physical AI
Fading Per-seat software
Ask Which layer is this on? Is money still arriving?
Builders
Rising Agents that do the work; AI in the real world
Fading Features the next model gives away
Ask Is money flowing here, from funders and customers?
Careers
Rising AI infrastructure, energy, robotics, AI in every industry
Fading Tasks AI now does for free
Ask Will this market be bigger in 3 years?
One question for all three: is the money flowing in, or out?
For discussion only. Not investment or career advice.
Which curve are you on?
Your money. Your product. Your career. Follow the money.
Bits driving atoms, all the way to orbit.
This talk ends where bits start driving atoms. Our thesis takes it one step further, to space, which is where we focus our advising and investing.