This week: The 10 trillion dollar question, AI infra numbers that will blow your mind by a16z & Brookings, plus upcoming events.
UPCOMING EVENTS
ROOFTOP SOCIAL | Thu Oct 15 | SoHo 6:30-8:30 PM
Join us at our favorite SoHo rooftop this Thursday for drinks, views and good conversation. With a 3x exited founder, friends from Prosper, a16z Speedrun, YC and more.
HARVARD CLUB BREAKFAST | Sun Oct 18 | 9:30AM
Back to where it all began. Join us at the Harvard Club in a unique and exclusive historic setting for a Founders Breakfast on Sunday. Share your wins, struggles, and exchange notes with fellow builders in our community. (10 spots)
FOUNDERS BREAKFAST CLUB x RHO | Th Nov 5 | 8-11AM
Join us at our favorite West Village cafe for coffee & co-working with Rho. Unlimited matcha and pastries included.
Gusto Build | Monday, November 2 | Spring Studios, Tribeca
A special invite from our friends at Gusto: They’re hosting their inaugural summit, an invite-only event for founders and small business owners, with real conversations about building your first team, leading when it's hard, and putting AI to work. Hear from Sloane Stephens, Morgan Housel, Alex Lieberman, and Corporate Natalie, get hands-on in workshops, connect with fellow founders, and request 1:1 time with business experts. Space is limited - register here.
THE 10 TRILLION DOLLAR QUESTION
How much is America spending on the AI buildout? We’re on track for $10 trillion over 2025–2032, according a recent Brookings paper. That’s 3.6% of GDP per year.
For historical perspective on infrastructure spending, that’s >3x the investment intensity of the US national highway system buildout and >1.5x that of the railroad boom.
It’s a scenario, rather than a forecast, but the scale is worth sitting with. Read the paper here.
Where does all that money actually go?
A useful chart in a16z’s latest State of Markets report illustrates the allocation of every $100 of AI infrastructure spending (aka hyperscaler capex): approximately $50 goes to semiconductors, $20 to power, $15 to networking, and the remaining $15 to cooling, facilities and construction. It’s an illustrative breakdown, but it makes an abstract investment boom much easier to picture. See the a16z State of the Markets deck here.
The Brookings paper breaks this down to a single campus: an illustrative 200 MW AI training facility costs around $8.2 billion, including its computing equipment. Roughly two-thirds is IT equipment (chips, anyone?), with the remainder going toward the facility and power infrastructure.
For founders, that raises a more interesting question than the headline number: what becomes difficult when this much money needs to turn into working infrastructure? Equipment needs sourcing. Construction needs coordinating. Power needs securing. Someone needs to manage maintenance, procurement, permitting and the information flowing between all these parties. Those are problems worth understanding before deciding what to build.
Want to get more context on current data center developments?
Check out RegLayer’s public research preview of their interactive data center atlas at atlas.reglayer.ai.
Stepping outside the bubble
Sometimes you might feel like you are lagging behind, and you’re too slow to keep up with all the latest AI news and developments. Don’t despair. The adoption lag is real. Only 2% of US households currently pay for AI. If we extrapolate that to the world at a realistic discount, say, 0.5% of global households, we land at 10 million households. So, from a macro perspective, take a breather: You’re probably ahead of the curve.
Bearish voices from within the industry
Semi maxis understandably didn’t appreciate the latest comments by Broadcom’s CEO that the current AI market and investment cycle feels like a casino.
Want to stay in touch and connect with other community members? Join our WhatsApp Group here.
Want to get involved, partner with us, or get featured in an upcoming edition? Ping us at [email protected].
See you soon,
Stephan, Jacqui & the F&O Team