Yes, we're going to talk about big blobs of compute.
Get it out of the way up front. Nobody is surprised; the point is that it frames every question about our portfolio and our model.
“I'd find something else to do.”
Michael Moritz, former chairman of Sequoia, asked if he'd be a VC today. October 2026.
Let the quote land. The answer to the question he was asked is the rest of this talk: who we are, what's different, and why speedstrapping is the version of venture that still works.
Agenda
Who we are, and why we're different
13 questions on the AI build-out
Changes to how we invest
Speedstrap, the book!
Third Sphere
We forge new paths for those reimagining global systems.
Twelve years. 100+ seed investments in climate. Speedstrapping as plan A since Fund IV.
Thesis evolution · 2014 to today
Hard-won lessons in building for the physical world.
Grey Rhino
EVOLOHNear Space Labs
Grey RhinoRenewell
Why hardware needs different underwriting
Not all hardware is the same. Complexity and customization stack risk.
Risk we underwrite wellNew frontier, AI-unlockedRisks override: avoid, or special structure
More complexity
Standardized complex
newly viable
Platform-based complex
newly viable
Complex product systems
traditional nuclear · none in the portfolio
🐉
Mass produced, complex
Nevoya · EVOLOH · Furno
Platform-based systems
Near Space Labs · Fire Dome · Waste*d
Complex-custom
newly viable
Simple, mass-produced
Future Motion · Plentify · Thalo
Simple, mass-customized
Revivn · Wonderfil · Renewell
Small-batch
newly viable
More customization
The team
Twelve years in. The most common thing we hear is that we're different.
Alphabet/X · Amazon · World Bank · JP Morgan · Lyft
LSE · Harvard · Stanford · Berkeley · UChicago
Top 2% of pre-seed investors on NFX Signal in nine categories: Hardware #6 of 529 · Climate #5 of 1,662 · Deep Tech #3 of 828 · Energy #3 of 673 · Manufacturing #7 of 458 · Impact #3 of 520 · Transportation #4 of 657 · Real Estate #4 of 968 · DTC #4 of 288. Oct 1, 2026.
Off the clock
We build, race, fight, and climb on the weekends.
Impact · 2025 portfolio report
Measured on the top 24 companies.
45.7M tonnes
CO2e avoided, lifetime · about 10 million cars off the road for a year
300M+ people
protected from climate risk · de-duplicated across companies
How impact is measured
mitigation, CO2e avoidedadaptation, people protectedboth
Bits and atoms
hardware and softwaresoftware or services
Credit that isn't venture debt
Most use revenue-based finance, equipment, project or asset-backed lending. None use venture debt.
Source: 2025 Third Sphere Portfolio Impact Report, Q1 2026. Activity data × emission factor; protected population from deployments × coverage.
Stanford case study
thirdsphere.com/stanford-case
The AI build-out
We know it's big, but what does it mean?
The AI build-out
Head-on AI looks like one thing. The structure only shows when you walk around it.
Section 2 · the deep dive
Thirteen questions about the AI build-out
Where value goes · Q1
If switching is free, where's the moat?
Where value goes · Q1
If switching is free, where's the moat?
Where value goes · Q1 · the answer
Nevoya: use automation to navigate supply chain and operational complexity to create a stand-out brand.
Where value goes · Q2
If attention is scarce, what does a brand look like?
Where value goes · Q2 · the answer
SwellCycle: a brand is an authentic person plus proof.
From the Speedstrap case study (Ch 12).
Where value goes · Q3
Falling costs always drove adoption. Never this fast. What should we expect in five to ten years?
Where value goes · Q3 · the answer
ReynKo: replace the wind tunnel with near-free generative AI tests.
Where value goes · Q4
How fast does it diffuse?
Where value goes · Q4 · the answer
Cove: the client never adopts AI. They get the benefits anyway.
Where value goes · Q5
Where does the value land?
Where value goes · Q5 · the answer
The stealth company: the AI stays in the lab. The deliverable is a product or a license.
Section 2 · block 2
What the physical world constrains
Physical constraints · Q6
Compute time to power
Physical constraints · Q6 · the answer
EVOLOH: few manufactured components, dropped anywhere with minimal construction.
Physical constraints · Q7
How do we reduce the cost of physical climate risk?
Physical constraints · Q7 · the answer
Fire Dome: take the physical risk out at the asset.
Physical constraints · Q8
Same El Niño, 29 years apart. The baseline moved. How do we act on warming?
Sea surface temperature anomaly vs the 1991–2020 average. NOAA PSL OISST v2. In 1997 the warmth was the El Niño; in 2026 the El Niño sits on top of a warm ocean.
Physical constraints · Q8 · the answer
Thalo: the buyer pays for fewer truck rolls and a lower power bill. The carbon comes free.
Physical constraints · Q9
What are we decoupling from, and why?
Physical constraints · Q9 · the answer
Future Motion: the winning side of reshoring.
Section 2 · block 3
Who pays
Who pays · Q10
Who pays for decarbonization now, and which policy support lasts?
Who pays · Q10 · the answer
Sonic Fire: the buyer pays because the alternative costs more.
Who pays · Q11
Who is funding the build-out?
Who pays · Q11 · the answer
Waste*d: the lender underwrote the contracts, not the story.
Who pays · Q12
What do exits look like when there are 8 megacorps?
Who pays · Q12 · the answer
Revivn: raise less, and the whole market can afford you.
Under all of them · Q13
What is the right way to think about VC fundraising?
How it changes how we invest
These questions are now our investment screen.
What changed
We took a year to rebuild how we underwrite. Every new company now has to pass the screen, one test per question.
The screen
Owns one of the five moats Q1
Earns attention in public, no paid acquisition Q2
Still works when intelligence is nearly free Q3
An ordinary team can run it Q4
Exit doesn't need AI-infra multiples Q5
Shortens time to power Q6
Works when the grid doesn't Q7
Sells without the word carbon Q8
Margin absorbs a tariff Q9
Someone pays with their own money today Q10
A lender would underwrite the revenue Q11
A premium buyer exists, or the cap table works at a cash-flow price Q12
Section 3
Why write a book now
Why write a book · 1
Too much focus on growth, not durability.
Why write a book · 2
Impact doesn't work if companies become undead, or fail. We count about seven of our own, and most were avoidable.
Why write a book · 3
There doesn't seem to be much VC left for everyone else.
Why write a book · 4
This isn't our idea. It's how companies were built before software was eating the world.
Why write a book · 5
AI changes what small teams can set out to do, and how much they need to do it.
Third Sphere · 2026
Speedstrap
A founder's playbook where VCs aren't the main characters.
Where this started · the story you know
The story you know: two funds, about $20M, an IPO at $600M.
Where this started · the story you know
"If not for what Sega did for Nvidia, Nvidia would not be here today."
Plan A isn't Series A
It's not a typo. Seedstrap and speedstrap both say Plan A isn't Series A. They diverge on what they will spend.
Five edges
Five edges. A ring, not a hierarchy.
Four parts, fifteen chapters
Four parts, fifteen chapters.
Six companies, close up
We learned from our founders. Now you can too.
For GPs
Efficiency buys the other half of the cycle.
For LPs
We don't know what share of companies will be built this way. It is already part of the mix, and it is working.
The book
A limited edition copy today. Everyone else on launch day.
Questions
Speedstrapping · AGM, October 8, 2026 · v1.20 · Made by Third Sphere & Claude Fable 5.1
EVOLOH
Near Space Labs
Grey Rhino
Renewell
SwellCycle: a brand is an authentic person plus proof.
EVOLOH: few manufactured components, dropped anywhere with minimal construction.
Thalo: the buyer pays for fewer truck rolls and a lower power bill. The carbon comes free.
Waste*d: the lender underwrote the contracts, not the story.
Revivn: raise less, and the whole market can afford you.
