I spent two years on the evaluation side, screening more than 150 early-stage startups and writing the memos, and the rest of my time building companies from zero. This page is where I think the next decade of value gets made: six big bets, the maps behind them, and the quiet markets I would back at pre-seed.
Six shifts I would build a fund around for the next ten years. Each memo says what the consensus misses, where the value pools, and where a seed check can still win.
Software sold seats to people. The next decade sells the finished work, and that budget is labor: more than ten times what the world spends on software.
Conviction HighScales by 2030 Read the memo 02$2.6T drug market by 2030AI made the first step of drug discovery cheap. The money goes to whoever owns the data, the lab loop and the clinical asset, not the model.
Conviction HighScales by 2031 Read the memo 03945 TWh of data center demand by 2030AI stopped being limited by chips and started being limited by power. This decade, speed to electricity is the product, and most of it will not come from new reactors.
Conviction HighScales by 2029 Read the memo 04$5T humanoid market by 2050 (Morgan Stanley)Robot bodies are getting cheap fast, mostly in China. The value pools in the models, the training data and the narrow jobs where a robot can work today.
Conviction MediumScales by 2034 Read the memo 05$1.5T US defense request for 2027Everyone is funding drones and autonomy. The constraint is production: factories, shipyards, rocket motors and cheap interceptors. That is where I would put the next check.
Conviction MediumScales by 2030 Read the memo 06$2T in stablecoins by 2028 (Standard Chartered)Stablecoins are not a crypto story anymore. They are dollars that settle in seconds, and the incumbents are already buying the rails. The opening is at the edges.
Conviction HighScales by 2030 Read the memoOne map for each big bet, the capital stack at pre-seed and seed, the quiet markets and the wider AI stack. Placements are checked against what each company ships.
As of Oct 2026
As of Oct 2026
As of Oct 2026
As of Oct 2026
As of Oct 2026
As of Oct 2026
As of Oct 2026
Sources: Kruze Consulting, Carta pre-seed data (Q2 2026) Y Combinator, The YC deal Techstars, accelerator terms South Park Commons, Founder Fellowship (Jul 2026) a16z, speedrun applications (Apr 2026) Afore Capital First Round Capital GamesBeat, PitchBook-NVCA midyear 2026
As of Oct 2026
As of Oct 2026
As of Oct 2026
As of Oct 2026
As of Oct 2026
As of Oct 2026
Sources: NVIDIA Q2 fiscal 2027 results Cerebras CoreWeave Nebius Crusoe Together AI Fireworks AI Baseten
As of Oct 2026
Sources: NVIDIA Jetson AGX Thor launch Hesai Ouster Orbbec Harmonic Drive Nabtesco maxon Apptronik
As of Oct 2026
Sources: Baltimore declaration on organoid intelligence, Frontiers in Science NSF invests $14M in organoid intelligence projects Brainoware, Indiana University (TechXplore) Braingeneers, UC Santa Cruz Cortical Labs commercial milestones (IT Brief) Cortical Labs CL1 launch (ChannelLife) FinalSpark Neuroplatform Biological Black Box (Entrepreneur)
The best pre-seed markets are quiet. Buyers are already paying to solve the problem, badly, and nobody with a term sheet is calling them.
Most early-stage sourcing starts from the technology and works out to a customer. I start from the customer and the money they already spend, and only then ask what the technology now makes possible. It is slower. It also finds markets before the market maps do.
What it looks likeBuyers already spend real money or staff time on a clumsy fix: consultants, offshore assistants, answering services, factoring, spreadsheets someone maintains full time.
How I measure itThe price of the workaround and how many buyers use it. A $40 an hour consultant is a stronger signal than a survey saying the problem is painful.
Where I lookService providers' pricing pages, job boards, freelance marketplaces, trade association benchmark reports.
What it looks likeThe work runs on phones, paper and people paid by the hour, while the rest of the industry's spend has moved to software.
How I measure itClerical and coordinator headcount in the vertical, and the share of a role spent on admin rather than the craft.
Where I lookBLS occupation data, Census business counts, job postings for coordinators, billers and schedulers in the vertical.
What it looks likeA capability got cheap enough to undercut what a small buyer pays for the human version today.
How I measure itCost per task with current models against the cost of the person or service it replaces.
Where I lookModel and API price sheets, latency benchmarks, interpreter and answering service rate cards.
What it looks likeA regulation, mandate, deadline or new API that forces buyers to change how they work, or opens data that used to be locked.
How I measure itThe date it binds, who it binds, and what it costs to comply by hand.
Where I lookFederal Register, state legislatures, agency rulemaking dockets, platform and carrier API changelogs.
What it looks likeThe buyer base is large and the number of venture-backed companies selling to it is small.
How I measure itMy crowding ratio: funded startups selling to the segment per 10,000 buyers. Low is good.
Where I lookFunding databases, public market maps, accelerator batch lists, conference sponsor lists in the vertical.
What it looks likeSomeone who has done the job, sold to the buyer, or grown up inside the community, and can get ten buyers on the phone this month.
How I measure itNames, not personas. Can the founder list ten buyers who will take the call?
Where I lookThe founder's own history, operator networks, trade associations, community groups.
Each signal scores 0 to 3. Multiply by its weight. The top score is 45.
I begin with occupation and business-count data and look for large groups of small buyers doing repetitive admin by hand.
For each candidate I look for the paid workaround and price it. No workaround, no memo.
I map every funded company selling to the segment and compute the crowding ratio before I form a view.
I cost the task with today's models and compare it to the workaround. The gap is the wedge.
Every memo states the thesis in one sentence, the evidence, and the signals that would make me drop it.
The founder-access score stays at zero until real buyers have described the problem in their own words.
Proof of paid demand against how many funded founders already sell to the buyer. My placement, from the evidence in each memo.
Each one ran through the screen. Each names the wedge I would back, who is already building, and what would change my mind.
Caregiver turnover falls from 84% to 53% when pay rises from about $16 an hour to about $19. A small agency can only fund that raise from its back office.
34/45 on the screenFull memo Conviction: High Read the memo 02Oct 2026I build hereA $45 a month policy pays the agent a few dollars a month. So the smallest businesses are left with one carrier's website or no cover at all, and 73% are under-protected.
32/45 on the screenFull memo Conviction: Medium Read the memo 03Oct 2026Fewer small firms win federal work every year. On November 10 the Pentagon's cyber rule starts requiring audits that cost a small supplier about $100,000. The software money went to proposals.
32/45 on the screenFull memo Conviction: Exploring Read the memo 04Oct 2026I build hereOwners pay general contractors in about 25 days. General contractors pay their subs in about 57. The money gets stuck one step down, at the firms least able to wait.
31/45 on the screenFull memo Conviction: Medium Read the memo 05Oct 2026I build hereMachine translation is about to be free. Getting a legal or medical intake right in Spanish, Vietnamese or Tagalog is not, and most voice AI is still built English-first.
28/45 on the screenFull memo Conviction: Medium Read the memoOne company from each wave, from data to biocomputing: the wedge, why now, the moat, the numbers, and what would change my mind. The full memos live on the Writing page.
Every company sits on data it cannot query fast enough, and AI made that gap expensive.
Public (NYSE: SNOW) CryptoCrypto trading is cyclical. Coinbase is trying to become the plumbing that earns through it.
Public (Nasdaq: COIN) AILaw firms bill by the hour. Harvey sells the software that shrinks the hour.
Private, late-stage venture ($15.5B valuation, Sept 2026) AIDoctors spend evenings typing notes. Abridge listens to the visit and writes them.
Private, late-stage venture ($5.3B valuation, Series E) Embodied AIRobots still need new programming for every task. Physical Intelligence wants one general brain.
Private, venture-backed (last reported price $5.6B, Nov 2025) BiocomputingAI burns megawatts. Cortical Labs is testing whether living neurons can compute on watts.
Private, early stage (Melbourne)The rules every memo on this page has to pass before it runs.
Nothing here is investment advice: these are my own notes on public information, written to sharpen judgment, not to recommend buying or selling anything.
Building in one of these markets, or think I have one wrong? Send it over.
Email me Book a call