Sector investing is a structural choice. A sector-thesis fund concentrates capital and brand into a narrow domain in exchange for informational edge and founder-side reputation. The trade-off is concentration risk — and getting the timing of a sector wrong can cost a vintage.
AI / ML — the 2024–2025 megacycle
The 2024 venture market saw extraordinary concentration: AI-related companies received roughly 37% of all global venture funding, the highest sector concentration since the 1999–2000 internet wave. Within AI, capital split between infrastructure (foundation-model labs, GPU cloud, training data) and application (vertical AI products, agentic systems). The infrastructure layer is dominated by a small number of capital-intensive players; the application layer is fragmented and competitive.
Diligence questions specific to AI: (1) Model moat — does this company have a structural advantage that survives the next foundation-model release? (2) Data network effects — does usage create proprietary training data that improves the product? (3) Cost structure — what is gross margin after model inference cost, and how does it evolve as token prices fall? (4) Go-to-market — is this a product the customer's existing IT can adopt, or does it require a multi-quarter integration?
Deep tech — long timelines, non-dilutive capital
Deep-tech companies (advanced manufacturing, biotech, novel materials, fusion, quantum) require 10–15 year development cycles and capital structures designed for that. The capital stack often includes government grants (DARPA, ARPA-E, NSF), strategic partners, tax credits and venture debt alongside primary equity. The diligence is technical: dedicated technical advisors, lab visits, and milestone-based financing rounds rather than time-based ones.
Commercial space
From 2014 onward, the commercial-space ecosystem has grown along three vectors: launch (SpaceX dominant; Rocket Lab, Blue Origin, others competing), earth observation (Planet Labs, BlackSky, Maxar), and satellite internet (Starlink, OneWeb, Amazon Kuiper). Each has different capital intensity, different revenue models, and different defence/dual-use dynamics. The sector's investability has expanded as launch costs fell from ~$10K/kg to ~$2K/kg.
Four worked sector theses — embedded finance, defense tech, applied AI infrastructure, and value-based healthcare
Embedded finance. The secular driver is the shift of financial services delivery from bank-owned channels to the software layer inside non-financial products — payroll, fleet management, e-commerce, and vertical SaaS platforms that now distribute lending, insurance, and payments natively. The structural advantage today is that Banking-as-a-Service (BaaS) regulatory infrastructure has matured enough to make sponsor-bank partnerships commercially viable at scale without a full bank charter. The target archetype is a vertical SaaS company with 5,000+ SMB customers that has not yet monetised its payment or credit surface — where the GP can fund the embedded-product build and underwrite the attach-rate expansion. The thesis is wrong if the BaaS regulatory environment tightens materially — specifically if sponsor-bank arrangements come under the kind of enforcement that has already hit several BaaS-heavy banks — in which case the cost and time to stand up compliant infrastructure rises significantly and the model does not close at the target margin.
Defense tech. The secular driver is the shift of US defense procurement from the traditional prime-contractor model toward commercially developed dual-use technology — autonomous systems, software-defined communications, AI-enabled ISR, and low-cost attritable platforms — accelerated by the lessons of the Ukraine conflict and the NDAA reform agenda. The structural advantage is that DoD Other Transaction Authority (OTA) contracting now provides a faster acquisition pathway that commercial-stage companies can navigate without a decade-long prime relationship. The target archetype is a founder-led company with a working prototype and an active DoD program-of-record sponsor at the GS-15 or SES level, raising a Series A or B on the basis of a signed CRADA or OTA. The thesis is wrong if procurement reform stalls and traditional prime incumbents successfully defend the acquisition process against commercial entrants — timeline risk is the dominant killer in this sector.
Applied AI infrastructure. The secular driver is the 2023–2025 explosion in enterprise demand for AI inference and fine-tuning capability outside of the hyperscaler APIs — driven by data-sovereignty requirements, latency constraints, and the margin math of running large workloads at scale against per-token pricing. The structural advantage is that the open-weight model ecosystem (Llama, Mistral, and successors) has created a supply of high-quality base models that infrastructure companies can build on without foundation-model-level capital. The target archetype is a company building the tooling layer — inference optimisation, model routing, fine-tuning pipelines, or vector retrieval infrastructure — that enterprises need to run AI workloads on their own compute at costs below hosted API pricing. The thesis is wrong if the hyperscalers discount their hosted API pricing aggressively enough to eliminate the unit-economics advantage of self-hosted inference — which is the single most important variable to monitor in diligence. See the sector thesis builder in this report's lab for a structured framework to stress-test any thesis against this kind of disqualifier.
Value-based healthcare. The secular driver is the long-run shift of US healthcare reimbursement from fee-for-service (pay per procedure) to value-based contracts (pay for outcomes) — a shift the CMS has been legislating since the ACA and which commercial payers have increasingly adopted for primary care, oncology, and chronic disease management. The structural advantage is that the enabling technology — risk stratification, care-gap identification, remote monitoring — has become cheap enough to deploy at the primary-care group level without hospital-system infrastructure. The target archetype is a primary care or specialty group with 50–300 attributed lives under value-based contracts already, a demonstrated MLR improvement over the fee-for-service baseline, and a management team willing to accept PE governance in exchange for the capital to expand the attributed-life base. The thesis is wrong if the company's MLR improvement cannot be replicated across new patient panels — if the performance is physician-specific rather than model-specific — in which case the roll-up economics collapse as soon as the founding physicians depart.