The Founder Tsunami Is Coming. Capital Just Got Harder. Access and Acquisition Are the New Moat.

Three things are happening at the same time, and most AI deal memos still only price one of them. One, AI made products cheap to build, which means the supply of founders and the supply of fundable-looking companies is about to surge, a founder tsunami the seed market has not yet absorbed. Two, capital at the early stage is getting harder, not easier, fewer boutique seed leads, longer cycles, tighter checks, multi-stage funds skipping seed entirely or rewriting it as Series A in disguise. Three, customers did not get cheaper to win, distribution is still scarce, attention is still finite, sales cycles in regulated and enterprise verticals didn't compress just because the model did.

Put those three together and the conclusion is uncomfortable for most current AI memos: the moat is not the product. The product is the cheap part. The moat is what sits between the product and the customer, the layers that decide whether the founder can actually reach a buyer and convert one. That's access (L4 and the distribution side of L3) and acquisition (L7 placement, L5 deep workflow, L8 retention). This piece is for VCs, an underwriting lens that sits above the deal memo and forces every other question into the right shape.

The framework calls the underlying force Current III: Capital Flows. Currents are not layers. They flow across the ten layers, and they decide whether a defensible position compounds into a business or stays a beautifully diagrammed line item on a deck.

The founder tsunami, in one sentence

When the marginal cost of building an MVP collapses, the marginal supply of founders trying does not collapse with it, it surges. That's the tsunami. Every solo developer with a Cursor seat and a Stripe account is a potential cap-table line. Every PM who can wire three APIs together is a potential pitch. Most of these will not be venture-scale companies. All of them will compete for the same scarce inputs: distribution channels, design partners, and the partner's attention at the next IC. The bottleneck is not the build. The bottleneck is everything that touches a customer.

Capital flowing the other direction makes the squeeze worse. The boutique seed funds that historically led these rounds are themselves struggling to raise their next vehicle. Multi-stage funds will write the $10M+ check for the obvious winner, but they don't price-discover the long tail. Supply of companies up, supply of capital down, supply of customers flat. That is the shape of the next 24 months at seed.

In that environment, the question "is this a good product?" stops being load-bearing. The product can be good and the company can still die for lack of a distribution edge. The question that replaces it is the access-and-acquisition question: who can this company actually reach, and what does it cost to convert and keep them?

Access as a moat (L4 + L3d/e)

Access is who you can structurally reach that a fresh founder with the same model cannot. It lives in two places in the framework:

L4 Access, the pipes layer. Connectors, APIs, agent interface protocols, identity and provenance. The companies that own meaningful L4 positions (the Plaids and Twilios of the agent era, the MCP-style connector hubs that get blessed by the platform layer) gate every agent above them. L4b (agent interface protocols) and L4e (agent identity and provenance) are quietly becoming the most under-priced sublayers on the map, because they look like infrastructure and price like SaaS but behave like toll roads.

L3d/e, the gatekeeping side of access. Editorial gates (who gets featured, who gets reviewed, who gets cited) and distribution gates (app stores, marketplaces, default placements, OS-level integrations). Law III says the surface captures attention; the gate above the surface captures the surface. A company without an access story is a company that has to buy its distribution forever. That's a P&L commitment, not a moat.

IC test for access: can this company name three structural reasons a buyer is more likely to encounter it than to encounter the next-cheapest competitor in 18 months? If two of the three reasons are "because we'll spend more on ads," the moat is rented.

Acquisition as a moat (L7 + L5 + L8)

Acquisition is the unit economics of actually landing a customer and keeping them long enough for the contract to compound. It lives across three layers:

L7 Surface, but only the parts that ride existing habit. A standalone chat surface in a crowded market is the most expensive acquisition shape there is. Surfaces that embed inside the buyer's existing tool (Slack, Outlook, the IDE, the CRM, the EHR) inherit the acquisition channel. The framework's read: L7 alone is a graveyard; L7 inside a habituated surface is leverage.

L5 Execution, deep enough that switching costs are real. A generic L5 skill ("summarize this," "draft that") gets absorbed by the L2 layer below. A deep L5 stack, L5a tool use + L5b reasoning scaffolds + L5d operating playbooks, is the company's actual product, and it's the layer that determines whether a customer renews. Renewal is acquisition you don't have to pay for again. That's where the unit economics live.

L8 Memory, even if thin in v1. Memory that compounds is the only acquisition asset that gets cheaper every quarter. L8b (user/entity profiles), L8c (aggregated network learning), L8d (institutional knowledge) are what convert a one-time install into a multi-year contract. Every cohort that stays makes the next cohort cheaper to land and easier to keep. L8 is acquisition flywheel mechanics dressed up as a database schema.

IC test for acquisition: take the company's projected CAC and LTV, and ask which layer the LTV expansion is supposed to come from. If the answer is "better marketing," the model is fragile. If the answer is "L5 depth and L8 compounding," the model is real.

The canonical read: capital is reflexive, layers are real

The definition that anchors the framework, intelligence is a supply chain; value accrues at the bottlenecks, has a financing implication most memos skip. A bottleneck isn't a bottleneck because it's technically hard. It's a bottleneck because capital can't route around it on the timescale capital expects returns.

L-1 (energy, fabs, water, skilled trades) is the cleanest example. Tens of billions flowed into L2 over the last three years and produced a generation glut, price per token kept halving. Near-zero flowed into L-1, and L-1 is now the binding constraint pinning the entire stack above it. That's not a market failure. That's capital chasing the layer with the shortest narrative arc and the cleanest comp, and ignoring the layer with a ten-year permitting cycle.

The lesson for an LP-facing partner isn't "go fund nuclear." It's that funding maps are distortion fields, not value signals. Where capital is piled up is usually where returns have already been priced in or past. Where capital is conspicuously absent relative to structural importance, today that means L1b proprietary data, L3 gates in regulated verticals, L4 access pipes, and L8 memory, is usually where the next decade's returns are sitting, untouched, because nobody's comp set fits.

The seed squeeze, sorted by layer instead of round size

Seed capital is contracting; founder demand isn't. Both halves are true. Neither half is useful until you stop sorting founders by check size and start sorting them by which layer they're actually trying to own. Sorted that way, the "seed split" everyone is talking about resolves into something more honest.

L0 / L2 ambitions (frontier models, foundation infrastructure, novel silicon). These were never a seed product. They are now Series-A-shaped rounds wearing seed clothing, led by multi-stage funds buying optionality on the next $200M round. Boutique seed funds were never going to compete here. Their absence from this segment isn't a contraction, it's a category correction. Underwrite this as a Series A, price it as a Series A, or pass.

L1 plays (proprietary data, behavioral data, outcome data). The most under-capitalized layer in the current market. Building a real L1b corpus is slow, contract-heavy, regulator-adjacent, and produces revenue late. It doesn't fit the multi-stage J-curve and it doesn't fit the micro-fund check. L1 is where capital is most distorted today. A patient, focused seed at this layer is the right instrument and almost nobody is writing it. If you have the conviction and the time horizon, this is the cleanest mispricing on the table.

L3 plays (compliance, quality gates, provenance, editorial / distribution gates). Underwritten as L5 wrappers, priced as features. Law IV says L3 above L2/L5 is structurally permanent in regulated industries. The boutique funds that understood this layer are exactly the funds struggling to raise their next vehicle, so the layer keeps getting served by generalists who don't underwrite it differently. The market is pricing L3 like a feature; the framework prices it like a moat. That gap is the opportunity for any fund still operating here.

L5 / L7 niches (vertical execution, embedded surface). Where the micro-fund and incubator world is correctly concentrating. AI did make the product cheap to build at L5 and L7. It did not make the customer cheap to win. The right capital shape is small, fast, and niche, with explicit acceptance of mid-cap outcomes rather than decacorns. The "smaller bets" half of the seed split is healthy financing for a healthy layer; treat it as a portfolio strategy, not a downgrade.

L8 plays (memory, institutional knowledge, learned world models). Almost no seed capital is targeted here yet, because L8 doesn't look like a company in year one, it looks like a schema. The funds that learn to underwrite L8 in the next eighteen months will own the next cycle's best book. Memory is the layer the framework calls the ultimate moat, and the financing market hasn't caught up.

The four laws, converted into an IC checklist

The four structural laws aren't operating advice for founders alone. Each one is a Capital Flows rule a partner can run a memo through in five minutes.

Law I, Intelligence Commoditizes Downward. A check whose only differentiation is generic L2 capability is a check the platform layer below will absorb. Wrappers don't go to zero; they get absorbed into the layer they were renting. Pricing a wrapper at a defensible-layer multiple is the most common Capital Flows mistake of the last 24 months. IC test: name the layer this company would have to own for the thesis to survive an 18-month L2 price collapse. If nobody at the table can name it, pass.

Law II, Value Accrues at Bottlenecks. Read the funding map as a distortion field. Where capital is absent relative to structural importance (L-1, L1b, L3, L8) is where the next decade's returns sit. Where capital is piled up (L2 foundation models, L6 orchestration frameworks, generic L7 chat surfaces) is where returns are already priced in. IC test: plot the company on the layer map; if it sits where every other check is going, the multiple has to be justified on something other than the layer.

Law III, Surface Captures Attention; the Chain Captures Power. A beautiful L7 demo is the easiest check to write and the hardest to defend three years later. Discount L7-only stories. Ask which deeper layer (L1, L5, L8) is being built quietly underneath. Only fund the surface if the deeper layer is real, instrumented, and on the roadmap before the Series A. IC test: cover the demo with your hand; if the deck still makes sense, the company has depth. If it doesn't, it's a surface bet.

Law IV, Generation and Verification Must Be Separate. In regulated verticals, do not fund the company trying to be both L2/L5 and L3. It will lose to a focused L3 player above it (Vanta over AWS, Snyk over Copilot, audit over the system being audited). Underwriting an integrated stack in a Law IV vertical is a category error, not a portfolio bet. IC test: in any regulated category, ask explicitly whether the gate-owner is in the deal or above it. If the founder is trying to be both, the company is structurally short the gate-owner that will eventually appear.

Those aren't four rules. They're a single underwriting checklist: name the layer, test for the bottleneck, discount the surface, respect the gate. Five minutes per memo. Most memos won't survive it. That's the point.

The meta-layer in one diagram

If a fund wants one operating habit out of this whole essay, it's this: before any AI memo gets read line-by-line, the deal partner answers four questions on a single page.

1. Which layer is this check buying? Not "AI." Not "agents." A layer. If the answer requires more than one sentence per layer, the team hasn't decided yet, and the round will fund indecision.

2. Is that layer structurally durable, contested, or absorbed? Durable layers (L1b, L3 in regulated verticals, L5 deep, L8 compounding) earn defensibility multiples. Contested layers (L6, generic L7) earn execution multiples at best. Absorbed layers (L2 wrappers, generic L5) don't earn defensibility multiples no matter how good the deck is.

3. Which Current is pushing capital toward or away from this layer right now? Demand Gravity (where the budget actually sits), Attention Economics (who owns the on-ramp), Capital Flows (the funding distortion field). If two of the three Currents are working against the company, the round has to be priced for that headwind, not against last quarter's comp.

4. What does the layer look like in 18 months if Law I plays out? This is the only question that matters for series math. If the answer is "absorbed by the layer below," the company is a feature, not a business. Price accordingly.

Four questions. One page. They sit above the memo, not inside it. Hence the meta-layer.

What the partner does with this on Monday

Three practical moves, none of which require a strategy offsite.

Re-tag the existing book by layer. Take the last twelve portfolio companies, ignore the sector tags, and write down the SCoI layer each one actually owns. Most portfolios discover they're triple-allocated to L6 or L7 and structurally underweight L1 and L8. That single audit changes the next three checks more than any thesis document will.

Add a layer line to the memo template. One required field at the top: "Primary layer owned (L#), secondary layer instrumented, layer this company will be absorbed into if Law I plays out." If the team can't fill it in, the memo isn't ready. If they can, every other section gets shorter.

Underwrite the unfashionable layers deliberately. L-1 (resources), L1b (proprietary data), L3 (gates in regulated verticals), L8 (memory). These are the bottlenecks the funding map is starving. They are also where the next cycle's returns will be concentrated, because the comp set doesn't fit and nobody is bidding the price up.

The closing read

The seed market isn't really splitting in two. It's being resorted by layer, and the round-size split is the visible artifact. Capital Flows is doing what the framework predicts it always does: overheating the fashionable layer (L2 still, L6 next), starving the unglamorous one (L-1, L1b, L3, L8), and forcing investors and founders to pick the layer first and the financing instrument second.

The meta-layer for VCs is not a new framework. It's a habit. Before the memo. Before the comp. Before the meeting. Name the layer, test the bottleneck, discount the surface, respect the gate. The four questions on the page above. The rest of the diligence does what diligence is supposed to do, but with the structural question already answered, the diligence converges instead of meandering.