Founder's Perspective

Regulators are not the first line of AI governance. Investors are.

By the time a frontier system reaches a regulator, its safety posture is largely fixed — set years earlier by the incentives in a term sheet, the control rights on a board, and the horizon of the fund that wrote the check. I examined the documented governance behavior of OpenAI, Anthropic, Google DeepMind, NVIDIA, Microsoft, and Scale AI to test that claim.

The research sorts investors into four positions: Accelerator, Guardian, Neutral Investor, and Bridge Builder. Each produces different observable behavior in the companies it funds — different evaluation pipelines, different release controls, different disclosure norms. The structural finding is a tension the industry hasn't resolved: acceleration-optimized capital and long-horizon stewardship pull in opposite directions, and most funds are built entirely for the first.

Opulense is built for the second. That isn't positioning. It's the conclusion of the research.

Read the full paper →

The defining feature of this AI cycle is concentration. A small number of frontier labs absorb an extraordinary share of global venture capital, and the gravity of those rounds pulls attention, talent, and follow-on capital toward companies that have already won. The headline totals look like abundance. To a founder at formation stage, they read as scarcity.

We think that's where the opportunity is — and not for contrarian reasons.

Frontier models are becoming infrastructure, and infrastructure is historically not where durable margin sits. It sits one layer out: in the systems that make the infrastructure usable, governable, and safe enough to run a business on. Compute orchestration. Memory and data persistence. Security for autonomous agents. That layer is unglamorous, capital-efficient relative to model training, and defensible in a way an application wrapper is not.

It also gets built at pre-seed and seed, by small teams, over long horizons — precisely the profile a market optimizing for deployment velocity is worst equipped to fund. Large funds cannot write small checks; their economics forbid it. The result is a formation-stage market with fewer serious buyers than at any point in recent memory, and a set of problems that still need solving.

We're built for that gap. Family-office patience rather than fund-clock urgency. Small checks, long horizons, few relationships, real involvement.

We may be wrong. If the frontier labs absorb the enablement layer into their own platforms, this thesis compresses. We think the governance, security, and reliability demands of regulated enterprises make that unlikely — but that's the bet, stated plainly.