Ramtin Naimi on AI Venture Capital’s Growing Funding Crunch

Patrick O'Shaughnessy asks Ramtin Naimi, founder of Abstract Ventures, to describe the venture market at deal level. Naimi says financings and follow-on rounds are happening at extraordinary speed, with a few sought-after firms receiving preferred allocations in popular companies.

Naimi says firms that miss the first tranche sometimes pay three to five times its price in a simultaneous second tranche instead of waiting for another round. In his view the structure is not automatically unhealthy: some AI businesses genuinely need more capital than they want to raise in the first tranche.

The speed also creates pressure to keep up. Naimi sees investors make offers quickly and sometimes withdraw or renegotiate them. Portfolio markups feel like merely keeping pace, while the biggest exits dominate attention only briefly. He sees a rhythm reminiscent of 2021, but says the present companies are creating substantial value.

Headline growth can obscure the economics. Some businesses growing 10 to 20 times a year have low or negative margins, concentrated customers, usage-based sales, or annualize a single month's data revenue. Naimi urges founders to compare growth with companies of the same type instead of treating every revenue figure as equivalent.

Data labeling can have 15–20% margins because collection is expensive, while some application providers pass much of their margin to foundation-model companies. Yet Naimi also sees margins improving quickly alongside growth. He gives a portfolio example moving from negative 15% to positive 35% margins while revenue grows sharply.

Some vertical software companies command very high revenue multiples. Naimi says the investment case can rest on a much larger labor budget rather than the current software budget—for example, legal work now done by paralegals. He acknowledges that he has not yet seen those companies actually replace that labor.

Asked to compare today with 2021, Naimi sees familiar signs: a rush of deals, public-market investors entering venture, and term sheets offered and renegotiated quickly. He cannot say whether the cycle resembles 2019 or 2021. The difference, he argues, is how rapidly today's products find users and become useful in everyday life.

O'Shaughnessy turns to limited partners, the institutions that supply venture funds. Naimi says a fund normally plans initial investments across 18 to 24 months and keeps reserves for follow-ons. Large firms have returned to fundraising much sooner because there are more deals, prices are higher and preserving ownership requires larger checks.

The portfolio trade-off is stark: deploy the same fund across fewer companies to preserve ownership, or keep the number of bets and accept smaller stakes. Fewer bets demand better selection; smaller positions limit the upside from a winner. Either choice changes the original portfolio construction.

LPs also expected more cash distributions than they have received. Naimi argues that some have substantial effective exposure concentrated in a few private companies, including SpaceX, Anthropic, OpenAI and Databricks. Liquidity from those names can matter more than the apparent breadth of the venture portfolio.

To illustrate the pace, Naimi describes one fund that raised $2.5 billion and expected to return to market roughly six months after starting to deploy it; another sought $5 billion nine months after receiving $3 billion. He says founders have also received commitments whose cash would arrive weeks or months later, once a new fund closed.

O'Shaughnessy asks whether compute-heavy AI businesses require new financing instruments. Naimi points to growth debt tied to customer value and investors offering compute instead of cash. Some founders value guaranteed capacity, but securing favorable pricing, allocating usage and accounting for hardware degradation make such arrangements complicated.

Naimi says new capital structures are difficult to introduce because founders and their investors generally prefer clean equity with understandable terms, rather than covenants that might create problems later. Still, he sees room for alternatives when billions in equity financing are destined largely for compute.

The conversation moves to a paradox: warnings about stranded compute capacity coexist with difficulty obtaining GPUs. Naimi says his portfolio companies often need help finding capacity, and some large neocloud providers are not interested in contracts below $200 million. The largest demand comes from hyperscalers, frontier labs and well-funded new model labs.

Naimi describes a growing group of new labs with unusually large capital needs. One he met was raising $1 billion for a single training run and expected to seek another $5 billion. He says it is hard to distinguish generalizable progress from systems tuned for benchmarks, especially when fundraising itself becomes a proxy for perceived success.

Could capital itself constrain compute demand? Naimi thinks financing conditions, particularly debt, may matter. Infrastructure builders can require debt several times greater than their equity, and rising borrowing costs could determine whether new capacity is economically viable.

On seed valuations, Naimi traces a succession of imagined best-case outcomes: $10 billion, then $100 billion, and now $1 trillion. A few exceptional companies change investors' sense of what is possible, but he doubts there will be dozens of trillion-dollar outcomes. Pricing every startup as though it could join them risks extrapolating from a handful of outliers.

His preferred alternative is to back exceptional founders pursuing markets that are not yet obvious. Crowded, widely recognized ideas command high prices. Proactive research into a neglected opportunity can create room for both a better entry price and a meaningful ownership position, although most overlooked opportunities will still fail.

He cites Conduit, a brain–computer interface startup Abstract backed before the category became fashionable. Naimi says its external wearable aims to infer intended text from brain signals, and describes early training data and an 80% accuracy claim. He imagines a future interface for agents, while recognizing that the company's future depends on its technology continuing to improve.

Etched is another example. Naimi says investors recognized founder Gavin Uberti's talent but many doubted a startup could compete with established AI chip makers. After a major financing made the opportunity more conspicuous, the category attracted much greater interest. His lesson is to investigate why a strong founder commits to a market before it becomes consensus.

A large number of meetings followed by only a few term sheets can signal an impressive founder working on an opportunity investors do not yet understand, Naimi says. That does not make every such deal good; it gives an investor a reason to examine the category rather than simply follow the crowd.

Valuation has become particularly hard to predict. Naimi says plausible offers for the same company can differ fourfold, and a bidder may pay three times the next offer to avoid losing a deal. Founders then compare themselves with friends who raised at similarly extreme prices, propagating those expectations.

O'Shaughnessy asks whether established, rapidly growing private businesses are a more rational place to invest. Naimi agrees they offer more evidence and can absorb large checks. But he cautions that access to a small group of obvious winners has itself become a crowded strategy, and great companies bought at high prices can still produce modest returns.

The two discuss AI data businesses. Naimi says demand from model builders has so far grown rather than disappeared, and difficult-to-collect data can command different economics. Some founders expect data sales to remain a large standalone business for many years; others sell data to fund their own robotics-model research.

O'Shaughnessy asks about a new generation of wealthy angel investors. Naimi argues that Silicon Valley's dense network of individuals willing to write personal checks before institutional financing helps new companies get started. More successful employees and angels could replenish that network.

Abstract Ventures has moved beyond seed into some Series A–C and passive growth investments. Naimi says its first fund was $100 million and its latest $800 million: LPs criticized the first as too large in 2018, while later praising the larger fund as restrained. The contrast illustrates how much the market has changed.

Naimi says the firm has often had conviction and access to its best companies but lacked enough capital to lead larger rounds. He wants to deploy available money well, then assess whether the firm's position supports raising more without weakening returns. In hindsight, he believes earlier funds could have deployed more capital with similar performance.

For a recurring closing exercise, O'Shaughnessy asks Naimi to rate the opportunity and the froth on a scale of one to ten. Naimi gives the opportunity a ten, citing the number of businesses he wants to back. He rates froth at 6.5 to seven, against ten for 2021, because he sees real growth and adoption supporting some of the enthusiasm.

Among companies and categories to watch, Naimi names AI-driven biology companies such as NewLimit, Chai Discovery and Isomorphic, as well as Applied Compute, prediction markets and Physical Intelligence. These are his examples and forecasts, not guarantees. He ends by emphasizing the difficulty and importance of retaining exceptional talent when competitors can offer attractive alternatives.