PitchBook’s Q2 valuation data has the cleanest description of the current venture market. In the first half of 2026, AI companies absorbed 86% of U.S. venture dollars, and rounds of $100 million or more took 87.5% of the dollars deployed. Fortune’s Term Sheet framed it as an AI world where everyone else fights for scraps. That is broadly right, but the more useful read is about access.

Venture capital is supposed to buy uncertainty. The investor accepts illiquidity and company-specific risk in exchange for convexity. A portfolio can miss often if the few winners pay for the errors. AI has not removed that logic. It has made the right tail so visible, and so expensive, that access to the anointed names now looks like the product.

The steelman is strong. AI is one of the few software categories with obvious budget urgency, executive attention, and evidence of willingness to pay. Coding tools are the cleanest example. Cognition is reportedly discussing a new round at a valuation of at least $40 billion, less than three months after raising $1 billion at a $26 billion valuation. PYMNTS, citing Bloomberg, says Cognition’s annualized revenue run rate is nearing $1 billion, roughly double its prior financing figure. Lovable just announced a $400 million Series C at a $13.3 billion valuation, with Menlo Ventures and EQT’s Scaleup Europe Fund leading and Tencent among the new investors.

If those revenue trajectories hold, these are not empty pitch decks. A coding agent that gets into daily engineering work can sit close to payroll budgets, cloud budgets, and software delivery schedules. That is a much better spending pool than a discretionary productivity app bought by a few enthusiasts. Investors are not irrational for wanting exposure.

The risk is that a good category can still produce a bad clearing price.

At Cognition’s rumored number, the shorthand multiple is around 40 times annualized revenue run rate. That is lower than the rough multiple implied by the $26 billion round if the revenue base really doubled, so the valuation jump is not as silly as the headline alone suggests. It still assumes that a very young company can keep converting intense adoption into durable gross margin, low churn, defensible distribution, and enterprise trust. Each assumption may be plausible. The joint probability is lower than the pitch version.

Lovable’s number tells a slightly different story. A $13.3 billion valuation on a $400 million round is large but not absurd if reports of several hundred million dollars of ARR are close. The investor list matters. U.S. venture money, a European scale-up vehicle, and Tencent all entering the same round says the market is buying more than software revenue. It is buying a possible control point in how non-engineers and companies create internal software. That is a strategic option, not just a SaaS multiple.

This is where the market starts to split. The top AI companies are being priced as scarce claims on a new production function. Everyone else is being priced as ordinary software again. PitchBook’s median valuation step-up was 2.2x for AI companies versus 1.6x for non-AI companies. At Series D and later, the gulf widens sharply, with AI at 6.6x. Emily Zheng at PitchBook told Fortune that median velocity of value creation at that stage rose from $108.9 million in 2025 to more than $1 billion in 2026.

That velocity number is doing a lot of work. A late-stage company that adds a billion dollars of paper value between rounds gives existing investors a mark, new investors a reason to fight for allocation, and employees a reason to treat options as currency. It also makes the private market self-referential. The valuation validates the category, the category attracts more capital, and the next round becomes a scarce ticket.

Liquidity is the awkward constraint. Fortune notes that the IPO window has not offered much broad confirmation beyond SpaceX and Cerebras. Acquisitions are better on aggregate, with PitchBook putting 2026 acquisition value at $375.4 billion, but the examples are uneven. ServiceNow buying Armis at $7.8 billion looks like a healthy step-up from the cybersecurity company’s prior $6.1 billion valuation. Capital One’s $5.2 billion Brex deal is a haircut from Brex’s $12.3 billion peak. Exit value exists, but it does not clear every old mark.

The secondaries market gives the bluntest pricing signal. On Forge, startups that raised in 2025 or 2026 are trading at a median discount of zero to 5%, according to PitchBook data cited by Fortune. Companies that last raised in 2021 or 2022 trade at median discounts of 54% and 59%. That is not just a quality screen. It is a vintage screen. The market is punishing the old price before it even gets to the business.

For founders outside the favored AI cluster, this matters because the headline venture market is less available than it looks. A $412.7 billion first-half deployment figure sounds liquid. If 86% went to AI, and 87.5% went to megadeals, the median founder is not living in that market. The cost of capital for an ordinary good company may be rising at the same time the press release economy looks euphoric.

For AI founders, the danger is subtler. Cheap capital can hide weak unit economics, but expensive capital can also force bad strategy. When investors pay for category access, management gets pulled toward the story that justified the allocation. The company may hire ahead of repeatable demand, subsidize customers to defend growth, or promise enterprise security and reliability before the product has earned that trust. The valuation becomes a constraint on the loss function.

The coding-tools category is the right place to watch this. It has real usage, clear budgets, and obvious labor substitution narratives. It also has brutal competition from Anthropic, OpenAI, Google, SpaceX, open-source models, IDE incumbents, and internal enterprise tooling. Revenue can grow quickly while margins compress. A company can be indispensable to users and still have less pricing power than the funding round implied.

My prior is that several AI companies deserve unusual valuations. Some of them are building genuine infrastructure for how software, research, and operations get done. I would not fade the entire category because the multiples look strange. Early power-law assets often look expensive before they look obvious.

I would separate that from the claim that the venture market is healthy. A market where almost all dollars chase a small number of AI megadeals is not broad risk appetite. It is concentrated access demand. That can be rational for the funds that get into the winners. It can be painful for everyone else, including the investors who buy the same exposure one round too late.

The practical question is who is really selling optionality. If the startup has durable distribution, expanding usage, and a path to strong gross margin after compute and support costs, the valuation may be an expensive but coherent call option. If the startup is mainly selling investors the fear of missing the next platform, the option premium is being marked by scarcity rather than cash conversion.

Those are different trades. The 2026 venture market is treating them as neighbors.

Sources: PitchBook, Fortune Term Sheet, PYMNTS, Bloomberg, Yahoo Finance.