From Narrative Premium to Proof Premium

AI is compressing the path from idea to revenue. The premium now goes to what compounds.

--  Justin Patel, Venture Investor at Decasonic

AI valuations are not cooling. The market is still paying aggressively for AI, but it is getting more precise about what it is paying for.

Anthropic raised $65B at a $965B post-money valuation, crossed $47B of run-rate revenue, and then confidentially submitted a draft S-1. OpenAI closed $122B in committed capital at an $852B post-money valuation and said it is generating $2B of revenue per month. These are not normal software financing events. They are infrastructure-scale financings for companies trying to own the next platform layer.

That does not mean every AI company deserves a platform multiple.

The better read is that the AI premium is becoming more segmented. At the top end, capital is concentrating around companies with model scale, distribution, revenue velocity, and infrastructure leverage. At the early stage, AI has changed the cost of building and the speed of learning. That is good for founders, but it also changes how investors should think about price.

The market used to pay for narrative. Then it started paying for product velocity. Now the premium is moving toward proof.

That matters even more in Web3 x AI. AI can make a product look useful earlier. Tokens can make a network look active earlier. Neither proves durability by itself. The companies that deserve a premium are the ones where AI improves the user loop, Web3 improves the economic loop, and the combined system compounds over time.

AI capital is concentrated, and that matters for early-stage pricing

The market headline is still very positive for AI, but the distribution is uneven.

PitchBook-NVCA reported that Q1 2026 U.S. venture deal value hit $267.2B, one of the largest quarters ever. The more important detail is concentration: the top five deals represented 73.2% of Q1 deal value. AI represented 88.8% of Q1 deal value, but only 42.5% of deal count.

Carta shows the same split from a private-company dataset. More than 60% of all venture capital raised by companies on Carta in Q1 went to AI companies. In SaaS, 83% of capital went to AI startups. Carta also highlighted how wide the valuation gap has become: a foundational model company at Series A might raise at a $300M median valuation, compared with roughly $55M for a non-AI startup at the same stage.

This creates a strange environment for founders. Capital is available, but not evenly. Valuations are high, but not because every company is getting a bid. The best AI companies are pulling the market up, while everyone else has to prove why they deserve to be priced like a beneficiary of the AI cycle instead of a company that gets competed away by it.

That is why the early-stage valuation question is so important. The market is not just asking whether a company uses AI. It is asking whether AI creates a real advantage in speed, distribution, data, revenue, or value capture.

AI compressed shipping

The first real change is that AI has lowered the cost of building.

Menlo Ventures estimates enterprise generative AI spend grew from $11.5B in 2024 to $37B in 2025, with the application layer capturing $19B. Coding was the breakout category, reaching $4B and accounting for 55% of departmental AI spend. 

In today’s market, the AI premium is showing up less as incremental headcount and more as direct compute spend. Salesforce is reportedly planning to spend roughly $300M on Anthropic tokens while keeping engineering headcount roughly flat.

At the same time, Microsoft is reportedly reducing reliance on Claude Code because of rising token costs, while GitHub Copilot is moving to usage-based billing.  AI is compressing software creation, but token spend is becoming a real operating cost that companies now have to manage like infrastructure.

Let’s take a look at a few case studies.  

Lovable is one of the clearest examples. The company said it reached $100M ARR just eight months after its first $1M ARR, with users creating more than 10M projects and 100,000 projects per day. Business Insider later reported Lovable reached $400M ARR, with 200,000 new coding projects created daily and only 146 employees.

Cursor shows the enterprise developer side. The company crossed $1B in annualized revenue, raised a $2.3B Series D at a $29.3B post-money valuation, and later became part of the broader xAI/SpaceX strategic orbit through a reported $60B acquisition option. 

These are not normal startup scaling curves. They are extreme examples, but extreme examples change investor expectations. Once the market sees companies go from product velocity to revenue velocity this quickly, every other AI company gets compared against the new curve.

That does not mean every founder needs to become Cursor or Lovable. It does mean “we shipped quickly” is becoming less differentiated than it was a few years ago and even a few months. YC’s recent batches having >10% solo founders and Matthew Gallagher’s Medvi are other proofpoints. AI makes it easier to launch, easier to prototype, easier to automate internal workflows, and easier for small teams to look larger than they are.

That is good for founders.

It also means investors need to look more carefully at what comes after the demo.

Revenue milestones are compressing too

The old software benchmark was that reaching $100M ARR took time, distribution, sales capacity, and organizational maturity. That is still true for most companies, but AI is changing the top end.

Bessemer’s Cloud 100 benchmark report says the average Cloud 100 company reached $100M ARR in 7.5 years, while AI companies averaged 5.7 years. Bessemer’s State of AI report breaks the market into “Supernovas” and “Shooting Stars.” The Supernovas averaged roughly $40M ARR in year one and $125M ARR in year two, while generating around $1.13M ARR per FTE. But those same Supernovas averaged only ~25% gross margins, often trading distribution for profit in the short term.

That second number is important.

Fast revenue can be real and still require a different underwriting lens.

Bessemer’s “Shooting Stars” look more like durable AI-native software companies. They average about $3M ARR in year one, $12M in year two, $40M in year three, and $103M in year four, with roughly 60% gross margins. That may be a less viral growth curve, but it may be a more financeable one depending on the category.

This is where the discussion around valuation needs more nuance. A company doing $10M of annualized usage-based revenue with low gross margins, high churn, and paid acquisition should not be valued the same way as a company doing $10M of contracted recurring revenue with expanding usage, improving margins, and deep workflow adoption. Both can be AI companies. The quality of the revenue is different.

The same is true for Web3 x AI, but the diligence bar is even higher.

A company can show wallet activity, token volume, agent transactions, or incentive-driven usage and still not have durable value creation. Those metrics matter, but they are first-order signals. The real question is whether the behavior persists when rewards decline, whether users return for the product rather than the incentive, and whether value accrues to the company, protocol, or token in a way that compounds.

This is where a lot of Web3 x AI underwriting can get sloppy. AI can make a product look more useful earlier. Tokens can make a network look more active earlier. But neither proves retention, willingness to pay, organic demand, or durable value capture by itself.

The better question is not “is there activity?” It is “what is causing the activity, and who captures value if it continues?”

That distinction matters. A token can bootstrap supply, liquidity, or participation. AI can improve automation, personalization, or decision-making. But if the only reason users show up is the reward, or if usage does not translate into revenue, fees, data advantage, network effects, or token value accrual, then the activity is not proof yet. It is a subsidy.

The strongest Web3 x AI companies will show that AI improves the user loop and Web3 improves the economic loop. When both are true, the model can become more defensible. When only one is true, the other is usually just narrative.

The speed of the curve gets attention. The quality of the curve earns the premium.

Early-stage valuations are moving up, but the market is bifurcating

This is the part founders feel most directly.

The median seed pre-money valuation reached $18.4M in Q1, more than double the 2021 figure. The median Series A pre-money valuation reached $62M, nearly triple the $21M recorded in 2020. Median Series A deal size rose to $19.6M from $7.5M over the same period.  Half of early-stage deals now exceed $10M, the highest share of large early-stage deals in the past decade.

From a post-money perspective. The median seed post-money valuation climbed to $24M in Q4 2025, up from $18M a year earlier and $16M two years earlier. At Series A, the median post-money valuation reached $78.7M, up 37% year-over-year.

The AI-specific numbers are more revealing. Median AI pre-money valuations of $18.7M at seed, $78M at Series A, $270.8M at Series B, $606.5M at Series C, and $4.7B at Series D+. Non-AI companies in the same chart were $18M at seed, $42.4M at Series A, $174M at Series B, $507.3M at Series C, and $1.285B at Series D+.

The seed-stage gap is not huge in that dataset. The gap starts to widen meaningfully at Series A and beyond. That makes sense. At seed, investors can still underwrite founder quality, insight, and speed. By Series A, they need to decide whether the company is actually becoming an AI winner.

That is why the $50M to $100M range matters.

A $10M or $20M valuation can be priced on founder quality, insight, speed, and a credible wedge early revenues. A $30M valuation usually needs a clearer reason to believe the team has early pull: revenue, users, design partners, a product loop, or a wedge that is starting to show market demand. The $50M to $100M range is harder because it sits between belief and breakout. It is often too expensive for pure narrative, but still too early for complete proof.

That does not mean $50M to $100M is wrong. Plenty of great companies should raise in that range. But the proof burden changes. At that price, the next round often needs to happen at $150M, $200M, or $250M. The company needs to create evidence quickly enough to make that next step feel natural.

For founders, the question is not just “what valuation can I get?” The better question is “what proof will this valuation require before the next round?”

AI changes this because it compresses the path to that proof. A founder can now ship faster, test more customer segments, build more product surface area, automate internal workflows, and learn from users with less capital. That should be an advantage.

But it also means investors will expect more clarity earlier.

At a lower valuation, the market can underwrite more uncertainty. The round can be about founder quality, insight, speed, and the right to keep learning. At a higher valuation, the company needs to show more than motion. It needs evidence that the product is becoming harder to replace.

That proof can take different forms depending on the category. In consumer AI, it might be retention, daily habit, organic sharing, or user-generated assets. In enterprise AI, it might be revenue quality, expansion, workflow depth, or gross margin improvement. In Web3 x AI, it might be transaction volume, wallet activity, liquidity, token value accrual, or network participation that persists after incentives decline.

The important point is that AI should not only help founders build faster. It should help founders learn faster. The best teams will use AI to shorten the cycle from product → usage → feedback → iteration → proof. They will know which metric matters, why it compounds, and what needs to be true before the next financing.

Every valuation is a promise about future proof. AI gives founders a better chance to create that proof faster, but it does not replace the need for it.

Every valuation is a promise about what proof will exist by the next financing.

The pre-seed market is also splitting

The earliest stage is not immune to this.

Q1 2026 pre-seed report says roughly 3,000 U.S. startups raised pre-seed funding in Q1, adding up to over $2.3B, with the total expected to reach around $2.9B as more data comes in. The total amount of cash is stable, but the distribution is less even. Says rounds between $1M and $2.5M fell from 24% of pre-seed rounds in Q1 2023 to 18% in Q1 2026. At the same time, AI reached 50% of all pre-seed dollars.

That is a useful signal. The pre-seed market is not dead. But the middle is thinner. Smaller checks are still happening. Larger AI-driven rounds are still happening. The $1M to $2.5M middle is becoming less common.

This matters because AI has changed what a pre-seed company can show. A founder can now build more product before raising, use AI to move faster with a smaller team, and show more customer development earlier. That can justify better terms for strong founders. It can also create pressure on founders who are still pre-product or pre-revenue, because investors may compare them against AI-native teams that already have a demo, early usage, and a faster learning loop.

The bar is not uniformly higher. It is more segmented.

Some founders will raise earlier than ever because talent, category, and narrative are enough. Others will need more proof than the same stage required a few years ago because AI made that proof cheaper to reach.

Crypto venture is selective, which matters for Web3 x AI

Web3 is in a different capital environment than AI.

Crypto and blockchain startups raised roughly $4B across 355 deals in Q1 2026. Capital invested fell about 50% quarter-over-quarter, while deal count declined only 16%, which suggests fewer mega-rounds rather than a total funding freeze. Valuations declined modestly from Q4’s high-water mark, while median crypto deal size reached a new all-time high above $4.5M, though valuation data was sparse and skewed later-stage. 

That is the backdrop for Web3 x AI.

AI capital is abundant and concentrated. Crypto capital is active but more selective. Founders building at the intersection need to speak to both markets without over-relying on either narrative.

In AI, investors are looking for speed, revenue, workflow ownership, and platform potential.

In Web3, investors are looking for usage, value accrual, token design, liquidity, community behavior, and whether the network creates something that cannot be replicated by a normal SaaS company.

A Web3 x AI company deserves a premium when those two sides reinforce each other. AI should improve the product, workflow, personalization, automation, creation, or decision-making. Web3 should improve ownership, payments, incentives, identity, settlement, access, provenance, or coordination.

If both layers improve the loop, the company can be more interesting than a normal AI startup or a normal crypto startup. If one layer is just decoration, the valuation should reflect that.

A token can amplify a premium that already exists. It cannot manufacture one from nothing.

Clean metrics will matter more

As revenue compresses, metrics get noisier.

Some AI startups are blurring ARR, CARR, and annualized run-rate revenue. In some cases,  CARR was 70% higher than ARR, even though some of that contracted revenue might never fully materialize. The problem with annualized run-rate revenue in AI, where usage-based or outcome-based pricing can make a single strong month look like predictable recurring revenue.

This is going to matter more.

AI companies can grow so quickly that the market wants to believe the numbers. But public markets, later-stage investors, and strategic buyers will eventually separate booked revenue, live revenue, usage-based revenue, gross margin, retention, expansion, and implementation risk.

Founders should be clear early. If revenue is annualized usage, say that. If it is contracted but not live, say that. If gross margin is temporarily low because the company is buying distribution, explain the path to improvement. If customers are expanding after the first use case, show the cohort data.

Clean metrics are not conservative. They are a competitive advantage.

In a market where a lot of companies will look impressive on the surface, the founders who can explain revenue quality will stand out.

What deserves a premium now

I think the premium moves to six areas.

Distribution. AI makes it easier to build product, but it does not automatically create demand. A company that owns a channel, community, workflow entry point, developer ecosystem, or consumer habit deserves a better valuation than a company relying only on product novelty.

Retention. If users come back without constant paid acquisition or token incentives, there is something real underneath. Retention matters even more in AI because the novelty curve can be steep. Many products feel magical the first time and replaceable the tenth time.

Compounding Data. The strongest AI applications should improve with usage. That can come from proprietary workflow data, user feedback, transaction history, domain-specific context, or network-level behavior. Data does not need to be massive to matter. It needs to be relevant, compounding, and hard for a generic model to replicate.

Transactions. Products closer to money movement, commerce, trading, settlement, creator monetization, or payments can capture more value if they own the workflow. This is especially relevant in Web3 x AI because wallets, stablecoins, tokenized assets, agent payments, and onchain settlement can turn usage into economic activity.

Trust. AI increases the amount of software, content, decisions, and automation in the world. That makes verification, security, identity, compliance, provenance, and reputation more valuable. In high-stakes categories, trust is not a feature. It is the product.

Network Effects. More users, creators, developers, agents, liquidity, data contributors, or node operators should make the product stronger. If usage does not strengthen the system, the business may still work, but it should not be priced like a network.

Speed. Velocity still matters. It just should not be the only premium. The best founders will use velocity to reach proof faster.

What this means for founders

This market is founder-friendly for the founders who understand the shift.

AI gives small teams real leverage. It can compress the time to prototype, launch, sell, support, and iterate. It can help founders do more with less capital and reach milestones faster than earlier software cycles allowed.

The mistake is using that leverage only to look bigger.

The better move is to use AI to learn faster. Ship faster, but measure the loop. Launch earlier, but understand retention. Get revenue sooner, but be honest about the quality. Use Web3 incentives, but prove the behavior survives beyond incentives. Raise at a strong valuation, but make sure the next milestone is reachable from that price.

The best founders are not just building faster. They are compressing the path to proof. That is the real opportunity.

The market is still paying for AI. It is paying aggressively. But the premium is changing.

Narrative gets attention. Shipping earns the first meeting. Proof earns the price.

In Web3 x AI, that is healthy. It pushes the market toward companies where AI improves the product and Web3 improves the economics.

That is what I am underwriting.  Not the label. Not the demo. Not the token by itself. The loop.

The content of these blog posts is strictly for informational and educational purposes and is not intended as investment advice, or as a recommendation or solicitation to buy or sell any asset. Nothing herein should be considered legal or tax advice. You should consult your own professional advisor before making any financial decision. Decasonic makes no warranties regarding the accuracy, completeness, or reliability of the content in these blog posts. The opinions expressed are those of the authors and do not necessarily reflect the views of Decasonic. Decasonic disclaims liability for any errors or omissions in these blog posts and for any actions taken based on the information provided.