Late-stage assets have never been priced higher relative to early ones. For a biotech with a year of cash, that changes what "cost" means, and where it's decided.

Most of my conversations with biotech leaders start in the same place: runway. How much does Phase II cost, how many months does it buy, what does the next round need to look like.

Lately a different question has been coming up more often, usually from the CBO rather than the CFO. Not what the trial costs, but what the asset is worth, and when. Because the answer to that second question has moved a lot in the last eighteen months, and I don't think most development plans have caught up.

So I went through the J.P. Morgan licensing and venture reports for 2025 and the first half of 2026 to see what actually happened to the price of clinical evidence.

The market repriced the clinic

Headline licensing value in biopharma reached $252 billion in 2025, up from $191 billion the year before, [1,2] and hit $166.7 billion in the first half of 2026 alone. [3,4] That part is well known. The part that matters for a biotech is how the money is split.

Only 6 to 7 percent of announced deal value is paid upfront. [2,4] Everything else is milestones, which means pharma is buying the option, and paying for it only when the data arrives. And within that thin upfront slice, the money concentrated hard. Large pharma paid an average of $185 million upfront for a Phase III candidate in 2025, up from $53 million in 2024. Over the same year, the upfront value paid for Phase I and Phase II assets fell. [2]

Venture followed the same logic. Seed and Series A rounds dropped from 228 to 191 between 2024 and 2025, and the dollars fell with them, while rounds above $100 million became routine for companies with clinical data in hand. [2] In the first half of 2026, investors kept favouring "later-stage assets, established programs and companies with clearer clinical and commercial catalysts," in J.P. Morgan's own words. [4]

The market didn't get more generous. It got more specific about what it pays for: a derisked asset, delivered on a timeline the buyer can plan around.

That reframes cost. The question stops being "what does it cost to run this trial" and becomes "what is the gap between what my asset is worth today and what it is worth the day the data reads out, and what is that gap costing me per month."

What it actually costs to climb the ladder

The direct numbers are smaller than people assume. Moving a small-molecule candidate from selection to an IND typically runs $2.5 to 5 million all-in. [5] The most cited per-trial estimates put an average Phase I at around $4 million, Phase II at around $13 million and Phase III at around $20 million, with wide therapeutic-area ranges. [6] Set those against the value step above and the picture is odd: the direct cost of a Phase II trial is an order of magnitude smaller than the upfront premium the market now attaches to a Phase III-ready asset. The clinic isn't where a biotech's money goes. It's where its valuation waits.

And most of it waits in vain. Of 100 programmes entering Phase I, about 48 reach Phase II, 14 reach Phase III and roughly seven are approved; Citeline puts the likelihood of approval from Phase I at 6.7 percent, the lowest it has recorded. [9] Read the two lines on the chart together and Phase II is the hinge. It's where the trial cost triples, where seven in ten programmes stop, and, since 2025, where the market pays least for what you have. Getting through it cleanly and quickly is the whole game.

The cost that isn't in the budget

If the direct trial cost isn't the problem, what is? Two things that rarely appear as line items: amendments and time.

Tufts CSDD's latest benchmarking wave found that 76 percent of protocols now require at least one substantial amendment, up from 57 percent in 2016. Protocols that are amended average 3.3 amendments each; in oncology Phase III the average is four. [10] The median direct cost of implementing one is about $141,000 for a Phase II protocol and $535,000 for Phase III, [11] and the interval from recognising the need for an amendment to final oversight approval now averages 260 days. Across the industry that adds up to an estimated $7 to 8 billion a year spent rewriting trials that have already started. [10]

The direct cost is annoying. The time is the real bill. Tufts' 2024 estimate puts the median value of one day of delay for a prescription drug at roughly $500,000 in lost sales, and higher in the areas biotechs concentrate on: about $840,000 a day in oncology, $880,000 in immunology, $1.3 million in haematology and $1.4 million in cardiovascular. On top of that, simply keeping a Phase II or III trial open costs around $40,000 a day. [12,13]

Put those together for a mid-size oncology biotech. One month of slippage on the path to a Phase III-ready readout is around $25 million in deferred sales value at the oncology median, plus $1.2 million in direct trial costs, before you count the extra months of runway, the dilution of the bridge round, or the partnering conversation that goes cold because the competing asset from Shanghai read out first. China-origin assets took 68 percent of upfront dollars in large licensing deals in the first half of 2026. [4] Speed to clean data is now a competitive variable, not just an operational one.

Tufts also found that about 23 percent of amendments are still classed as avoidable. [10] Those are the ones driven by design decisions, and they were all knowable before the first patient was dosed.

Runway turns delay into dilution

For a pharma company, a delayed readout is a line in the annual report. For a biotech, it decides when you raise and at what price.

EY's latest count found that 33 percent of public biotechs in the US and Europe had less than a year of cash at the end of 2025. [14,15] That is a slight improvement on 2024, when it was 39 percent, the worst figure EY had recorded in at least six years, [16] and part of the improvement is simply that there are fewer companies left to count: BioSpace tallies 219 biotechs lost over four years of the cash squeeze. [15] The share with more than five years of cash fell from 24 percent in 2021 to 18 percent. Rounds are tranched to milestones, extensions are the norm, and, in the words of EY's Rich Ramko, investors are "holding management teams' feet to the fire." [16]

Now put the pieces together. A biotech raises at the rung it's standing on. If your Phase II readout slips six months, you don't raise after it at Phase III pricing; you raise before it, at Phase II pricing, in the year the market marked Phase II assets down. Or you take a tranched extension with a milestone attached to the very readout that just moved. Either way the cost of the delay isn't the $40,000 a day the trial burns. It's the extra slice of the company you hand over to reach the same data a few months later. Deferred sales are a future P&L problem. Dilution is the founder's problem, and it's permanent.

Delay doesn't just defer the value of the asset. It decides who owns it when the value arrives.

What this means for AI in clinical development

I've spent a lot of time talking to biotech leaders about pricing, and I've come to think the most important pricing decision a biotech makes isn't in the term sheet. It's in the protocol.

Go back to the avoidable amendments and the slipped timelines and they trace to a short list of design decisions, made months before first patient in. Eligibility criteria written for the patients you'd like to enrol rather than the ones who exist at your sites, so screening fails and criteria get loosened by amendment. Site and enrollment assumptions carried over from a feasibility survey rather than from what comparable trials actually recruited, so the timeline in the board deck is wrong from day one. Endpoints and sample sizes chosen without a clear view of what regulators accepted in this indication, so the design gets renegotiated after the trial has started. Each of these is a question with an answer in existing trial data. Each of them is still, in most biotechs, answered by a static feasibility report, a handful of KOL calls and institutional memory.

That worked when the market paid for potential. It works less well when the market pays, almost exclusively, for delivered evidence on a timeline the buyer can plan around. This is where I think AI earns its place, and it isn't in replacing the clinical team. It's in giving that team, before the protocol is locked, the same depth of evidence a pharma partner will later bring to diligence: what comparable trials enrolled and where, which criteria drove amendments in this indication, which endpoints were accepted, what the realistic timeline looks like versus the one in the plan. If that removes even one amendment or shortens the path to a derisked readout by a quarter, the runway maths above says it is the highest-return spend in the company, and by a wide margin.

The market has already priced your molecule. The only question left is when you get paid, and how much of the company you still own when you do. Both are decided earlier than most teams think.

I'll keep having this conversation. Curious how others building biotech companies are thinking about it.