The drug royalty business thrives on a bet that an approved biologic will keep printing money along a predictable schedule. Patent expires here, biosimilar enters there, revenue declines at roughly this rate. For the better part of two decades, that bet has paid off handsomely. The announced royalty funding market hit a record $10 billion in 2025, Royalty Pharma reported at the J.P. Morgan Healthcare Conference in January. The company alone deployed $2.6 billion across the year, holding roughly 40% of the market.

Then, in April, a 40-person startup most of those royalty analysts have never heard of did something that should make them pull up their spreadsheets. Converge Bio, an Israeli AI lab founded in 2024, took cetuximab, a widely used cancer antibody worth over $1 billion a year in global sales, and produced an improved version in eight hours from a single prompt. No multi-year R&D campaign. No dedicated research team. Six edits to the original sequence, a 2.1x improvement in binding affinity over the originator and 4.4x over a competing antibody, validated in the lab, provisional patent filed.

The question worth asking now is not whether the science works. It is whether the financial infrastructure built on top of biologic drug timelines is ready for what happens when the science works this fast.

The Asset Class Behind The Curtain

A pharma royalty deal, stripped to its bones, is a claim on a drug’s future revenue. An investor pays cash today for the right to collect a slice of sales tomorrow. Biotech venture debt works on a parallel logic: lend money to a clinical-stage company, collateralized against the expectation that the asset will reach a value-creating milestone before the cash runs out. Both structures depend on the same underlying assumption: that the cashflow tail of a biologic can be modeled with enough confidence to underwrite against.

Royalty Pharma collected $3.25 billion in portfolio receipts in 2025, a 16% increase over the prior year, across royalties on more than 35 commercial products. Alongside it sits biotech venture lending: Hercules Capital closed 2025 with over $3.9 billion in new commitments and roughly $5.7 billion in assets under management, with a significant share committed to life sciences companies. These are real, large pools of capital. They price risk using depreciation curves that assume the timeline of biologic competition moves at a pace set by chemistry, regulation and patent law.

That pace, historically, has been slow enough to model.

Why The Curve Always Had Some Give

Duration risk in pharma credit has never been perfectly smooth. AbbVie filed roughly 250 patent applications around Humira — 90% of them after the drug was already on the market — and pushed U.S. biosimilar entry back to 2023, years past the original patent expiry. IV-to-subcutaneous reformulations, secondary patents and simple regulatory inertia have all bent the depreciation curve before.

But those were known variables. Credit underwriters could see them coming and price them in. Patent thickets take years to build. Reformulations require clinical programs. Biosimilar uptake delays follow patterns that actuaries can study. The wobble was bounded, continuous. What AI-generated biobetters introduce is the possibility of a discontinuous shift.

Six Edits, Eight Hours

Here is what Converge Bio actually did. Using ConvergeAB, its antibody design platform, the company fed in cetuximab’s sequence and target receptor as inputs. In a zero-shot setting (meaning no task-specific training or manual tuning), the platform generated 100,000 candidate antibodies, and the top 10 were tested in the lab using surface plasmon resonance, a standard binding measurement technique. The winning design carried six edits across framework and CDR regions and showed binding affinity 2.1x stronger than cetuximab and 4.4x stronger than a competitor’s AI-designed variant.

A few things give this result extra weight. Cetuximab’s European patent expired in 2014, yet the drug still generates roughly $1.68 billion a year without a single approved biosimilar in the U.S. or Europe. Researchers have attributed this gap to the antibody’s unusual structural complexity, which makes demonstrating biosimilarity harder than for drugs like trastuzumab or rituximab. AI-driven design may be suited to exactly this kind of problem.

The result was internal and validated by the company’s own lab. That context matters, and this article will not pretend otherwise. But Converge Bio CEO and cofounder Dov Gertz says the cetuximab experiment was not an isolated case. “We’ve now succeeded with the same workflow against four additional targets: PD-L1, CD19, HER2 and the SARS-CoV-2 spike protein,” Gertz says. “In roughly 50% of the molecules we input into ConvergeAB, we achieve at least a 2x improvement in binding affinity in a single optimization cycle, testing only the top 10 sequences.”

On cost and time, Gertz says: “If a royalty fund or a biosimilar manufacturer handed us 10 marketed biologics tomorrow, we could realistically deliver optimized lab-validated leads on all 10 within roughly a quarter, at a fraction of the cost of a single traditional discovery campaign.”

The Risk Cuts Both Ways

This is where the royalty fund analyst’s spreadsheet starts to sweat.

On one side, a third party can now plausibly produce a patentable, materially improved version of a marketed biologic in a timeframe that standard cashflow models never anticipated. That compresses the revenue tail of any drug a fund holds; the originator loses market share faster than the depreciation curve assumed.

On the other side, the originator can do the exact same thing to defend its franchise, extending the curve by deploying a prompt-driven biobetter before generics arrive. The AbbVie playbook — spend years building patent thickets around your blockbuster — might become a weekend exercise with an AI platform.

The trouble for anyone underwriting these instruments is that the uncertainty itself is the problem, more than the direction. Credit instruments can absorb bad news. What breaks them is variance they cannot hedge.

Two recent regulatory moves sharpen this. On September 9, 2025, the FDA finalized updated guidance on comparative analytical assessments for therapeutic protein biosimilars. On October 29, the agency published a draft framework reducing the clinical study burden for biosimilar approvals. Both lower the cost of follow-on biologics at the same moment AI is collapsing the design step. The timing is not coincidental; it is structural.

What The Money Is Saying

Eyal Lifschitz, founding general partner of Peregrine Ventures and chairman of the Israel Early-Stage Investors Association, offers a measured counterpoint to the enthusiasm. He points out that stronger binding does not automatically mean a better drug.

“To postulate that cetuximab efficacy in cancer therapy can be solely reduced to its binding affinity to EGFR is inexact,” Lifschitz says. “The changes to the sequence may provide not only changes in affinity but also in toxicity and in fine bringing a lesser therapeutic outcome than original cetuximab, and only clinical studies would bring an answer.”

In practical terms, Lifschitz estimates that AI-driven lead optimization could save $2 million to $4 million and one to three years in the early discovery phase of developing a biosimilar. Those are meaningful savings, but a fraction of the total cost, which remains dominated by preclinical and clinical development. “It does not fundamentally change the development of the biosimilar from a preclinical and clinical perspective, which is where most of the money is spent and the only results that matter for future revenue,” he says.

He also flags what the article’s thesis requires: these tools are double-edged swords. “The solutions may be used by a competitor to develop a biobetter or by the originator to enhance the lifecycle of a biological drug,” Lifschitz says.

What Has To Go Right For The Thesis To Hold

Honest accounting. The Converge Bio result is one experiment, validated internally, on one target, with a provisional rather than granted patent. For the broader claim to hold, that AI-generated biobetters create unhedgeable duration risk for pharma credit, several things need to come true.

The results have to generalize across targets. Early signals from Converge Bio and competitors suggest they will, but the track record is thin. Patent litigation has to allow biobetter variants outside the originator’s IP thicket, an open legal question that will be fought case by case. And originators themselves have to respond in ways that current underwriting models cannot capture. If they defend with AI-driven biobetters, the curve extends. If they do not, it compresses. Either way, the old model is wrong.

Somewhere between $200 billion and $400 billion of branded biologic revenue is expected to lose exclusivity between 2025 and the early 2030s. If biobetter generation becomes a routine, prompt-driven workflow within that window, the depreciation curves backing the current generation of royalty deals and venture debt will be mispricing risk on both sides. The question is simply whether the repricing happens orderly or not. The funds that recognize the shift earliest will adjust their books before the rest of the market catches on. The ones that wait may find out that the financial infrastructure they trusted was sitting on top of a workflow that just got rewritten in eight hours.

STANDOUT FACT: Cetuximab’s European patent expired in 2014 and the drug still generates over $1 billion a year with no approved Western biosimilar, partly because its structural complexity has kept competitors at bay. AI may have just kicked that door open.