AbCellera Spent $1 Billion Building an Antibody Factory. ABCL635 Is Its First Real Test.

The easiest way to misunderstand AbCellera is to call it an artificial-intelligence antibody company.

Software is part of the system, but the system is the company: immunization, single-cell screening, antibody libraries, wet-lab assays, protein engineering, translational biology, process development, clinical operations and a 130,000-square-foot manufacturing plant. AbCellera says in its 2025 annual report that it has invested approximately $1 billion over nearly 15 years to connect those pieces into a machine that can take an idea for an antibody drug from target selection to early clinical manufacturing.

The second easy mistake is to call it a contract-research business. That is what funded and trained the machine. It is no longer the main reason the machine exists.

AbCellera began reducing ordinary discovery partnerships in 2023, committed to an owned pipeline in late 2023 and brought its first two internal candidates into the clinic in 2025. The old model collected research fees now and retained milestones and royalties if a partner succeeded later. The new model spends AbCellera's own money to choose targets, create candidates and own much more of the eventual drug.

That is why the June 2026 quarter can show only $4.1 million of revenue against $46.0 million of R&D without describing a broken services company. The revenue contraction is partly the strategy working. It is also why the investment has become much riskier: AbCellera has exchanged a larger stream of small fees for a smaller number of clinical bets that can fail.

The company in one frame

AbCellera is now three assets under one roof:

1. An owned pipeline. ABCL635 and ABCL575 are clinical; ABCL688 and ABCL386 are in IND/CTA-enabling work; more than 20 other internal discovery programs remain mostly undisclosed.
2. A partner portfolio. At June 30, partners led 35 programs that AbCellera believed were progressing and in which it held a downstream stake. Twelve partnered molecules with those stakes were understood to be in the clinic.
3. A reusable drug-creation platform. The company has substantially finished the large facility build and is shifting capital from constructing capabilities to using them.

Those numbers must not be added together as though 35 partner programs, 12 clinical molecules and 20 internal starts were equivalent lottery tickets. They are at different stages, AbCellera does not control the partner programs, and most drug programs fail. They show breadth, not probability.

What the antibody factory actually does

An antibody drug begins with a biological proposition: bind this target, in this location, with this kind of activity, without creating an unacceptable safety or manufacturing problem. Finding something that attaches to the target is only the first filter.

The eventual molecule also has to bind the right epitope, produce the intended biological effect, avoid unwanted cross-reactivity, remain stable, survive formulation, be manufacturable at useful yield and behave predictably in the body. A candidate that excels at one property and fails another is not a drug.

AbCellera's process joins six layers.

1. Generate biological diversity

The company primarily starts from natural immune responses rather than relying only on a synthetic sequence library. Its immunization systems and Trianni humanized rodents generate antibody-producing immune cells against a selected target. The logic is that an immune system has already applied some biological selection for specificity and viable antibody structure. AbCellera's platform overview shows how those discovery systems connect to engineering, development and manufacturing.

This is not intrinsically superior for every target. Computational design and synthetic display can search spaces natural immunity does not produce. It is AbCellera's chosen starting advantage, especially when combined with several discovery methods rather than used alone.

2. Search one cell at a time

AbCellera's microfluidic and single-cell systems isolate antibody-producing cells, screen large populations and recover the genetic sequences behind interesting antibodies. The benefit is not merely speed. Different assays can ask functional questions during discovery instead of waiting until a short list has already been chosen.

3. Characterize function, not just binding

Hits are tested for potency, specificity, kinetics and biological activity. For difficult membrane proteins or multispecific antibodies, this can require engineered cell systems, high-throughput functional assays and in-vivo models. The output is a panel of candidates with measured trade-offs, not one supposedly perfect antibody.

4. Engineer the molecule

Protein engineering changes affinity, valency, format, half-life, Fc activity and manufacturability. ABCL575, for example, carries an Fc modification intended to silence Fc-mediated activity and extend half-life. A T-cell engager may combine a tumor-binding arm, a CD3-binding arm and sometimes a co-stimulatory component in a structure whose geometry materially changes its behavior.

5. Select for development

The company brings translational science and process development into candidate selection. That matters because discovering an elegant molecule that cannot be manufactured, formulated or dosed is an expensive way to fail. Experimental data flow into a common system with computational tools used to compare candidates and improve future decisions.

6. Manufacture early clinical material

The Vancouver facility can perform GMP cell banking and make drug substance for early trials. In-house manufacturing does not make late-stage development cheap, and AbCellera still relies on outside clinical organizations. It does reduce handoffs between discovery, cell-line development, process development and first-in-human supply. For novel multispecific formats, control of that handoff may be as important as screening speed.

The moat, if one exists, is therefore not a single machine or model. It is the accumulated integration of biology, assays, data, engineering and manufacturing. That is harder to copy than a screening instrument, but also harder for an investor to verify. Clinical output is the test.

Solution one: antibodies for GPCRs and ion channels

G protein-coupled receptors and ion channels sit in cell membranes and regulate signalling across neurology, metabolism, immunology and many other systems. They are rich drug targets, but their multi-pass structures can be difficult to purify in a native shape and difficult to present to conventional antibody-discovery systems.

AbCellera has made these complex transmembrane proteins a major internal focus. Roughly half of its internal programs target GPCRs or ion channels, according to the 2025 annual report. The platform uses specialized immunization, membrane-protein presentation, single-cell screening and functional assays to find antibodies that do more than bind an isolated fragment.

Two disclosed assets come from this capability: ABCL635 and ABCL688. That makes ABCL635 more important than an ordinary lead program. It is the first clinical readout from one of the company's claimed areas of technical differentiation.

ABCL635: block the thermostat signal with a long-acting antibody

During menopause, falling estrogen removes part of the restraint on KNDy neurons in the hypothalamus. Neurokinin B continues activating the neurokinin-3 receptor, or NK3R, which can overactivate heat-dissipation pathways and produce hot flashes and night sweats.

ABCL635 is an antibody designed to bind NK3R and block that signal. The target is already clinically validated by oral small molecules. The product thesis is different delivery: use an antibody's long half-life to replace a daily tablet with an infrequent subcutaneous injection, while potentially avoiding small-molecule liver toxicity and monitoring.

That last sentence remains a hypothesis. The approved landscape already includes Astellas's once-daily NK3 antagonist Veozah and Bayer's once-daily NK1/NK3 antagonist Lynkuet, alongside hormone therapy and inexpensive off-label medicines. Veozah carries an FDA boxed warning for rare serious liver injury and requires liver testing; Lynkuet also requires hepatic monitoring. ABCL635 may eventually differentiate on safety, efficacy or convenience, but a 92-patient trial cannot establish that commercial profile.

The Phase 2 result was nevertheless strong. One 600 mg subcutaneous dose reduced moderate-to-severe events by 8.8 per day at week four, versus 3.5 for placebo: a placebo-adjusted difference of 5.3 events per day, with p<0.001. Frequency fell 83% from baseline versus 33% for placebo. Severity also met its primary endpoint, and no serious or severe adverse event or discontinuation was reported through four weeks. Headache, fatigue and injection-site reaction were the most common events occurring more often than on placebo.

The limitations are just as important: 46 treated participants, four-week primary data, top-line rather than final results, no active comparator and no Phase 3 yet. A successful pivotal program must reproduce the effect in many more patients, characterize repeat dosing and longer-term safety, and establish a regimen people prefer to oral competitors.

AbCellera is also exploring oncology-related vasomotor symptoms, where hormone therapy may be unsuitable, with a Phase 2 start contemplated for 2027. That is a sensible extension of the mechanism, not yet a second validated market.

ABCL688: the second GPCR/ion-channel proof point is still hidden

ABCL688 targets an undisclosed multi-pass membrane protein for an undisclosed autoimmune indication. It entered IND/CTA-enabling work in the second quarter of 2025, and AbCellera anticipates a Phase 1/2 patient trial in 2027.

This is simultaneously encouraging and impossible to underwrite. It shows the complex-membrane platform has produced a second development candidate. Without a target, indication, preclinical package or trial design, nobody outside the company can judge differentiation, competition or commercial value. Until disclosure or clinical entry, ABCL688 is evidence of platform throughput rather than an asset with a defensible valuation.

Solution two: multispecific T-cell engagers

A conventional antibody usually binds one biological target. A T-cell engager is designed to bind a target on a diseased cell and CD3 on a T cell, physically bringing the immune cell into position to kill. Multispecific formats can add another tumor target, tune CD3 activity or introduce a co-stimulatory signal.

The opportunity is precision immune activation. The difficulty is controlling it. An engager can activate T cells too broadly, hit healthy tissue carrying the same antigen, exhaust the immune response or create a molecule too complex to manufacture reliably. Format, binding strength, geometry and assay selection are part of the drug, not packaging around it.

AbCellera says it spent roughly five years building a platform with proprietary CD3 binders, co-stimulatory arms, multiple protein formats, functional assays and in-vivo models. This is not a first-mover field: Amgen, Genmab, Regeneron and many other large biopharma companies already have deep multispecific and T-cell-engager capabilities. AbCellera's claimed edge is the breadth of interchangeable components and the ability to screen many combinations as an integrated system, then manufacture a selected format for early clinical work.

The best external validation so far is not an AbCellera-owned clinical TCE. It is who has paid to use the platform.

• AbbVie has a multi-target oncology collaboration announced in January 2025, following an earlier multi-target agreement.
• Jazz Pharmaceuticals committed to two GI-cancer and solid-tumor programs plus a third due to start within 12 months. AbCellera receives $56 million for the first two and $28 million for the third, then may receive up to $792 million per program in option and milestone payments plus mid-single-digit to low-double-digit royalties.
• Vertex is paying $28 million upfront for multispecific TCE work in autoimmune diseases and other conditions, funding the research while retaining development and commercialization rights.

Headline milestone totals are not contract value today. Most require years of successful development and sales, and many will never be earned. The upfront cash matters because it shows sophisticated drug developers performed diligence and chose the platform. It also helps finance internal programs without surrendering every internal asset.

The rest of the disclosed pipeline

ABCL575: useful biology, but not currently a monotherapy franchise

ABCL575 binds OX40 ligand, interrupting the OX40/OX40L signal that helps activate inflammatory pathways and form memory T cells. The molecule is Fc-silenced and half-life extended, with atopic dermatitis as the lead indication and possible use in other inflammatory or autoimmune conditions.

The Phase 1 trial in healthy participants completed dosing, with top-line safety and pharmacology data expected in the fourth quarter of 2026. It does not test efficacy in atopic dermatitis.

More importantly, AbCellera says it does not presently anticipate developing ABCL575 past Phase 1 as a monotherapy. The program began as a co-development project with EQRx, and AbCellera took control when EQRx was acquired. A clean Phase 1 could support out-licensing, combination work or another strategic use, but it should not be treated as a second fully funded commercial program.

The competitive bar is high. Atopic dermatitis already has approved antibodies from Regeneron/Sanofi and Lilly, oral agents from AbbVie and Pfizer, and clinical antibody programs from Kyowa Kirin, Apogee and Bristol Myers Squibb. Less frequent dosing alone will not be enough unless the biology produces comparable or better control.

ABCL386: an oncology candidate with no target disclosed

ABCL386 is in IND/CTA-enabling work for an undisclosed oncology target and indication, with a Phase 1/2 start anticipated in 2027. The same rule as ABCL688 applies: its existence supports the claim that the engine is producing candidates; its value cannot be analyzed until AbCellera discloses what the candidate is trying to do.

The 20-plus discovery programs

The company reports more than 20 active internal discovery programs across several indications and modalities. This is the reservoir from which future candidates should emerge. It is not a pipeline table of 20 investable assets. Targets, development criteria, ownership structures and spending are mostly undisclosed.

The platform thesis will become more credible when AbCellera repeatedly converts that reservoir into named development candidates, files CTAs or INDs on schedule and produces clinical data without R&D growing in direct proportion to program count.

Partners, clients and the portfolio AbCellera does not control

AbCellera has worked with large pharmaceutical companies including Lilly, AbbVie, Regeneron, Gilead, Novartis, Pfizer, Biogen, Moderna and Teva, as well as smaller biotechnology companies such as Rallybio, Abdera, Denali and Kodiak. The list is evidence of broad customer acceptance, but historical partnership announcements are not the same as active revenue-generating contracts.

The old economic model had three layers:

• access and research fees during discovery;
• milestones when a partner advanced a candidate;
• royalties if a drug reached the market.

The 2025 contract portfolio's 5th-to-95th-percentile royalty range was 1.5% to 9.0% for agreements signed from 2020 through 2025, compared with 0% to 4.0% for older 2015–2019 contracts. Some deals also include equity or co-investment rights.

That portfolio can produce extraordinary but irregular outcomes. Lilly's COVID antibodies bamlanivimab and bebtelovimab proved that AbCellera could discover useful molecules quickly and generated a royalty windfall. They did not prove a durable annual royalty stream. At June 2026, the 35 partner-led programs and 12 clinical molecules still offered diversified downstream exposure, but the partners decide what to advance, disclose and discontinue.

This is why partnership revenue is a poor quarterly scoreboard. A program can be scientifically alive while producing little current revenue, and an upfront payment can arrive in cash while remaining deferred on the income statement. Jazz's upfront caused free cash flow to look strong in the June quarter without turning the quarter profitable.

Where AbCellera sits in the competitive landscape

There is no clean public peer because AbCellera overlaps four groups.

The distinctions matter.

Absci emphasizes de-novo computational design and a rapid AI-to-wet-lab loop. Alloy distributes enabling technologies broadly. CROs sell execution capacity. Regeneron and other large drug companies own integrated platforms but primarily use them for their own pipelines. AbCellera's position is between those models: natural-immune discovery plus computation, external partnerships plus owned programs, and early manufacturing rather than full commercial production.

Its strongest competitive claim is vertical integration around hard antibody problems. Its weakest position is downstream. It has not independently run Phase 3, obtained approval for an owned medicine, built a commercial organization or launched a proprietary product. Large pharma partners have all of those capabilities and far more capital.

The competitive question is therefore not whether AbCellera can find antibodies. More than 100 partner starts and the COVID molecules answered that. It is whether the integrated platform selects better drugs often enough to compensate for the cost and risk AbCellera now keeps.

The business-model transition: from fees to ownership

Under the old model, a partner supplied the target and financed most downstream development. AbCellera earned modest research revenue, retained a thin claim on a broad portfolio and avoided the most expensive clinical years.

Under the new model, AbCellera chooses the target, finances discovery and early clinical development, and decides whether to keep, partner or out-license the asset. Success captures much more value. Failure consumes much more cash.

The June quarter shows the transition numerically:

• partner-led programs with downstream stakes fell from 44 at December 2025 to 35 at June 2026;
• revenue fell to $4.1 million;
• R&D rose to $46.0 million;
• SG&A fell to $13.9 million, consistent with a company reorganizing around pipeline execution;
• the manufacturing plant began depreciation, adding cost after the construction phase ended.

This is not a SaaS transition where recurring revenue temporarily hides behind deferred billing. It is a biotechnology transition where revenue is deliberately replaced by clinical probability. The correct operating indicators are candidate nominations, IND/CTA filings, trial starts, data quality, partner advancement and R&D per successful stage transition.

The scalability test is especially important. If several new programs enter the clinic while R&D remains around today's level, the fixed platform investment is being spread across more shots. If each additional program requires another dedicated organization and R&D rises one-for-one, the factory has less operating leverage than the metaphor suggests.

What being Canadian changes

Vancouver is not an incidental headquarters. It shaped the capital structure, facilities and workforce.

Canada subsidized the build, but not all of the money is free

Since 2020 AbCellera has received commitments for CAD 475.6 million of government financing: CAD 175.6 million connected to pandemic-response and antibody infrastructure, then CAD 300 million from Canada and British Columbia toward an eight-year, CAD 701 million campus and pipeline project.

The second package is intended to build Canadian capability through Phase 1 and support up to 17 internal programs. It offsets eligible R&D and infrastructure spending: AbCellera recorded $3.9 million of grants and incentives in Q2 2026.

Calling all of it non-dilutive is correct in the narrow equity sense and incomplete economically. Parts are repayable or conditionally repayable. Of the federal 2023 commitment, CAD 56.2 million is non-repayable, CAD 78.8 million repayable and CAD 90.0 million conditionally repayable, with repayment beginning in 2033. British Columbia can receive up to CAD 64.0 million from 2032 if revenue thresholds are met. The earlier federal package also contains a conditionally repayable portion tied to revenue.

The state has effectively co-financed a strategic domestic asset in exchange for facilities, jobs, clinical capability and potential future repayments. That is cheaper than issuing the equivalent equity during the build, but it is not a gift with no strings.

Vancouver concentrates the operation

AbCellera had 562 full-time employees across Canada, Australia and the United States at year-end 2025; 65% were scientists and 13% engineers or data scientists. Management argues Vancouver and Sydney combine strong universities, livability and deep technical talent with less-developed local biotech sectors, reducing competition for some recruits.

Concentration also creates exposure. Most revenue is denominated in U.S. dollars, while a substantial portion of expenses and government balances is in Canadian dollars. The company does not hedge that foreign-exchange exposure. A stronger Canadian dollar raises the U.S.-dollar cost of Canadian work; a weaker one does the opposite.

Canada provides a credible Phase 1 regulatory route—Health Canada cleared the first two internal trials—and a publicly supported manufacturing base. It does not remove the need for U.S. FDA development, global trial sites, late-stage manufacturing and commercial partners. The largest drug market and many of AbCellera's customers remain outside Canada.

There is also a strategic constraint: as a Canadian business with sensitive technology and government-supported infrastructure, a future acquisition can face Investment Canada Act review. That does not prevent a transaction, but it can add a policy gate that a comparable U.S. target may not have.

The balance sheet and what the market is already paying for

AbCellera ended June with $540.1 million of cash and marketable securities, then raised approximately $187 million net in August. Adding the new common shares and nearly share-like pre-funded warrants gives roughly 327 million diluted economic shares and about $727 million of pro-forma cash and securities, or $2.22 per share. The $110 million of still-available government funding is reimbursement capacity, not cash and not part of that number.

At the $12.50 close on August 25, the equity was worth approximately $4.09 billion and the cash-adjusted enterprise value approximately $3.36 billion. The market is not valuing AbCellera as cash plus a research shop. It is already paying billions for ABCL635, the partnered stakes and the possibility that the platform creates more drugs.

Our AbCellera model makes that tension explicit. Its base case values the shares at $5.18, 59% below the August 25 price. It assumes ABCL635 does not generate revenue until 2030, ramps toward a Veozah-sized annual sales base rather than a multibillion-dollar prevalence fantasy, keeps quiet licensing revenue quiet and gives no explicit product value to ABCL688, ABCL386 or the 20-plus discovery programs.

That is not proof the stock is worth $5.18. It is a description of what $12.50 requires beyond the conservative base. ABCL635 must become materially larger, launch sooner, carry better economics, or the rest of the platform must produce value the model deliberately does not count. Raising the exit multiple cannot substitute for answering those questions.

The August offering improved runway after the strongest clinical result in the company's history, but it also diluted existing holders by roughly 6.3% on the post-offering economic share count. That is the other side of ownership: when AbCellera funds Phase 3 rather than handing the asset to a partner, shareholders fund it too.

What would prove the factory works

ABCL635 has established target engagement and an unusually strong early efficacy signal. It has not established a repeatable drug factory.

The next two years offer a clean test:

1. ABCL635's full follow-up and Phase 3 design. Repeat dosing, longer safety, trial size, comparator, endpoints, timing and who pays will determine whether the Phase 2 result becomes a drug program.
2. ABCL575's Phase 1 readout. The relevant outcome is not only clean safety. It is whether the pharmacology is good enough to attract a partner or support a new development strategy after AbCellera stops monotherapy development.
3. ABCL688 and ABCL386 entering the clinic in 2027. Target disclosure and on-time CTAs would turn two labels into testable programs.
4. Partner TCE candidates. Jazz, Vertex and AbbVie advancing molecules would provide independent evidence that the TCE component works beyond discovery contracts.
5. Pipeline throughput without proportional spending. More clinical starts on roughly stable platform cost would demonstrate reuse. Delays and rising R&D with no candidate output would demonstrate the opposite.
6. Movement in the partnered portfolio. The count of progressing programs matters less than partnered molecules reaching new clinical stages and producing milestones.

The bull case is not that every one of those branches succeeds. Drug development does not work that way. It is that the same integrated system can generate enough independent, differentiated candidates that a minority of successes pays for the failures.

ABCL635 is the first reason to take that claim seriously. It is also only one reason.

Platform scope, pipeline status, partner counts, royalty ranges, workforce and Canadian financing terms are from AbCellera's 2025 annual report and June 2026 quarterly filing. ABCL635 results are company-reported top-line Phase 2 data from August 10; comparisons across drugs are not head-to-head. Jazz and Vertex economics are from the companies' collaboration releases; milestone maxima are contingent, not present value. Cash, expenses and revenue are reported; post-offering shares, cash per share, market capitalization, enterprise value, dilution and the 59% model gap are R40 arithmetic. The $5.18 fair value, 2030 launch and ABCL635 commercial ramp are R40 assumptions, not company guidance. Price is the August 25, 2026 close.