The $74 Billion Bet on a Patent Cliff

On January 3, 2019, Bristol Myers Squibb announced it would buy Celgene for roughly $74 billion, the largest pharmaceutical acquisition to that point.1 Celgene holders would receive one BMS share plus $50.00 in cash — an upfront value of $102.43 per share against BMS's January 2 close, a 53.7% premium — and, for each share, one tradeable contingent value right worth $9.00 if three drugs cleared the FDA on time. The deal closed on November 20, 2019, leaving legacy Celgene holders with about 31% of the combined company.1

Most write-ups of this deal grade it on a single axis — the stock went up, or the revenue went down — and reach a confident verdict. That confidence is the mistake. A retrospective needs two grades, not one: was the bet reasonable at the price paid, and did the outcome pay? Decision quality and result are different things, and this deal is a clean case of the two diverging.

The thesis they sold

The rationale was cash flow. Celgene threw off enormous near-term earnings, led by the multiple myeloma blockbuster Revlimid, and BMS pitched the deal as buying a decade of cash to fund its own next generation — with roughly $2.5 billion in projected cost synergies to sweeten the math. The problem, which BMS acknowledged and the activist investor Starboard Value made the centerpiece of its opposition, was that Revlimid faced a defined loss of exclusivity. Starboard's own filing put pre-deal BMS three-year total shareholder return at −18.1% against the S&P 500's +30.6%, and argued the company was buying one of the largest patent cliffs in industry history.8 This was not a hidden risk. It was the risk, in the prospectus, in daylight.

The asset's fate — a clock, and a miss

The most revealing structure in the deal was the CVR. Its $9.00 was all-or-nothing across three FDA approvals, each with its own deadline: ozanimod (Zeposia) and lisocabtagene maraleucel (liso-cel / Breyanzi) by December 31, 2020, and idecabtagene vicleucel (ide-cel / Abecma) by March 31, 2021.3 Miss one, and the entire right expired.

Two milestones cleared comfortably. Ozanimod was approved 281 days early; ide-cel came in five days ahead of its deadline. Liso-cel did not. A pandemic-delayed inspection of a third-party viral-vector plant pushed its approval to February 5, 2021 — 36 days past its December 31, 2020 deadline.4 Because the structure was all-or-nothing, the near-miss on one drug voided the payout on all three. The CVR terminated automatically on January 1, 2021, delisted from the NYSE, and the roughly $6.4 billion of potential payments went to zero.5

The lesson a good deal structure teaches here is about fragility, not about the drugs. Every one of these therapies was eventually approved and marketed. The value destruction was entirely in the joint condition — three independent regulatory timelines multiplied into a single fragile promise, where a logistics delay on one plant was enough to zero the whole instrument.

Did it pay — the accounting

Revlimid did exactly what everyone said it would. It peaked at $12.8 billion in 2021, its last full year of exclusivity, then fell as Teva launched the first U.S. generic on March 7, 2022 — to roughly $9.98 billion in 2022 and about $6.10 billion in 2023.67 The cliff arrived on time.

And yet here is the fact that should unsettle anyone grading this deal by revenue: BMS has never taken a goodwill-impairment charge on the Celgene acquisition.7 Impairment is the accounting system's confession that a company overpaid — and in the deal that supposedly bought a doomed cliff, the confession never came. Revenue fell exactly as predicted, and the balance sheet recorded no mistake. Two "yardsticks" pointing in opposite directions, on the same deal.

Did it pay — the market

Acquirer total shareholder return is the yardstick reporters reach for most, and here it is nearly useless. Celgene was folded into a company many times its size, so BMS's share price in the years after the close reflects its entire portfolio, its own patent cliffs (Eliquis, Opdivo), and the market's mood far more than the merits of this one transaction. BMS did lag the broad market for years afterward — but attributing that lag to Celgene specifically is a story the data cannot actually support. When a metric can't isolate the thing you're trying to judge, using it anyway is not rigor.

The context

Two footnotes matter. First, antitrust: to clear the FTC, Celgene had to divest the psoriasis drug Otezla, which it sold to Amgen for $13.4 billion in cash — proceeds earmarked for paying down the deal's substantial debt.2 Second, litigation: CVR holders, through the trustee UMB Bank, sued BMS for breach of the CVR's "diligent efforts" clause, arguing the company had not moved fast enough on the liso-cel filing. The suit sought the lost ~$6.4 billion; it was dismissed in 2024.5

The yardstick

Revenue trajectory is not deal quality. On this deal, every single-metric verdict is wrong in a different direction.

Grade it on Revlimid's revenue and you call it a disaster; grade it on goodwill and you find no error was ever booked; grade it on acquirer TSR and you measure everything except the deal. The honest reading is the two-grade one. The bet was defensible but expensive — a sound "buy cash flow to fund a transition" logic executed at a 53.7% premium for a cliff that was fully visible, financed with an expensive bridge. The outcome was genuinely middling: the pipeline it paid up for did reach the market, but the headline contingent payout was lost to a 36-day operational slip, and the core asset eroded on schedule. C+ and C+ — not because the deal was unremarkable, but because it refuses to be summarized by any one number.

Method

Every retrospective here reconstructs a deal from free primary documents — SEC filings, Drugs@FDA, FTC records, company disclosures — and assigns two grades: Bet (was it reasonable at the price?) and Outcome (did it pay?), each judged by the yardstick that actually fits the deal. Grades are the author's judgment, defensible but arguable; the underlying figures are sourced below.

Retrospective analysis of public information for educational purposes; not investment advice. Figures are rounded from public reporting — confirm exact values in the underlying filings before relying on them. Views are the author's own and not those of any employer.