Hyunjun Park spent nearly a decade teaching machines to write in DNA. In October 2025, he stepped down as chief executive of Catalog Technologies, the Boston startup he co-founded out of MIT in 2016.
Catalog raised $54.3 million along the way. It built the first industrial-scale machine for writing digital data into synthetic DNA, and encoded 200,000 words from eight Shakespeare tragedies into molecules it could search by query. What it never landed was a paying data center. The company's last press release ran in August 2025. Weeks later, Park was gone.
In March 2026 the assets found a buyer. Biomemory, a Paris-based competitor founded in 2021, acquired Catalog's technology, patents, equipment and core engineers. It opened a Boston hub for research and production. The stated deadline: a first commercial DNA storage solution ships in the second half of this year.
That window is open now. The pitch is no longer a laboratory demo.
It is a rack.
The physics case is settled: extreme density, zero power at rest, decades of guaranteed readability. The economic case is not: writing costs and access latency remain far from tape-class economics.
The investor signal to watch is narrow and testable. Either a named customer runs a named workload on a rackable DNA archive before this promise window closes, or the milestone slips into 2027 and the valuation story resets.
The mechanics have not changed since researchers first stored text in synthesized strands. Binary code maps onto the four chemical bases of DNA, adenine, cytosine, guanine and thymine. A synthesizer produces the strands, a vial or card holds them, and a sequencer reads them back decades later. What changed in 2026 is who controls the industrial process end to end, and what that control costs to buy.
From laboratory curiosity to rack-mountable
Biomemory's approach skips the slowest step in traditional synthesis. Instead of building strands base by base, one nucleotide at a time, the company assembles pre-manufactured DNA blocks using enzymes. Olivier Lauvray heads industrialization at the company. His claim: the block method writes faster and cheaper because the expensive chemistry happens upstream, in bulk, before any writing machine runs.
Retention floor on the DNA Card
Three retention tiers of 50, 100 or 150 years, an uncorrectable bit error rate beyond 10E-16, plus lifetime readability assurance · Biomemory, 2026
Retention is the quiet advantage. A hard drive lasts five years. An archive tape demands periodic migration every generation of hardware. Synthetic DNA claims half a century minimum in Biomemory's card, drawing no power while it waits. Its first cards went on sale in December 2023 at $1,000 per kilobyte. A souvenir price. It tells you how early this market is and how much cost curve remains.
What the Catalog acquisition actually buys
Per Lauvray, integrating the acquired kit takes roughly a year. Getting printers and cartridges into enterprise-ready production takes two to three more. Biomemory had planned to operate its own storage facilities as a service from next year; the deal pulled that schedule forward by about six months.
Confirmation criteria: a rackable write-store-read appliance, running in at least one facility it does not own, by end of 2027.
"Biomemory is the first company in the DNA data storage industry to have implemented a complete end-to-end industrialized process. It spans the digital file or object ingestion to DNA writing and storage with a data retention of 50 to 150 years."— Olivier Lauvray, VP of industrialization, Biomemory
The claim that matters commercially arrived through the Scality partnership in June. Scality builds the software layer large organizations use to manage archive data across flash, disk and tape tiers. Wiring DNA into that stack positions the molecule as one tier among several, called in by policy when data must survive decades. Integration beats replacement in enterprise sales. Always has.
The cold storage math starts to move
The demand side of this story is not speculative. Research firm Gartner projected that the world would need roughly 12 million petabytes of storage capacity by 2030, then revised the figure upward by an additional 20 million petabytes. IEEE Spectrum, surveying those numbers, concluded the world is not on track to manufacture enough conventional media to fill the gap. Factories, not just disks, are the bottleneck.
Gartner's 2030 storage capacity need
A 20 million petabyte upward revision against earlier projections, driven by data creation outrunning manufacturing capacity · Gartner via IEEE Spectrum, 2024
Most of the world's archive already sits on magnetic tape. Tape keeps winning tenders because nothing matches its cost per petabyte. Against that benchmark, Lauvray projects roughly 10 times lower capital expenditure for a DNA estate of about 100 petabytes. He adds claims of a 20 times reduction in ten-year total cost of ownership and a tenth of the physical footprint. Treat those as vendor numbers until an operator confirms them. The direction is plausible. The magnitude is unproven.
The energy logic is harder to argue with. Data centers absorb around 2% of global electricity, by Biomemory's own accounting. Archive copies sit in those halls drawing power for cooling and redundancy, waiting for requests that rarely come. Molecular media idles at effectively zero. We flagged that pressure point in another storage layer: as we wrote in August, Noon Energy booked 1.1 gigawatt-hours of ultra-long-duration battery storage in four months on AI-driven demand. Every tier of the storage stack is being repriced at once.
Institutions have noticed. The Library of Congress began a DNA storage trial. The DNA Data Storage Alliance presented that detail at its 2026 meeting there. The alliance counts more than 40 members, split evenly between academia and industry. Standards work on codecs and interoperability is underway. Archives move slowly, then all at once.
The skeptics have real numbers too
The Storage Networking Industry Association published a technology review of DNA data storage in 2025 and expressed doubt that commercialization arrives before the decade turns. Writing is the reason. Reading DNA got cheap and fast on the back of genomics; writing it at archive scale did not, and latency rules out anything beyond cold tiers regardless of cost.
Holographic storage compounds the problem. Competing archival formats already demonstrate terabyte-capacity media and aim for mass production around 2027, which means DNA will not enter the market uncontested. The competition is two-sided: entrenched tape below, maturing holographic formats above. DNA promises densities orders of magnitude beyond both. So far it ships kilobytes.
Atlas Data Storage carries $155 million in initial funding from its spinout out of Twist Bioscience's storage program. In December 2025 it launched Eon 100, billed as the first scalable DNA storage service. In April, Taiwan Semiconductor Manufacturing Company (TSMC) selected Atlas to present its chip-enabled writing platform at the chipmaker's annual technology symposium. Yet its appointed chief executive departed around September 2025. No replacement is listed today. Two flagship companies, two leadership vacuums inside twelve months. That pattern suggests capital is still discovering how long the road to revenue really is.
Does DNA earn a rack by 2030?
Probability: 65% — Biomemory publicly targets appliance deployments in external facilities for 2030-2031, the Scality integration path already exists, and institutional pull is documented. The discount: no vendor has published write-cost curves at petabyte scale.
✅ Arguments for
Enterprise integration runs through Scality's installed object-storage base, so DNA lands as policy-driven tier, not rip-and-replace.
Capacity forecasts leave room for a new medium; conventional manufacturing cannot close the 2030 gap alone.
Confirmation criteria: first external-facility pilot converts to paid production workload before mid-2028.
❌ Arguments against
SNIA's own 2025 review doubted commercial viability before 2030, and nothing public since has overturned the write-cost math.
Leadership churn at both flagship vendors suggests timelines slip more often than they hold.
Disconfirmation criteria: the H2 2026 launch misses, or ships without a named external customer.
Whether Biomemory names and prices a commercial system before December 31, 2026
The first enterprise archive deployed through the Scality partnership, with a disclosed customer
Atlas and imec publishing write-throughput or cost-per-gigabyte figures for their chip-enabled platform
The next SNIA technology review, and whether it revises the pre-2030 commercialization doubt
Development scenarios
🟢 Optimistic scenario (25%)
Implications: DNA storage becomes a fundable category again; expect a financing round at materially higher marks than the $23.2 million Biomemory has raised to date.
🟡 Base-case scenario (55%)
Implications: patient capital wins; the inflection asset is the first audited total-cost comparison against a live tape estate.
🔴 Pessimistic scenario (20%)
Implications: a second consolidation wave follows, and DNA storage retreats to niche preservation contracts like the Library of Congress trial.
