While data centers drown in electricity, archivists are hitting a different wall: tapes fill up, disks rot, everything demands space and power. AtlasBase proposes to untie this knot at the molecular level; its new Thalia platform encodes digital information into synthetic DNA strands built from adenine, cytosine, guanine and thymine bases. As Blocks & Files (blocksandfiles.com) reports, the starting point is 100 terabytes in an LTO-cartridge footprint, with the roadmap reaching 5 exabytes in the same space, equal to compressed LTO-10 capacity.
The heart of the hardware is a biochemical reaction surface mounted on a standard 16-nanometer CMOS layer. The chip's 5.6 billion electrochemical synthesis sites decide each cycle, via electrical signals, which base gets added; a software codec translates digital data into a four-letter DNA blueprint while polymerase chain reaction produces cheap copies. Per All About Circuits (allaboutcircuits.com), Thalia is a 5.6-billion-well design aimed at breaking the power and density bottlenecks of AI data centers.
The scale figures are audacious: the architecture promises a 700-fold increase over older synthesis technologies, with the potential to synthesize in a single day the synthetic-DNA volume the whole industry once produced in a year. In Quantum Zeitgeist's (quantumzeitgeist.com) analysis, the platform compresses petabyte-scale production into 24 hours, talking of a day's output equal to a year's DNA production. If that pace materializes, DNA storage graduates from lab curiosity to industrial feedstock.
The storage format is equally unfamiliar: synthesized strands are purged of liquid, dehydrated, then hermetically sealed into stainless-steel capsules. The cartridge design holds 100 capsules in a 10-by-10 array, with per-capsule capacity starting around 1 terabyte and rising toward 50 petabytes on the roadmap. In GenomeWeb's (genomeweb.com) product note, Thalia is listed as a CMOS DNA-synthesis chip carrying 5.6 billion synthesis sites.
The density metrics take the breath away: 16.8 billion devices with a synthesis rate of 250 gigabytes per chipset per 24 hours. But Blocks & Files attaches a critical caveat in the same article: whether the chip exists beyond a design concept is unknown, and the question was put to the company. In hardware journalism, the gap between an unverified slide and working silicon matters more than the rest of the file.
Context makes the claim meaningful: the LTO roadmap stretches to 913 terabytes compressed at LTO-14, so Thalia's 5-exabyte target passes tape's ten-year horizon today. Back in March 2026 Atlas Data Storage and imec joined forces; per imec's press release (imec-int.com), the Belgium-California partnership will scale permanent DNA storage. The target customer is clear: critical archives kept for decades, from legal records to scientific data and cultural heritage.
The realistic picture is this: read-write speed, drive cost and availability figures are still missing, and as TechRadar's (techradar.com) original file notes, speed and price remain undisclosed. DNA may become the king of cold archives but cannot approach hot data. Still, with storage appetite growing endlessly in the AI age, writing to molecules is no longer science fiction; it is an engineering claim demanding scrutiny.
| Topic | Why it matters |
|---|---|
| Thalia bio-silicon platform | A CMOS chip writing to DNA with 5.6 billion synthesis sites. |
| Road from 100 TB to 5 EB | Scale targeting a decade past the tape roadmap. |
| Unverified hardware | Without working silicon a roadmap counts as a wish. |
AI commentary
"What caught me in this file is the audacity of the scale claim: 5 exabytes per cartridge leapfrogs a decade of tape roadmaps in one stroke. Such claims oblige questions, so I wrote the doubts about whether the chip exists beyond concept right into the piece."
AI assessment
The hardest question is the burden of proof: 700-fold, 250 gigabytes in 24 hours, 5 exabytes per cartridge; none of it verified in an independent lab. The Blocks & Files chip caveat may be the file's most honest sentence: without working silicon, a roadmap is a wish. The startup's track record of shipped products and customers is also absent; first-time readers owe themselves skepticism.
The second limit is physics: however fast DNA writing gets, reading still means sequencing, with latencies possibly slower than tape. Random access, rewrite and error-correction costs pass through the file's margins but not its headlines. Acceptable for cold archives, a weakness in any general-storage claim.
The sourcing leans one way: every figure comes from the company or outlets quoting it; the imec partnership is genuine validation but not a technical audit. TechRadar's speed-and-price questions stand unanswered. That does not mean the startup is lying, but after a decade of archive-technology disappointments every demo deserves distance.
The practical payoff still matters: storage cost and energy have become the hidden tax of the AI economy, and molecular density is a genuine way to cut it. The short-term lesson for buyers is to make the seller speak with lab evidence; the long-term one is to build an archive strategy tracking the tape-to-DNA transition. Even if Thalia fails, the direction is right.
Sources
6 links; no other published story cites them. Stories sharing a link do not confirm each other; a source's origin is not inferred from how often it is cited.
- @techradar.com TechRadar — 100TB DNA capsules storage story
- @blocksandfiles.com Blocks and Files — AtlasBase Thalia chip
- @quantumzeitgeist.com Quantum Zeitgeist — AtlasBase synthetic DNA scale
- @allaboutcircuits.com All About Circuits — AtlasBase bio-silicon debut
- @genomeweb.com GenomeWeb — AtlasBase Thalia product note
- @imec-int.com imec — DNA-based storage press release
dna storage · atlasbase · thalia · archives · data center