Data Dump - Durable DNA

The idea to use DNA as a storage medium was first proposed in 1964 by the Soviet physicist Mikhail Samoilovich Neiman, and demonstrations since the 1980s have confirmed its viability. Its advocates say it makes for a remarkably efficient and durable solution. A single gramme of DNA could theoretically store up to 215 petabytes (PB, each equal to a million GB) of data for thousands of years.

Transforming bytes into nucleotide bases turns out to be surprisingly simple. "You take your digital data and map it onto the building blocks of DNA," says Thomas Heinis, a professor in data management at Imperial College London. DNA's four base letters, A, T, C and G, are converted to 01, 00, 11 and 10. "Then you synthesise a molecule – the actual physical representation of this – and store it for as long as you want."

A favourite line amongst DNA data storage researchers is that "you could store all of the data in the world in a teaspoon", says Heinis. In reality, though, it would be very difficult to locate the data you were after in this undifferentiated globule, he says.

Crucially, though, the storage requirements aren't energy intensive. "It's energy efficient, because if you store it in a reasonable place, you don't need to cool it at all," says Heinis.

Startups are now cropping up in the DNA storage space, and progress has been made in recent years in bringing down the cost of "reading" DNA, says Heinis. But overall cost remains a hurdle. "It is still far too expensive," he says, especially when it comes to synthesising the DNA. "On the 'write' side, we haven't seen a major breakthrough yet, so that really needs to happen," says Heinis. "Once it's cheap enough, everything else will fall into place."

Silica and DNA technologies are unlikely to replace conventional storage for everyday computing or AI workloads anytime soon – Tania Malik

While Heinis describes Kazansky’s memory crystals as a "direct competitor to DNA storage", where DNA might have the edge is that "we will always be able to read DNA", due to its wide-ranging medical applications. "With other technology, the question is how long the read device will be around," he says.

Heinis points out it's now increasingly difficult to read the likes of floppy disks – which launched in the 1970s but were pretty much obsolete by the early aughts. "There are companies offering data storage for more than 100 years. But which of these companies will still be around in 100 years' time?"

Of the tech giants, Microsoft has taken the keenest interest in experimenting with new kinds of data storage. In 2016, the company announced it had stored 200MB of data in DNA, including a database of seeds held in the Svalbard Global Seed Vault, as well as the Universal Declaration of Human Rights in more than 100 languages. In 2020, Microsoft and other companies founded the DNA Data Storage Alliance.

"The demand for long-term data storage in the cloud is reaching unprecedented levels, and we are reaching the limit of what's possible with existing storage technologies," a Microsoft spokesperson told the BBC.

Microsoft also sponsored Kazansky's research group at Southampton University as part of its Project Silica from 2017-2019. "We proved the core principle together, after which they continued developing the technology independently," says Kazansky.

In February 2026, Microsoft published a Nature article detailing a new achievement on this front. The company managed to store data in borosilicate glass, which is found in kitchen cookware and oven doors, in addition to the standard fused silica glass. Borosilicate glass is much cheaper – making the idea more financially tenable – while also being very durable. The company claims this data could be stored for up to 10,000 years.

Microsoft's spokesperson told the BBC that although its proof-of-concept tests have shown promise, it is not currently commercialising this research.

Rethinking computing 

Of course, solving the long-term data storage problem is only part of the solution to energy-guzzling data centres. Silica and DNA are "very attractive from a sustainability perspective", acknowledges Tania Malik, assistant professor at the School of Informatics and Cybersecurity at Technological University Dublin in Ireland. "However, these technologies are unlikely to replace conventional storage for everyday computing or AI workloads anytime soon."

Malik says there are more practical ways of addressing the problem of "hot data" energy consumption in the near term. "One important area is improving infrastructure efficiency, for example through more energy-efficient processors and advanced cooling techniques such as liquid cooling or free-air cooling," she says.

At the same time, she adds, there is "growing recognition that efficiency must also be addressed at the software and workload level, not just the infrastructure level".

"In high-performance and cloud computing, performance has traditionally been the dominant metric, but energy efficiency needs to be treated as equally important," Malik says. "This means designing algorithms and applications that are energy aware." It also means using the appropriate amount of computing power for the task in hand, she says. "Not every task needs the largest possible AI model or the fastest possible runtime."

But in the face of exponentially accumulating data, a different kind of radical rethink may also be required, says Malik. Do we really need all of the data that we produce? Increasingly, part of the solution, she says, "is being more intentional about what we choose to keep".

- Author: Laurie Clarke, BBC