Engineering the invisible

Metal-organic frameworks have won a Nobel Prize in Chemistry, and AI start-ups now raise rounds of $500 million to discover new materials. Abhiram Kannan of Bidra wished them well and expects them to find no buyer. A company that discovers a new drug can be acquired. A company that discovers a new material cannot. "If you don't deliver materials by the truckload, by the ton, by the kiloton, you're really not an acquirable business," Abhiram said. The founders at the table who make MOFs spent more time on factories and customers than on models.

Abhiram and Vincent Brillault of Calderion had asked whether MOFs are living through the moment AI had when large language models arrived. Vincent's interest began with electrolyser catalysts, whose amorphous structure nobody could describe, so nobody could predict how they would behave. MOFs are crystalline. Their makers know exactly what is inside and can tune the properties finely. Some capture CO₂, and some can be made conductive. A carbon-capture founder whose company uses them called MOFs chemical filters, tuned at the molecular level to grab one gas.

They have been around for years, yet they are rarely seen at scale outside academia. Abhiram blamed the gap between the lab and the plant. No MOF occurs in nature, so every one is synthesised, and academics over-engineer the synthesis to prove a material works. Industry wants the simplest, cheapest and most repeatable process it can find.

AI has been here before

People who had worked with MOFs saw less that was new. One noted that CuspAI's only commercial project shared so far is MOFs for water purification, and asked how interesting water filtration is to venture investors. Machine learning on these materials is not new, the carbon-capture founder said, and IBM has been producing lists of candidates for a decade. The models miss feedback from the field. A model of adsorption for one gas struggles once ten contaminants compete. A new material must also be made on existing equipment at a competitive price, and as a drop-in replacement it has to match the incumbent on cost or perform two to four times better.

Abhiram listed the bottlenecks. Cost comes first, since a 10–20% gain in performance rarely justifies a premium. Cycle life is second. Bankable projects need proof of tens of thousands of cycles, which is hard to get without an industrial installation. Third is the choice of application. Carbon capture, water harvesting, hydrogen storage and olefin–paraffin separation are so price-sensitive that a MOF has to sit at the bottom of the cost curve. Asked why a crystal of metal and carbon should be costly at all, Abhiram said things discovered in a computer rarely translate to the real world, and the practical learning comes only from trying 10,000 things.

Hydrogen shows the trap. One of the hosts put the market at about 100 million tonnes and €200 billion a year, with 93% used where it is made. Production, compression, and transport and distribution each account for about a third of the delivered cost, so moving hydrogen other than by pipeline at least triples it. The host's team had looked at MOFs for hydrogen storage and could not make the case work without free energy.

If you don't deliver materials by the truckload, by the ton, by the kiloton, you're really not an acquirable business.

Where nothing else works

Abhiram suggested looking where incumbents are weak or absent. A future with more nuclear power will need to separate hydrogen isotopes. Chipmakers need xenon and krypton, present in air at parts per billion, and during the AI build-out no cost is too high to recover more from air separation units. Defence forces want textiles that protect soldiers from toxic gases, an application for which NuMat has won an Air Force grant. Vincent added that the distillation columns for krypton–xenon separation stand 80 metres high.

A participant added separating rare earths of similar weight and slow-release pesticides, but saw no killer idea yet. Chemistry driven directly by light, which could make hydrogen without an electrolyser, came up as a long shot. It is thermodynamically possible, but nobody has solved the kinetics.

The carbon-capture founder would not accept that cost is the barrier. "I think it's unfair to say that the cost of MOFs are prohibitive," they said. BASF has made material at $20–30 a kilo, and in the founder's systems it accounts for a single-digit share of operating costs. The harder work now is scaling the process technology around it. The founder also warned that niche applications may be too small to return the 10x a venture fund expects. Abhiram agreed there is a balance to strike. A billion-dollar niche might open a larger market, but capturing the value takes a company that engineers the whole system.

A founder who scales nanomaterials from a gram to a tonne offered the most down-to-earth market. Catalysts care less about material price, because the catalyst's cost disappears into the processing cost. Some of the founder's customers buy MOFs to burn them, as precursors for catalysts of finely controlled composition. The founder admitted it is less noble than the fancy applications, but it is a practical one. They track demand by order size, from 10 grams to 100 grams, then a kilo, then five. By Abhiram's test, that is the direction that counts.

This Ripple was hosted by Vincent Brillault (Calderion) and Abhiram Kannan (Bidra, OCP Venture Capital) at The Drop 2026 on 16 September.