HomeContact
Keep track of advancements in every frontier

TheTechno Optimist.

MIRACuspAI—

MIRA

by CuspAIGB

PilotStage 3 of 5

In use with industrial partners; scaling after the Series B.

Updated 20 Jul 2026Checked 25 Sep0 updates this week

Milestones

Next · Materials validated in partner labs
  1. Materials validated in partner labsNowCurrent milestone.

Most important, last 3 months

  • 20 Jul 2026

Current obstacles

  • From prediction to productPartners must still make, scale up and approve each designed material, which is slow and outside CuspAI's control.

Physics limits

  • Models inherit the errors of their training physicsMachine-learned simulators are trained on approximate quantum calculations, so binding energies can be off by amounts large enough to change the ranking.
  • Capturing CO2 has a thermodynamic minimumSeparating CO2 from air takes at least ~20 kJ per mole, less from flue gas. Better materials can approach that floor but never beat it.

How it works

3 parts
Racks of the Galileo supercomputer in Italy: models like MIRA run on computers like these, generating and checking candidate crystal structures
Racks of the Galileo supercomputer in Italy: models like MIRA run on computers like these, generating and checking candidate crystal structuresPhoto: CINECA · CC BY 2.0 (opens commons.wikimedia.org)
Generate

Ask for properties, get structures

A generative model proposes candidate materials, such as porous frameworks, aimed at a target property like capturing CO2.

Simulate

Physics checks the ideas

Fast machine-learned simulations, trained on quantum-chemistry data, estimate each candidate's stability and how strongly it binds CO2.

Filter

Partners test the best

Only top candidates go to partner labs to be made and measured, and the results feed back into the models.

Update log

1 update

Mon 20 Jul

  • Major: BlogFunding

About CuspAI

The team behind MIRA

CuspAI

AI materials design platform

CuspAI (Cambridge, UK), co-founded by Chad Edwards and Max Welling, uses generative AI and physics simulation to design new materials, starting with ones for carbon capture. It raised $450M in July 2026 at a $2.6B valuation.

  • $450M Series B (July 2026) led by Kleiner Perkins and NEA, with Bezos Expeditions.
  • Formed a chemical research group including NVIDIA and Samsung Electronics.
  • Its MIRA platform aims to cut materials design from years to months.

Company profile →

Founded
20242 yrs
Headquarters
United Kingdom
Status
Private
Valuation
$2.6BprivateJul 2026
Raised
$450M1 round
Last round
Series B · $450M20 Jul 2026$2.6B post
Works in
MaterialsAI materials discoveryAI
Coverage
1 program · 1 updatelatest 20 Jul 2026checked 25 Sep
Lead investors
Kleiner PerkinsNEA
Partners
NVIDIASamsung Electronics (opens samsung.com)
People
Chad EdwardsCo-founder & CEOMax Welling (opens en.wikipedia.org)Co-founder
Primer