Apoha Raises $36M to Build a Data Layer for Physical AI

Apoha’s $36M funding round gives the London deeptech company a chance to turn Liquid State Intelligence into a commercial data layer for physical AI: tiny liquid-interface measurements that record how drugs, foods and materials behave under stress. The Series A was led by Singular, with Draper Associates, Redalpine, Seedcamp, Wilbe, Nucleus and Innovate UK backing also disclosed.

The cheque lands as AI for science runs into a data problem that software alone cannot fix. Apoha has to generate its dataset in the lab, then convince pharma, food and materials customers that those measurements can move product decisions earlier.

A New Round for Scarce Physical Data

The company came out of stealth on June 3, 2026, at SXSW London with a pitch that sits between biotech tools, materials discovery and AI infrastructure. Its claim is simple enough to grasp and hard enough to prove: companies already know a molecule’s sequence and can often model its structure, but they still struggle to measure how it behaves once pressure, temperature, liquid interfaces, formulation changes and time get involved.

Companies House, the UK corporate registry, shows the corporate trail around the financing. Apoha Limited filed a share allotment dated March 25, 2026, and Raffi Kamber, co-founder and general partner at Singular, was appointed as a director on the same date in the Companies House filing history. The same registry lists Apoha as a private limited company incorporated on April 19, 2021, with research and experimental development on biotechnology as its business activity.

For a startup emerging from stealth, that mix matters. Venture money pays for scale, but the public grant record shows an effort to turn the platform into equipment that scientists outside Apoha can use.

A Drop Becomes the Dataset

Apoha calls its measurement layer Liquid State Intelligence. On Apoha’s science page for molecular states, the company sets the idea against two older biological data classes: sequence, which identifies a molecule, and structure, which describes its shape. The new layer is behaviour, the response of matter when it is pushed away from rest.

The first commercial product is Variations in Interfacial Behaviour under Excitation (VIBE, a readout that turns wave responses at a liquid interface into molecular behaviour data). A sample is dropped onto a prepared liquid surface, the contact perturbs the interface, and an imaging system captures the emissions that move across it. The company says the process is label-free and can run with as little as 10 micrograms of material.

Shamit Shrivastava, Apoha’s co-founder and chief executive, traces the physics to work on how interfaces carry information. In Shrivastava’s founder essay on matter, he describes contact senses such as touch, taste and smell as data channels built from matter meeting matter. Apoha’s lab version of that idea is controlled, optical and machine-readable.

The platform’s commercial promise depends on a difficult sequence. First, the hardware has to produce repeatable waveforms. Then the software has to turn those signals into descriptors that customers can compare. After that, the models have to connect those descriptors to failures that would otherwise appear later in development.

Why Antibodies Came First

The earliest paid use case sits in antibody developability. A UK Research and Innovation project record for Apoha describes an active Innovate UK project to develop a site-deployable Liquid Brain prototype for high-throughput antibody screening. UK Research and Innovation (UKRI, Britain’s public research and innovation funding body) puts the project cost at £499,750, with the funded period running from May 2025 to August 2026.

The same record states that poor developability is responsible for one third of clinical-stage failures of biologics, with a cost of about $450 million per biologic developed. It defines the early screening problem in practical terms: thousands of candidates can be screened for target binding, while viscosity, hydrophobicity, self-aggregation and other properties remain expensive and material-hungry to test at the same stage.

Apoha’s Boehringer Ingelheim antibody case study gives the clearest company-published evidence so far: 71 clinical monoclonal antibodies, 8 micrograms per replicate, three replicates per antibody, six true positives and zero false positives. It also lists 12 false negatives for the single VIBE1 feature, with VIBE2 through VIBE5 still in validation.

The caveat sits in the data itself. A conservative early warning tool can save work when it flags a doomed candidate, but missed liabilities still have to be caught by other screens until more features are validated.

Customers Are Testing the Same Signal Elsewhere

Pharma gives Apoha a demanding first market because failures are visible in clinical, formulation and manufacturing records. The company’s disclosed customer list points to other markets where a product can pass a chemistry check and still disappoint in use.

• In mRNA delivery, Apoha says it is working with Ethris on the link between in-vitro tests and animal behaviour for lipid nanoparticles carrying messenger RNA.
• In food, the company says THIS, the plant-based food group, used the technology to find a protein replacement for a product headed to supermarket shelves.
• In materials and ingredients, Apoha’s public materials describe work across formulations where behaviour under stress affects stability, texture, performance and shelf life.

Those markets have different buying habits. Pharma customers will ask for validation against internal datasets and regulatory expectations. Food companies care about speed, cost and sensory match. Materials customers may need a readout that survives heat, shear, storage and supplier changes. The common thread is that each buyer wants a measurement early enough to change a formulation decision.

This is where Apoha’s pitch moves beyond a single instrument sale. If the same measurement class can travel across antibodies, proteins, lipid particles and industrial formulations, the company gets more examples for its models and more chances to prove the readout against products already moving through customer pipelines.

The AlphaFold Comparison Ends at Behaviour

AlphaFold gives readers a useful benchmark for Apoha’s ambition. The AlphaFold Protein Structure Database provides open access to over 200 million protein structure predictions. That database grew because the field already had decades of experimentally determined structures to train on and validate against.

Apoha is working on a younger dataset. Its measurements concern state and behaviour, areas where public, standardized corpora are far thinner. The distinction is easiest to see side by side.

That comparison explains the investor logic without flattering the risk. Apoha has to sell the measurement, the instrument, the model and the workflow integration in the same customer conversation. A software demo can spread quickly. A lab data business spreads through reproducibility.

A Hardware Company With Software Economics to Prove

The funding will be used to scale Liquid State Intelligence across biologics, food, materials and physical-world AI. That means more hardware, more sample types and more models trained on measurements the company controls. It also means slower proof than a pure software startup would face.

The business case depends on a compounding loop in which each measured sample gives the model another labelled example. The lab risk is just as concrete. Instruments drift. Sample preparation varies. Pharma customers ask for validation across their own molecules. Food and materials buyers care about combinations of texture, taste, durability and storage that rarely fit into a single metric.

Apoha has raised enough money to move from stealth science to commercial scrutiny. The proof now has to survive customer after customer, droplet by droplet.