Osmo is giving computers a sense of smell — and already selling AI-designed fragrances at Target

Osmo is giving computers a sense of smell — and already selling AI-designed fragrances at Target

Alex Wiltschko has spent twenty years on a single problem: teaching computers to smell. He did his PhD in olfactory neuroscience at Harvard under a lineage that runs directly to Richard Axel, who won the Nobel Prize for discovering smell receptors. He then ran the digital olfaction team at Google Brain for six years before spinning Osmo out through Lux Capital and GV. The company's $70M Series B, led by Two Sigma Ventures, funded the construction of a fragrance manufacturing facility in New Jersey.

“We raised our Series B, putting an additional $70M in the bank with Two Sigma leading and Lux. We have a robot the size of a school bus that makes a new fragrance every 100 seconds. We have 5,000,000 sniffs digitized, over a quarter million physical samples created, and have digitized about 6,000,000,000 fragrance molecules. You can go into Target and buy a product that has our fragrance in it today.”

Why smell is hard to digitize

The core challenge is the absence of a map. Color has three channels — RGB — which is why cameras, screens, and CMOS sensors all work. Sound maps cleanly onto frequency. Smell has more than 300 channels of receptor information, and no equivalent organizing structure existed until AI made it tractable. Wiltschko argues the field genuinely had to wait for modern machine learning before a 300-dimensional map of scent could be extracted from raw chemical data.

Osmo built that dataset from scratch, because there is effectively nothing on the open internet — the fragrance industry has kept its data proprietary. The company has digitized 5 million sniffs, created over 250,000 physical samples, and digitized roughly 6 billion fragrance molecules. Mass spectrometers run 24/7, and the company maintains sensory panels domestically and abroad, shipping crates of samples to human evaluators.

The business model

Osmo designs and manufactures fragrances for consumer brands. A robot the size of a school bus produces a new fragrance every 100 seconds. What leaves the factory is either a steel drum of fragrance oil or, for brands that want it, a fully assembled bottled product. The company completed its transition from R&D to commercial operations last summer, has built out both a manufacturing and a sales organization, and already has product on shelves at Target.

Flavor is a natural adjacency — Wiltschko says Osmo's olfactory intelligence models work on flavor "surprisingly well," since roughly 90% of what people perceive as flavor is retronasal olfaction rather than taste. For now, the company is staying focused on fragrance.

Sensor miniaturization

Osmo's current portable sensor is the size of two shoe boxes. It can distinguish real from counterfeit sneakers — counterfeiters use cheaper glues that produce a different chemical signature — and can identify a shoe's factory of origin with 93% accuracy. Getting from two shoe boxes to one is a clear near-term path. Getting to AirPods-case scale requires engineering work Wiltschko says is visible. Getting a sensor into a phone requires breakthroughs he can't yet see clearly, though he notes human noses already prove the physics is possible.

Scaling dynamics

Wiltschko frames the company's development in S-curves. Data is the current constraint, not model size, which is why Osmo is running its manufacturing robot and mass spectrometers around the clock. The proprietary dataset is both the bottleneck and the moat — building it was hard, but that means no one else has it either.