Follow
Subscribe via Email!

Enter your email address to subscribe to this platform and receive notifications of new posts by email.

Osmo Builds AI Smell Model Amid Growing Infrastructure Pressures

Osmo has developed a foundation model for smell in New Jersey to give computers olfactory capabilities, as broader technology sectors confront severe energy drops and network disruptions.
A stylized sculptural human nose set against a dark background with subtle botanical textures and floral patterns.

Osmo has built an AI smell model in a New Jersey laboratory to turn physical odors into digital maps, as broader technology sectors face mounting strains from power shortfalls and network outages [1]. The startup claims its system has reached human-level olfactory intelligence by pairing chemical analysis with descriptive sensory tags. While Osmo pursues commercial scent formulation, energy data shows hydroelectric generation falling in dry regions.

Osmo Develops AI Smell Model

Researchers at the company’s New Jersey facility spend their days testing scents to catalog every distinct smell on Earth. Founder Alex Wiltschko said the main goal of the venture is “to give computers a sense of smell.” Laboratory teams record sensory notes such as “nutty,” “peachy,” and “marine,” then match those observations against chemical breakdowns of each sample [1].

Teaching software to perceive scent involves solving problems that don’t exist in computer vision or speech recognition, because smells depend on airborne molecules interacting with biological receptors inside the human nose. In the latest edition of The Download, reporting showed how Osmo built its AI smell model by matching molecular structures directly to sensory descriptions, creating a digital map where chemical compounds correspond to predictable scents [1]. Wiltschko claims the system has already attained “human-level olfactory intelligence,” allowing computers to recognize complex mixtures that previously required trained human evaluators. It treats volatile organic compounds like language tokens. By organizing odors on a mathematical coordinate plane, the software can predict how an untested molecular structure will smell before chemists synthesize it in a laboratory.

Commercial Scope of the AI Smell Model

Fragrance companies have already started working with Osmo to create new scents for consumer products. Beyond perfumes, the startup plans to use the system for disease detection and counterfeit product verification, where faint chemical traces reveal fake goods [1]. Reporter Amit Katwala noted that while commercial partnerships are expanding, questions remain about how far automated olfaction can go.

The challenge of refining an AI smell model comes from the complexity of real air, where temperature, moisture, and competing odors alter molecular behavior before sensors can capture a sample. Human noses adapt to background smells within seconds, but digital sniffers often struggle to separate target compounds from surrounding air pollution and household chemicals [1]. Not everyone in the industry is convinced the startup can deliver on its broader medical promises. Identifying a pure chemical in a quiet New Jersey lab is fundamentally different from detecting early disease markers in human breath, where target molecules appear in tiny concentrations mixed with moisture and organic compounds. Independent researchers caution that commercial fragrance formulation can’t guarantee success in clinical diagnosis.

Osmo maintains that a general foundation model can overcome these hurdles by learning chemical principles instead of memorizing individual formulas. If the software succeeds, it will give computers a practical way to monitor physical environments [1]. Yet verifying whether it works outside perfume labs will take years of testing in field conditions.

Hydropower Declines in the Southwest

While software firms explore digital smell, basic energy systems powering the tech industry face sharp declines from weather stress. New figures from the US Energy Information Administration (EIA) revealed that the American Southwest generated 28% less electricity via hydropower between April and August compared to its 10-year historical average [1]. Reporting by Casey Crownhart in MIT Technology Review showed that severe drought has left water levels dangerously low in the region’s two largest reservoirs, curbing output just as summer computing demands surged. Most American hydropower plants depend on river dams to hold back water in deep reservoirs before channeling it through electricity turbines. When persistent drought starves those reservoirs of seasonal runoff, power output falls steadily across multiple southwestern states.

Climate change has intensified the problem by shrinking winter mountain snowpacks and accelerating evaporation from open water surfaces. As reservoir levels fall toward dead pool thresholds where water can no longer spin generator turbines, hydroelectric facilities lose their ability to provide steady base power during heatwaves [1]. Utilities must then search for backup electricity.

Osmo laboratory equipment used to analyze chemical odor data for an AI smell model
Osmo researchers analyze chemical odor data in New Jersey to train an olfactory foundation model. (Credit: MIT Technology Review)

Losing more than a quarter of regular hydropower output strains energy grids that also feed data centers and electric transit. When renewable hydro drops, power companies often turn to gas plants, raising carbon emissions and operating costs [1]. The southwest drought shows how sustained environmental stress threatens the physical foundation of modern computing infrastructure.

Solid-State Transformers for Power Grids

Rising electricity demand from artificial intelligence clusters, electric vehicles, and factories has exposed vulnerabilities in American power transmission lines. Network operators rely on heavy transformers to step voltages up and down over long distances, yet the basic hardware design hasn’t changed much in about a century [1]. Today those conventional units are complicated to build and in short supply, leaving power companies waiting years for new shipments. To solve this bottleneck, engineers are building solid-state transformers that use semiconductors instead of heavy metal cores. These devices can cut power losses, manage shifting loads from solar farms, and handle two-way electricity flows between neighborhood batteries and regional substations. The technology is still in development, but it could become an important part of the future electrical grid.

A recent episode of the MIT Technology Review Narrated podcast, published on Spotify and Apple Podcasts, examined how solid-state conversion could help power the ongoing artificial intelligence expansion. Moving electricity along transmission corridors without heat loss is now a top priority for utilities building lines to supply new data centers [1]. Faster production would help relieve utility backlogs.

Smaller footprints also allow technicians to install solid-state equipment in cramped urban spaces where massive oil-filled transformers can’t fit. Built-in digital monitoring gives operators real-time data about voltage spikes, enabling automated switches before localized faults cascade into blackouts [1]. Better hardware design could prevent physical failures as computing demand continues climbing in major metropolitan centers.

Infrastructure Outages and Grid Pressures

Physical grid strains coincided with multiple digital disruptions this week, hitting both government operations and major consumer platforms. A classified communications network used by a United States government agency went completely down according to CNN, while a software update disrupted Apple Music, the App Store, and AppleCare as reported by the Associated Press (AP) [1]. Marco Rubio said the network failures appeared unrelated and were not hostile acts, but reporting from Vox and MIT Technology Review indicated that artificial intelligence tools are accelerating vulnerability discovery and online crimes. The incidents join a wave of outages that grounded commercial flights, locked retail customers out of bank accounts, and knocked out an NSA network. As systems grow more interconnected, software bugs and automated scanning tools create disruptions that shut down services quickly.

Commenting on the wave of disruptions, Donald Trump shrugged off the situation during remarks reported by NPR. “It’s life. When you’re big and powerful and you have all the money, things like that happen. It’s fine,” Trump said [1]. The casual dismissal comes as tech managers struggle to harden networks against automated attacks and power instabilities.

International policy struggles to manage these overlapping challenges. At the COP31 climate conference, only nine countries backed a voluntary pledge governing artificial intelligence emissions, while Nvidia doubled its revenue but halted shipments of H200 chips to China amid new export limits [1]. Elsewhere, San Diego moved to shut down over 500 Flock surveillance cameras, and the Cambridge Dictionary selected Human as its word of the year, capturing cultural unease as algorithmic automation accelerates across modern society.

Sources
  1. ONLINE NEWS Macaulay, T. (2026, November 18). The Download: AI learns to smell and hydropower struggles. MIT Technology Review. [Article Link]
Leave a Comment

Related Posts
Total
0
Share