Whatnot is a livestream shopping app where sellers go live on video to auction off collectibles, fashion, electronics, and increasingly, almost anything. Buyers bid in real time, products sell out mid-stream, and shows can start and end within minutes. It closed 2025 with over $8 billion in sales, more than double the year before, and now holds close to 60% of the live shopping market across North America and Europe.
The company just acquired Shaped, an AI startup whose core job was helping online platforms show the right items to the right people at speed. Shaped built its system to process behavior data and make relevance decisions in under 50 milliseconds. Its previous clients included Outdoorsy and QVC.
The problem Whatnot is trying to solve is different from what Amazon or eBay face. On a standard shopping site, the catalog is stable. A product listing from Tuesday is still there on Thursday. On Whatnot, a seller's entire inventory can be gone in twenty minutes. The software that suggests what to watch next needs to know that, and it needs to know it now, not after a nightly data update.
Whatnot's VP of Data and AI said the company spent six years cutting the delay between a buyer's action and a new recommendation from roughly a day down to a few minutes. The Shaped acquisition is meant to push that further toward real time. The platform currently processes over 500,000 hours of live video and millions of buyer interactions every week.
The wider trend here is that recommendation quality is becoming the main competitive battleground in online retail. Poshmark replaced its social-following feed with an algorithm-driven one in early 2025. eBay launched an AI shopping agent that proactively guides buyers rather than waiting for a search query. Both moves reflect the same pressure: buyers expect the platform to know what they want before they type anything.
Whatnot's position is that it has more behavioral data per user than almost anyone. Its buyers spend an average of 95 minutes a day in the app, with over 80% coming back month after month. That is significantly more daily engagement than YouTube or TikTok gets from a typical user. The challenge is converting all that attention into recommendations specific enough to keep each buyer watching the right show at the right moment.
Shaped's founder, Tullie Murrell, built recommendation systems at Meta before starting the company. He will now lead Whatnot's newly formed Applied AI Research group, alongside about a dozen engineers and researchers joining from Shaped.
For anyone selling through Whatnot, or watching it as a distribution channel, the practical implication is straightforward. Better recommendations mean more buyers finding your show without searching for it. The platform's fastest-growing categories in 2025 include beauty, up nearly 800% year on year, electronics up over 400%, and women's fashion up over 200%. Sellers in those categories are arriving on a platform where the discovery engine is about to get meaningfully sharper.
The broader live shopping market in the US is projected to reach roughly $68 billion this year, still tiny compared to China where the format already accounts for nearly a fifth of all online retail sales. The gap between those two figures is essentially an argument for every investment Whatnot is making right now. The company is not yet profitable, but it is building the infrastructure that would let the US market eventually look more like China's.