Anthropic–Decart: A Compute Deal With Consumer-Grade Consequences
The Anthropic Decart acquisition is a prospective deal in which Anthropic aims to buy AI startup Decart to gain compute efficiency software and real-time video AI capabilities, reshaping how large models are trained and how consumers may interact with visual retail assistants in everyday shopping and support experiences.
Anthropic is reportedly in talks to acquire Decart for about $6 billion, in what would be its largest acquisition so far if it closes. This is not a vanity purchase. It is a blunt admission that AI compute efficiency has become as strategic as the models themselves. Decart’s software reduces the cost of training models by helping chips work more efficiently, which could help Anthropic’s infrastructure absorb far more demand without a matching surge in hardware spending. In a market where leading labs have committed to spend tens, if not hundreds, of billions of dollars on data centres filled with costly chips, owning such efficiency tools is a competitive survival strategy, not a side project.
Why Compute Efficiency Is the New Frontier in the AI Race
Anthropic has been spending heavily on computing power to develop new products and serve customers, and those costs are mounting as it grows. The same is true for its biggest rivals, who have committed eye-watering budgets to data centres and specialised chips to train and run their largest models. Those expenses may weigh on them as they approach public markets, which are far less patient with open-ended infrastructure burn.
Against that backdrop, buying Decart reads like a bet that owning the software that squeezes more performance from every chip is better than endlessly renting more compute. Decart’s tools are designed to make training and inference more efficient, lowering costs and expanding capacity at the same time. That is a direct answer to competitive pressure in AI infrastructure, where every major lab is racing to reduce computational costs before those costs define their valuations. If Anthropic can train and serve models more cheaply than peers, it gains pricing power, faster iteration cycles, and the freedom to experiment with more compute-hungry features like real-time video AI.
From Chat to Camera: Real-Time Video AI as Anthropic’s Next Edge
Where this acquisition becomes strategically interesting is Decart’s work in real-time video AI, not just its infrastructure smarts. Decart focuses on generative video, using so-called world models capable of modifying live video feeds instantly. Its Lucy model can transform live video in real time, with use cases ranging from virtual try-on and product placement to gaming and advertising. On its own website, the company says it builds “the infrastructure and models that make AI run at the speed of reality.”
That philosophy aligns cleanly with Anthropic’s conversational strengths. Until now, most deployments of its models in retail have centred on text: search, recommendations, order questions, and returns. These help, but they do not solve the core problem in many shopping journeys: customers need to see, not just read. Folding real-time video AI into Anthropic’s stack would let it move beyond chatbots into experiences where language and live visuals work together in a single, continuous interaction.
What Real-Time Retail AI Could Look Like for Shoppers
The most obvious near-term impact for ordinary users shows up in how they shop. Decart’s virtual try-on tools already allow shoppers to use a live camera feed to see clothing and accessories on their body, without relying solely on a static uploaded photo. Instead of guessing from a product page, a shopper could stand in front of their phone and see garments adapt to their movement and body shape during a live session.
Layer a conversational Claude-style assistant on top and the experience changes again. A customer could describe an upcoming event, share a budget and style preferences, then watch as recommended items appear on their live image while the AI explains fit, fabrics, and delivery options. Beyond fashion, beauty brands could let customers test makeup shades, eyewear retailers could offer live frame comparisons, furniture sellers could show how a piece might look in a home, and consumer electronics brands could guide setup or identify the right cable by analysing video of the user’s space. For shoppers, this is less about novelty and more about reducing uncertainty, boosting confidence, and cutting down on mistaken purchases and returns.
Trust, Privacy, and the Real Stakes of Anthropic’s Move
Real-time retail AI will only matter if people trust it. A virtual try-on that quietly stretches proportions, smooths fabric, or misrepresents colour to flatter a product will do more harm than good. The risk is that real-time video becomes yet another glossy filter that widens the gap between expectation and reality, instead of narrowing it. If that happens, conversion may rise in the short term, but returns and disappointment will follow close behind.
Privacy is the other fault line. Live-camera shopping and visual support raise fresh questions about consent, storage of images, biometric data, and how footage might be reused or combined with other profiles. Retailers will need clear policies, easy opt-outs, and accessible non-visual alternatives so camera-based journeys enhance rather than replace existing options. For Anthropic, this deal is a bet that owning AI compute efficiency and real-time video AI will define the next era of retail AI applications, from virtual try-ons to visual customer service. The companies have not framed the talks as a retail strategy, and the acquisition is not yet final, but the direction of travel is obvious: the next big AI battleground will be who can make powerful models feel instant, trustworthy, and useful in the most ordinary, everyday interactions.
