이미지: 러버블
50 Million Projects and the "Latency" Wall
Lovable is a platform for building apps and websites through conversation with AI. Users describe what they want, and the AI generates a working prototype in real time, which can then be refined through simple feedback and deployed. Since launching in November 2024, the number of projects built on Lovable has surpassed 50 million.
The platform's core loop is "describe → see result → iterate." But when waiting time creeps in between each step, it breaks the flow of work. This "latency" is precisely what Lovable's infrastructure investment is meant to address.
How Cerebras's Wafer-Scale Engine Works
Typical AI systems spread a model's memory across dozens of separate chips. Every time chips exchange information with one another, it incurs a "communication cost," which is one source of response latency. Cerebras took a different approach. Its Wafer-Scale Engine places an entire model on a single, massive silicon wafer, eliminating inter-chip communication altogether.
Cerebras says this architecture is especially effective in environments with repeated back-and-forth exchanges, such as software creation, and that it accelerates the slowest part of the existing GPU pipeline. Andrew Feldman, co-founder and CEO of Cerebras, said, "When AI responds in real time, users do more, stay longer, and carry out more valuable work."
Phased Rollout Plan and Target Timeline
Lovable is aiming to substantially cut response times by 2027. Rather than an immediate, wholesale switch, the company has chosen a phased approach, first migrating the most latency-sensitive workloads to dedicated Cerebras capacity. Progress on each phase of the transition will be disclosed sequentially.
Anton Osika, co-founder and CEO of Lovable, said, "We built Lovable on the idea that the person closest to a problem should be able to solve it themselves," adding that "Cerebras is helping Lovable respond as fast as our customers think."
Pricing, Support Scope, and Limitations
The current source material does not mention specific pricing changes, billing structure, or the scope of supported regions or models related to this infrastructure transition. No timeline has been disclosed for when the full migration to dedicated Cerebras capacity will be completed or for each phase's schedule. The company said further progress updates will be shared at a later date.
