Lola Vision Systems says manually preparing an AI model for new hardware can take roughly 200 hours before testing begins. Founded in 2024 by Tayo Adesanya, the Washington, D.C. company is developing software and semiconductors for running AI models on devices. The software translates a customer's code and model into instructions that its chip can execute.
What slows model testing on new chips?
Adesanya told TechCrunch that the software layer connecting an AI model to a chip is a major bottleneck. Teams can spend roughly 200 hours configuring a model on unfamiliar hardware before they can start testing it. That is setup time, rather than a measure of model performance or a reported reduction achieved by Lola Vision Systems.
The company says it has rebuilt this software layer to automate more of the process. A client supplies its code and the AI model it wants to use, which may be custom-built or open source. Lola Vision Systems' software then translates both into instructions the client's chip can run.
How did the company reach this approach?
Adesanya's work with microchips and AI processors began almost 12 years before he founded the company. He mainly helped large manufacturers decide which chips to use in their hardware. He told TechCrunch that this work gave him an early view of demand in AI computing, and that his company was a bet on what he saw ahead.
Lola Vision Systems is also developing its own semiconductor chips, although the source does not give a release timetable or technical specifications. The company says it aims to automate more of the path from model and code to chip instructions. It has also partnered with SCALE, a microelectronics workforce development programme, to work with more semiconductor labs.
Where could the software fit?
The company is one of several startups seeking alternatives to Nvidia technology for running AI on devices. Adesanya said many companies begin with Nvidia Jetson, a line of compact computing modules, or open source models. He said these options can require teams to spend days or weeks getting a system running, followed by further debugging before models become usable.
Adesanya also said some device deployments run into power limits or insufficient computing capacity for medium or large models. He described possible consequences as recognition systems missing targets or misidentifying objects. Edge computing means processing AI directly on a device, such as a camera or drone, instead of sending work to a remote data centre.
What evidence is available so far?
According to Lola Vision Systems, a dozen corporate customers have signed letters expressing interest in buying its chips when they become available. The company also says it has one signed customer, but the source does not identify that customer or disclose commercial terms. Interest letters indicate stated interest; they do not establish that chip sales have taken place.
Adesanya said aerospace and other mission-critical businesses need accuracy and reliability because those factors can affect regulatory review and operation in the field. The source does not provide independent performance results, power measurements, or customer deployments. It also does not specify which models or chips have been tested through the software.
Lola Vision Systems says it will license its software on existing hardware to bring revenue in sooner. The available material does not state the licensing price, launch date, or customer terms. The next points to watch are software availability, evidence from deployments, and the timing of its own chips.