What if the phone in your pocket could help run AI? A Mumbai-based startup wants to cut AI computing costs by tapping into millions of smartphones.
Jash Jetly, co-founder and CEO of EVE NET, points out that when Google started in 1999, storing data was expensive, much like artificial intelligence (AI) computing is today. Yet Google made web search free. Imagine if it had charged one dollar for every search—most people would never have used it. That made Jash Jetly question why AI should be any different. AI is trained on public data and, in many cases, on users’ own data, yet people have to pay to access information that already belongs to them. This idea became the foundation of EVE, intending to reduce AI computing costs and make AI freely accessible to everyone.
Imagine every smartphone in the world as a house with an empty room. Today, companies like Amazon Web Services (AWS) and Google build huge hotels (data centres) where everyone rents a room to store data or run AI. EVE wants to use those empty rooms in millions of houses instead. Its EVE Box connects phones into a shared network so they can lend their unused storage and computing power to each other. In places such as universities or remote villages, the EVE Box first creates a local network that works even without the internet. As more devices join, the company aims to build a large mesh network where information can hop from one device to another.
When asked how EVE Box differs from peer-to-peer technologies such as ESP-NOW, Jash said, “ESP-NOW is only a communication layer protocol, allowing devices to exchange data directly over Wi-Fi or Bluetooth.” EVE’s orchestration technology goes further, he explained, by splitting a file into hundreds of small encrypted fragments spread across phones on the network. A 6MB image, for example, is divided into pieces of a few kilobytes each, with every phone storing one fragment. When a user requests the file, the system reassembles it using stored coordinates. This reduces the storage burden on individual devices and improves security, since anyone intercepting the data would get only a small, unusable fragment.
EVE NET’s core architecture is built around this orchestration, which distributes AI workloads and data according to each device’s available resources. If one phone has 2GB RAM and 64GB storage while another has 8GB RAM and 256GB, the larger device is assigned a bigger portion of an AI model, while the smaller phone handles only what it can process without overheating or slowing down. The system monitors the network and redistributes models and data whenever devices join or leave. Jash says that if a phone’s battery runs low, the system shifts its data to the next available device. If a phone breaks down or is lost, other devices already hold copies of every fragment, so the file can be rebuilt without losing any data.
Building an offline network was not the hard part, he adds. The real challenge was developing orchestration technology that can run AI across millions of devices with different computing capabilities, and shift workloads automatically as devices join or leave. Recent advances in AI model compression helped make this possible.
The company currently outsources the manufacturing of its EVE Boxes because it does not yet have its own manufacturing facility. Founded only months ago, it sources PCBs and other hardware components from suppliers in Japan and China, while external partners handle assembly.
On revenue, Jash said, “We have recorded ₹600,000 in revenue in our first three months of operations and expect to reach ₹2.4 million by the end of this year. Most of it comes from two construction sites in Mumbai, with 10 EVE Boxes at each site. We have also onboarded three universities, with 15 EVE Boxes at each campus.”
The company is bootstrapped. Its initial capital came from the acquisition of the founder’s previous startup, Velvette AI, for US$50,000, which has funded EVE NET’s research and product development so far.
Looking ahead, the focus is to reduce the cost of the EVE Box while expanding the EVE NET ecosystem. The company is preparing to launch a software development kit (SDK) that will allow developers to host applications and store data on the EVE NET platform instead of cloud services such as AWS or Firebase. The company expects this to increase the number of connected devices beyond the nearly 30,000 already deployed across universities, with a target of 50,000 to 100,000 by onboarding popular applications. Its primary focus is growing its developer community rather than expanding hardware distribution through channel partners.
