EnCharge AI

A chip company that sells less power, not more

Most of the AI industry is busy building bigger. Bigger clusters, bigger cooling towers, bigger quarterly power bills. EnCharge AI is selling the opposite proposition: a chip small enough to slide into a laptop, frugal enough to run on a battery, and capable enough to run a generative model without phoning a server farm. Its first product, the EN100, arrived publicly in May 2025 in two flavors - an M.2 module for laptops and a PCIe card for workstations.

The pitch is unglamorous and, for once, specific. Move AI inference out of the cloud and onto the device. Keep the data local. Cut the energy. Verma, the CEO, puts it plainly: advanced, secure and personalized AI can run locally, without relying on cloud infrastructure. It is the kind of sentence that sounds modest until you remember the entire industry is currently built on the opposite assumption.

AI got smart. Then it got hungry.

Here is the tension that EnCharge exists to resolve. The machine-learning models everyone wants - chatbots, real-time vision, image generation - are enormous, and the operation they perform most is deceptively dull: multiply a number, add it to a running total, repeat billions of times. On a conventional chip, the expensive part is not the math. It is the commute. Data shuttles back and forth between memory and processor, and every round trip burns energy.

Scale that across a data center and you get the headline of the decade: AI's appetite for electricity is starting to look like a small nation's. The standard answer is to build more power plants. EnCharge's founders asked a more impertinent question - what if you simply stopped moving the data around?

Three people, one stubborn idea

The idea came out of Naveen Verma's lab at Princeton, where he is still a professor of electrical and computer engineering. The technique is called analog in-memory computing: do the multiply-and-add right inside the memory itself, by reading the electrical charge that accumulates on precise metal capacitors. No commute. The catch, historically, is that analog is finicky - temperature, manufacturing variation and noise all conspire to make it imprecise. Verma's bet was that capacitors, unlike the transistors others had tried, are stable enough to be trusted. More than eight years of research, across multiple generations of silicon, went into proving it.

An academic, an IBM lifer, and a semiconductor operator walk into a startup. The joke writes itself; the silicon, less so.

They spun the company out in 2022. The U.S. Defense Advanced Research Projects Agency noticed and wrote a $18.6 million check. Then, in February 2025, the venture world noticed too.

The EN100, in two sizes

The headline number is the one that makes hardware engineers do a double-take: 200+ trillion operations per second inside an 8.25-watt budget. That is the M.2 laptop module - it does more AI math than many desktop accelerators while drawing less power than some phone chargers. The bigger sibling, a PCIe card with four neural processing units, pushes toward roughly 1 petaOPS, putting GPU-class throughput on a workstation without the GPU-class electricity bill.

The cap table is the tell

You can learn a lot about a chip company from who is willing to fund it. EnCharge's oversubscribed Series B was led by Tiger Global, but the more telling names sit beside it: Samsung Ventures (consumer electronics), RTX Ventures (defense), the Foxconn-linked HH-CTBC (manufacturing), and In-Q-Tel, the venture arm tied to the U.S. intelligence community. When a company's investors include both a phone-maker's fund and a spy agency's fund, it is usually a sign the technology is genuinely cross-cutting - or genuinely strange. Here, a bit of both.

AI compute from the edge to the cloud, for every business

That is the company's own line, and underneath the marketing gloss sits a real argument about access. If running capable AI requires a data center, then capable AI belongs to whoever can afford a data center. EnCharge frames its work as democratizing - making advanced AI cheap enough, efficient enough and local enough that it can be woven into ordinary devices. The version of the future the company is selling is one where the intelligence stays on your machine, your data never leaves it, and nobody has to build a new power station to make it happen.

It is worth being skeptical here. Plenty of chip startups have promised to dethrone the GPU and quietly disappeared. Analog computing in particular has a long graveyard. EnCharge's answer is not a pitch deck but silicon that has been measured, and a roster of strategic investors with every incentive to verify the claims before wiring the money.

The commute, eliminated

Step back to that Santa Clara office, where the arithmetic is being done the old way - by reading charge instead of counting bits. If EnCharge is right, the heaviest part of AI stops being a thing that happens somewhere else, on a machine you rent, powered by electricity someone else burns. It becomes a thing that happens in your hands, on a chip that sips power, doing 200 trillion operations per second while your battery barely notices.

That is the bet. Not that EnCharge will out-muscle the data center, but that it will make the data center unnecessary for the work most people actually do. The industry spent a decade learning to move data faster. EnCharge's wager is that the smarter move was to stop moving it at all. The heresy, it turns out, was the point.