Ramin Shirani Has Spent a Career Teaching Machines to Talk

Ramin Shirani’s favorite professional achievement happens before most people notice that anything has happened. Plug an Ethernet cable into a computer and the devices conduct a quick, private negotiation. They announce what they can do, settle on a compatible speed, then begin the conversation. Shirani helped devise that ritual at National Semiconductor in the early 1990s. Auto-negotiation became part of the Ethernet standard and found its way into billions of devices. It is a modest sort of magic: two strangers meeting on a wire, agreeing on terms, and getting to work.

That mechanism is also a neat miniature of Shirani’s career. For nearly four decades he has worked in the narrow, essential territory between machines. His companies have built the physical-layer chips and network architecture that move information from one point to another. The speeds changed from 10 megabits to 100, then to multi-gigabit links. The destination moved from office networks and data centers into cars, robots, and factory systems. The underlying question stayed put: how do you make complicated equipment communicate quickly, predictably, and without drama?

A working education

Shirani came to the United States from Iran with his family when he was 14, shortly before the Iranian Revolution. He has described the change plainly: a comfortable life became a start from scratch. As a teenager, he delivered pizzas to help make ends meet. He attended high school in San Francisco, went to college in the Bay Area, and did most of his university work at the University of California, San Diego, where he earned a master’s degree in electrical engineering. School and work ran side by side.

In 1987, he joined National Semiconductor. Ethernet was young enough to feel like open country. National was building 10-megabit technology, and Shirani became involved in the move to 100 megabits. He recalls the first decade of his career as a prolonged learning sprint. He crossed the usual engineering borders: digital design, analog design, product testing, and customer meetings. Sixteen-hour days are not a life philosophy worth laminating, and Shirani does not present them as one. For him, they were an accelerated extension of school, a chance to absorb the entire business surrounding a chip.

“You have to have a healthy arrogance.”Ramin Shirani on founder conviction

The phrase sounds dangerous until he supplies its counterweight: keep an open mind. Shirani’s version of founder confidence is not the belief that every first answer is correct. It is the belief that the problem is worth staying with while the answers improve. Build relationships. Recruit people who are better than you. Empower them, then move out of their way. His career contains enough rejection to make that advice more than conference-room embroidery.

One no, repeated a hundred ways

In 1996, Shirani and colleagues left National to start the LAN division of Enable Semiconductor. The timing caught the transition from 10-megabit to 100-megabit Ethernet. Lucent acquired the business in 1999, and Shirani moved from hands-on chip design into running a transceiver group. He was suddenly responsible for a team and a sizable profit-and-loss statement. The engineer was becoming an operator, although the urge to build another company had not gone away.

The next attempt arrived in a far colder funding climate. Between roughly 2001 and 2004, while developing the idea that became Aquantia, Shirani says he heard no more than 100 times from venture capitalists and corporate investors. The market had reasons. The dot-com crash had scorched enthusiasm for communications silicon, and a 10-gigabit transceiver over ordinary copper looked technically severe. A sensible person could have taken the hint. Shirani took notes.

Aquantia was founded in 2004. Shirani served on its board and led engineering, first as vice president and later as senior vice president. The company pushed high-speed Ethernet over twisted-pair copper for data centers and enterprise networks. In November 2017, Aquantia listed on the New York Stock Exchange. From the podium, Shirani learned a charmingly mechanical detail about financial theater: there was no physical bell for the team to pull. The famous sound was automatic. After years of making devices negotiate, his company had arrived at a ceremony that negotiated itself.

The car is a crowded conversation

Aquantia’s main market was the data center, but the team could see Ethernet edging into vehicles. Cars were accumulating cameras, radar, lidar, displays, processors, and software, all joined through a patchwork of specialized networks. Shirani has described a midrange vehicle as containing dozens of networks and well over a hundred electronic control units. Many could not easily talk to one another. The more intelligent the car became, the more antique its internal conversation looked.

Shirani left Aquantia in 2018. With fellow Ethernet veterans, he incorporated Ethernovia and supplied the initial seed capital. Then they did something refreshingly slow for a chip startup: they spent about a year speaking with automotive customers in Europe, North America, and Asia before deciding exactly what to build. The inquiry shifted the plan from a single fast component to an end-to-end system of chips and software. They wanted to connect sensors to CPUs and GPUs over a more unified Ethernet backbone, while preserving the safety, security, and timing guarantees a moving vehicle requires.

This is where the popular picture of artificial intelligence gets incomplete. Attention gathers around the brain: the processor, the model, the decision. Ethernovia works on the nerves. Sensor data must arrive at the processor intact and on time. Control instructions must travel back with predictable latency. When traffic surges, an entertainment stream cannot crowd out a safety-critical message. A robot reaching for an object and a car reacting at highway speed both depend on timing that “best effort” networking was never designed to promise.

“Control, perception and decision-making are only as good as the timing and integrity of their inputs.”Ramin Shirani on physical AI

Ethernovia’s argument is that Ethernet can become a governed, observable fabric for the whole machine. Its products include single-pair automotive Ethernet PHYs, deterministic packet processors, and software. The company uses a 7-nanometer process for lower power and greater integration, useful in electric cars and battery-powered robots where every watt has somewhere else to be. The goal is not speed in isolation. It is bandwidth with bounded delay, time synchronization, traffic isolation, diagnostics, and the ability to update a system without rewiring it.

The nervous system leaves the car

In May 2023, Ethernovia raised a $64 million Series A with backing that included Porsche SE, Qualcomm Ventures, AMD Ventures, and others. The company also announced work with Continental on a 7-nanometer automotive switch. In January 2026 came a $90 million Series B led by Maverick Silicon, with returning strategic investors and several new funds. The money is intended to advance packet processors, PHYs, compute devices, software, customer support, and hiring.

The company’s vocabulary has expanded with its market. It now speaks less narrowly about automotive Ethernet and more about physical AI: autonomous vehicles, humanoid robots, industrial automation, and other machines that perceive, decide, and act in the real world. The common requirement is deterministic connectivity. A packet processor designed for a vehicle’s zonal architecture may also live in a robot’s torso or an industrial control cabinet. Different machine, familiar argument: the network deserves to be designed as carefully as the compute.

There is a pleasing recursion here. As a young engineer, Shirani worked on the short exchange that helps two Ethernet devices discover how to speak. At Ethernovia, he is enlarging that question to an entire machine. Which sensors may speak? When? At what priority? What happens during congestion or failure? How can the architecture change through software without turning unpredictable? The answers reside in packet queues, timing budgets, and silicon. The consequence is legible to anyone: a machine that responds when it should.

The open mind clause

Shirani’s technical role model is Bob Metcalfe, the Ethernet co-inventor. What impressed him during their encounters was not merely the famous invention. It was Metcalfe’s continued presence at technical meetings, listening to presentations and offering constructive feedback decades into his career. Shirani reads that behavior as an injunction against expertise becoming a locked room. Anyone, at any point, can still learn and grow.

Away from work, Shirani has spoken about making time for his family, traveling with his wife, and joining his younger son’s soccer training. He played soccer himself and once arranged a summer camp in England connected with Chelsea. Even there, the operator’s reflex peeked through: when an interviewer noted an electric-vehicle maker’s sponsorship around the sport, Shirani joked that perhaps he could do some business too. The founder brain, like Ethernet, is difficult to unplug.

His career resists the neat myth of one blinding insight. It looks more like compounding attention. Learn every corner of the first job. Notice the next technical transition. Hear no, revise, return. After an acquisition, learn to run a business. After an IPO, go back to the empty whiteboard. Keep enough arrogance to believe you can build the missing thing, and enough humility to spend a year asking customers what that thing should be. The handshake matters. So does knowing when to listen.