A Conversation with Arbe CEO Kobi Marenko
Radar has been around for nearly a century, but its role in modern mobility is only now coming into focus. In a recent IsraelTech interview, Yoel Israel sat down with Kobi Marenko, CEO and Co-founder of Arbe, to unpack how radar is evolving from a simple detection tool into a core perception layer for autonomous systems.
What followed was a conversation that moved fluidly between deep tech, semiconductors, automotive reality, and even philosophy.
Radar Is Old Technology. The Use Case Is New.
Radar dates back to World War II. At its core, the idea has always been simple: transmit a signal, wait for it to bounce back, and analyze what returns.
What has changed is not the physics, but the ambition.
Arbe focuses on millimeter-wave radar, transmitting signals whose reflections allow the system to build a detailed picture of the environment. Unlike cameras, radar does not depend on light. Unlike lidar, it works reliably in rain, fog, dust, and harsh sunlight.
According to Marenko, radar’s real strength is consistency. It works in conditions where other sensors fail, which makes it critical for autonomous driving.
Why Cameras Are Not Enough
Cameras can solve most driving scenarios. Marenko estimates around 95 percent.
The remaining 5 percent is the problem.
Edge cases include:
- Heavy rain or fog
- Glare from low sun angles
- Long-range detection
- Small or partially obscured objects
These are exactly the scenarios where safety matters most. Radar is often the only sensor that can reliably confirm whether an object is actually there.
This is why Arbe treats radar not as a supporting sensor, but as a primary perception system.
What Makes Arbe Different
Early automotive radars were built for simple tasks like adaptive cruise control or blind-spot detection. They did not need high resolution. They only needed to answer one question: is something in front of me?
Arbe’s approach is fundamentally different.
The company set out to build imaging radar, capable of constructing a detailed environmental map. That meant dramatically increasing the number of transmit and receive channels, which directly increases resolution.
The problem was that no existing silicon could support this level of performance.
So Arbe built its own.
When a Radar Company Becomes a Semiconductor Company
Arbe did not initially plan to design chips. The team assumed suitable radar silicon already existed.
It did not.
High-resolution imaging radar requires massive parallel processing, extreme data throughput, and precise signal handling. The available chips simply could not deliver.
That realization forced a strategic shift. Arbe became a semiconductor company by necessity, not by design.
The result is a proprietary radar chipset that enables ultra-high-resolution sensing, something traditional automotive radars were never built to do.
4D Radar Explained
Arbe’s system operates in four dimensions:
- Horizontal position
- Vertical position
- Depth
- Velocity
Velocity is especially important because radar measures it directly through Doppler effects, rather than inferring it across multiple frames like cameras or lidar.
The challenge is scale.
A single radar frame can theoretically generate enormous amounts of data. Arbe’s core intellectual property lies in compressing and processing this data efficiently, producing a usable point cloud that downstream systems like Nvidia platforms can work with in real time.
From Prototype to Production Is the Hard Part
Demonstrating radar performance in a lab or demo vehicle is one thing. Making it automotive-grade is another.
Marenko describes automotive development as a long and unforgiving process:
- Systems must work continuously, without resets
- Failure rates are measured in parts per million
- Reliability is a life-or-death requirement
He notes that roughly 90 percent of the challenge is scientific. The remaining 10 percent, industrializing the solution to meet automotive standards, can take just as long.
This is where Tier-1 suppliers come in. Arbe provides the chips. Tier-1s handle manufacturing at massive scale and integrate the radar into production vehicles.
Working With Nvidia and the Automotive Stack
Arbe collaborates with companies like Nvidia and Qualcomm, which provide the central compute platforms used in autonomous vehicles.
The relationship is complementary:
- Arbe delivers the radar data
- Tier-1s build the radar systems
- Nvidia processes perception and decision-making
No single company owns the entire stack. Autonomous driving only works when sensors, silicon, and software evolve together.
Semiconductors Are Getting Harder for Startups
Marenko is candid about the state of the semiconductor industry.
Advanced process nodes are becoming prohibitively expensive. A single tape-out at 3 nanometers can cost an order of magnitude more than older nodes. Talent is scarce, salaries are extreme, and competition with global giants is brutal.
This environment makes it harder for startups to innovate at the silicon level, even as demand for specialized chips continues to grow.
Israel, however, remains a global center of gravity thanks to deep talent pools and the presence of major R&D centers.
Beyond Tech: Philosophy, Risk, and Perspective
The conversation also veered into less common territory for a tech interview.
Marenko studied physics and philosophy, with a focus on Eastern thought. He practices yoga and meditation and speaks openly about humility, luck, and the dangers of ego in tech.
His view is simple: success in startups is rarely just skill. Timing and luck matter more than most people admit.
That perspective informs how he invests, how he builds companies, and how he thinks about technology’s role in society.
Why Radar Still Matters
Autonomous driving has taken longer than many predicted. But Marenko remains confident that radar will play a central role when it arrives at scale.
Cameras see well most of the time. Radar sees reliably all of the time.
That difference matters.
