Innoviz

What a camera does not know, and why it matters at high speed

A camera measures light, not distance. It infers depth from context: relative size, perspective, movement between frames. That inference is very good in most cases and fails exactly in the hard ones: low light, a low sun facing the lens, an object the same color as the background, a smooth surface with no texture.

The difference between "there is something in front of me" and "the object is 62 meters away and getting closer" is the difference between a warning and a decision to brake. At 100 kilometers per hour, every tenth of a second of uncertainty is almost three meters.

A laser skips the inference. It sends out a pulse, measures the time until it returns, multiplies by the speed of light and divides by two, because the pulse made the trip twice, and gets a direct distance to every point, even in total darkness. This is where the logic of a complementary sensor comes from: not because the camera is bad, but because it measures something completely different. Two systems that rely on different physical principles are also less likely to fail at the same moment and in the same way. That is why the automotive LiDAR category exists at all.

Why it is so hard to make automotive LiDAR cheaper

A laser that measures distance is an old idea. What is hard is putting it into a car made in millions of copies. The early generations were mechanically spinning units: expensive, large, and sensitive to vibration and heat. A car is not a lab. The component has to survive a decade and a half of bumps, sun, cold and shocks, and pass a carmaker's quality standards.

In its newer programs, Innoviz works as a direct Tier 1 supplier, meaning it sells to the carmaker itself and not through an intermediary. It develops automotive-grade sensors without large spinning parts.

There is also a second problem that gets less attention. A sensor that produces too much data pushes the bottleneck further down the line, to the bandwidth and to the computer that must process everything in real time. So the company also packages a product that combines the laser with an Nvidia processor, to compress the information at the edge rather than send it on raw. This pattern repeats across all of sensing: once the sensor improves, the problem moves to whoever has to read its output.

Three sensor generations, and why each exists

The first generation, InnovizOne, is the one that went into BMW's Level 3 autonomous driving program. In that program Magna was the Tier 1 supplier to BMW and Innoviz supplied through it, while in later programs the company sells directly to carmakers.

The second generation, InnovizTwo, was aimed at higher performance at a lower production cost, and it is the one now being shipped for Daimler Truck's autonomous trucks.

The third generation, InnovizThree, was unveiled in 2026. What is interesting about it is the problem it is meant to solve: mounting behind the windshield instead of on the car's roof. This requires a smaller volume and lower power consumption, so as not to hurt the car's design and cabin. It sounds like a minor consideration, but it is often what decides whether a carmaker will agree to integrate a sensor at all. This version also includes data fusion with a camera, to save the carmaker part of the integration work.

In parallel, the company developed a line for non-automotive uses, including perimeter security for facilities and smart transportation systems, a market where the sales cycle is much shorter.

The customers: BMW, Volkswagen, Daimler Truck

The announced customer list is the company's main asset, and it is concentrated in a few large programs. BMW's Level 3 driving program uses InnovizOne. The Volkswagen Group chose the company's technology for its software arm, and the autonomous ID. Buzz, which runs on the Mobileye Drive platform, is in a Level 4 program whose series production is planned for 2026.

In 2025 the company was selected by Daimler Truck and Torc Robotics for series production of sensors for heavy Level 4 autonomous trucks, meant to run on highways in North America. Shipments of InnovizTwo for this fleet have already begun.

One feature here shapes the whole financial picture: the time from winning a program to mass production is measured in years. The company itself notes that several carmakers are running Level 3 tenders in 2026 with production planned for 2028, meaning revenue that is recorded two years after the win.

The 2025 numbers and the 2026 target

On 25 February 2026 the company reported its 2025 results: revenue of $55.1 million, more than double the previous year, with gross margins the company describes as a record. Its liquidity at the end of 2025 was about $72.1 million.

The targets it set for 2026 are revenue of $67 million to $73 million, two to three new program wins, non-automotive sales of up to 10% of revenue, and new development payments of $20 million to $30 million.

Important financial background: the company went public in April 2021 through a merger with a special purpose acquisition company called Collective Growth, at a valuation of about $1.4 billion, raising about $371 million. This route was very popular at the time, because it shortened the path to public capital compared with a regular IPO, and it almost completely disappeared afterward. Five years later, revenue is still far smaller than the valuation set back then.

Why many LiDAR companies shut down, and this one survived

In the early 2020s there were dozens of LiDAR companies around the world, based on forecasts that autonomous driving would reach the mass market within a few years. Those forecasts did not come true at the promised pace, and many of the companies, including well-known names, shut down or were absorbed.

What separated the survivors from those that fell was mainly the type of revenue. A company that lived on one-off trial orders from carmakers testing the market was left with nothing when the trials stopped. A company with a series production contract gets visibility for years ahead, even if the actual revenue comes late.

This is also Innoviz's risk, not only its advantage. Its revenue depends on carmakers' schedules, which it does not control. Revenue of $55 million against liquidity of $72 million describes a company that has shown significant growth, but whose real test, actual mass production, is still ahead of it.