Video summary
Lidar Fell From $75,000 to $200: How 64 Lasers Became a Chip
Main summary
Key takeaways
Summary
The video explains how lidar technology evolved from a roughly $75,000 unit in 2007 to Hesai’s ATX, which sells for around $200. The price drop involved more than higher-volume manufacturing.
How the original lidar worked
Velodyne’s spinning unit used 64 laser-and-receiver pairs, each aligned at a slightly different angle. Each laser measured distance by timing how long a light pulse took to return. Millions of measurements formed a 3D “point cloud.”
The 64 lasers traced 64 lines, but the spacing between them left substantial gaps at long range. Hand assembly and calibration also limited production.
Lasers became chips
Older lidar lasers emitted light from the cut edge of a wafer, so each had to be separated, tested, and installed individually. VCSELs emit light from the wafer’s surface, allowing many lasers to be made and tested together and packed onto a chip.
The video notes that VCSELs are relatively weak, so conventional side-emitting lasers can retain an advantage in long-range applications.
Sensors became chips, too
Older receivers used avalanche photodiodes and required amplification circuitry for each line. SPADs can register single photons, but sunlight can trigger them as well.
To distinguish a real return from background light, the system repeatedly fires at the same spot and accumulates photon arrival times into a histogram. A consistent peak indicates the reflected laser pulse. Because SPAD output is digital, sensing pixels and timing circuits can be integrated into stacked chips.
Why chip-based lidar costs less
Combining lasers, receivers, drivers, and signal processing into chips reduces the number of components and simplifies assembly. Machines can align and calibrate the resulting units. The video cites an automated factory cycle of about eight seconds per unit and a 2023 Hesai claim of 45 seconds. It also reports that Hesai shipped 1.62 million lidars in 2025.
The ATX uses a rotating mirror rather than spinning an entire sensor head, making it smaller and lighter than the old Velodyne unit.
More lines, but not all models are equivalent
With chip-based receivers, line count can be determined by pixel rows rather than by the number of individually fitted lasers. RoboSense’s Phoenix and Hesai’s Picasso chips were described as producing 2,160 lines; Hesai’s ETX lidar was said to reach 4,320. These high-resolution chips were slated for mass production, but the video says their prices had not been announced.
The sub-$200 ATX, by contrast, tops out at 256 channels. It is not a 2,160-line unit.
Tesla and the limits of the price argument
The narrator challenges the claim that Tesla’s camera-only approach is now merely stubborn because lidar has become cheap. Elon Musk has argued that lidar is unnecessary, while the video says the cost objection has weakened. It also reports that lidar has been used on Tesla test vehicles as a reference for checking camera distance estimates.
Volvo removed lidar from the 2026 EX90, and its supplier Luminar later filed for bankruptcy protection. The narrator argues that the broader test is safety evidence—whether lidar measurably reduces crashes—not simply resolution or price. The video cites Hesai chief scientist Sun Kai’s view that 500 lines and 300 m of range is a reasonable benchmark.
Review or guide content
This is an explanatory technology analysis, not a product review or hands-on tutorial.
Main speakers and sources
- Main speaker: The video’s narrator, who explains the engineering history and offers the closing analysis.
- David Hall: Quoted discussing the original 64-laser design; his 2007 patent is referenced.
- Company and industry sources: Hesai, RoboSense, Ouster/Velodyne, Sony Semiconductor Solutions, Luminar, and Volvo, as represented through statements or product details cited in the video.
- Other named voices: Elon Musk on Tesla’s camera-only position; Hesai chief scientist Sun Kai on lidar performance benchmarks.
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.