Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Amnon Shashua’s January 2022 AnandTech interview described an ambitious route from camera-based driver assistance to affordable Level 4 autonomy. The plan depended on efficient, purpose-built EyeQ compute, independent camera and radar-lidar sensing, REM mapping, and the Responsibility-Sensitive Safety (RSS) framework. It also attached bold targets to EyeQ Ultra: about 176 TOPS, less than 100 watts, a system cost below $5,000, and consumer autonomous-driving hardware near $10,000.
Those numbers and dates were projections, not guarantees. By August 18, 2026, Mobileye still presents a graduated path from ADAS to SuperVision, Chauffeur and Drive, but its public product descriptions emphasize EyeQ6H-based systems and a planned U.S. robotaxi launch in 2027. The interview remains useful as a record of Mobileye’s architecture and safety philosophy—and as a test of how autonomous-driving predictions survive contact with deployment, regulation and business reality.
As an Amazon Associate I earn from qualifying purchases.
What the 2022 interview was really about
Published by AnandTech on January 4, 2022, Dr. Ian Cutress’s interview took place around CES and focused on Mobileye’s announcement of EyeQ Ultra. Shashua, Mobileye’s co-founder and chief executive, presented the chip as an all-in-one platform for Level 4 autonomous vehicles. He argued that driverless capability would not come from a larger neural-network number alone. It would require a complete system: efficient accelerators, diverse sensors, redundant computation, detailed maps, a formal safety policy and economics that could work in a production vehicle.
The conversation covered Mobileye’s history in computer-vision ADAS, the transition from a single forward camera to multi-sensor autonomy, robotaxis versus privately owned autonomous cars, legal and ethical responsibility, and broader questions about machine intelligence. Shashua is both a computer-vision and machine-learning researcher and a Hebrew University professor. Mobileye’s 2026 investor material still identifies Professor Amnon Shashua as president and CEO (Mobileye investor release).
#1 Best Overall
The original interview is therefore best read in two ways: as a historical statement of Mobileye’s strategy in 2022, and as a benchmark against the company’s publicly described position in 2026.
EyeQ Ultra: the 2022 bet on efficient automotive compute
In the AnandTech discussion, EyeQ Ultra was described as a 5-nanometer Level 4 system-on-chip intended to consolidate the processing needed for autonomous driving. Shashua gave the following specifications and targets:
| Item | Statement made in the January 2022 interview |
|---|---|
| Accelerators | 64 accelerator cores across four accelerator families |
| CPU | 12 RISC-V CPU cores |
| Other processing | Arm GPU and DSP components |
| Deep-learning compute | Approximately 176 TOPS |
| Process | 5 nm |
| Target system power | Below 100 watts |
| First silicon | Near the end of the fourth quarter of 2023 |
| Production ambition | Automotive production extending into 2025 |
| Economic targets | Complete Level 4 system below $5,000; consumer option around $10,000 |
Every date, specification and price in this table is a historical company statement from that interview. It is not a current retail price, a verified bill of materials or proof that the proposed production schedule occurred exactly as described.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
The design goal was not simply to put a powerful general-purpose processor in a car. An automotive chip must process camera, radar and lidar data with predictable latency, stay within a thermal envelope, support safety monitoring and continue operating—or reach a safe state—when something fails. Consolidating functions can reduce component count and power, but it also makes architectural redundancy and fault containment especially important.
Why Shashua said TOPS is not enough
TOPS (trillions of operations per second) is a useful headline measure of neural-network throughput, but it is not a complete measure of an autonomous-driving system. In the interview, Shashua argued that comparisons based mainly on TOPS ignore several variables:
- Different workloads and numerical formats can make nominal operations incomparable.
- Sparsity can reduce the work a network actually performs.
- Perception is only one part of the problem; planning, prediction, control and safety monitoring also consume compute.
- Memory bandwidth, interconnects, sensor input rates and latency can limit a system before arithmetic capacity does.
- Thermal behavior, functional-safety mechanisms and redundancy affect what can be delivered continuously in a vehicle.
- Purpose-built accelerators may perform a particular automotive workload more efficiently than a larger but less specialized processor.
That argument does not prove that Mobileye’s 176-TOPS design is more efficient than a rival. The interview contains no independently controlled, like-for-like benchmark. A serious evaluation also needs the operating domain, sensor throughput, end-to-end latency, validation evidence, software maturity, mapping coverage, power consumption and production cost.
From cameras to “true redundancy”
Mobileye’s safety case in the interview depended on independent sensing and computation rather than on one fused perception result. Shashua described two main streams:
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitches- A camera-only stream.
- A separate radar-and-lidar stream.
The streams were intended to remain independent so that a failure or weakness in one would not automatically contaminate the other. EyeQ Ultra was also described as having internal compute redundancy, alongside an external safety microcontroller and a separate fail-operational stream.
Mobileye still calls this approach True Redundancy. Its current Chauffeur description identifies two independent systems: a camera system and a radar-lidar system.
Four kinds of redundancy
- Sensor diversity: cameras, radar and lidar observe the world using different physical principles.
- Algorithmic diversity: separate software paths interpret the environment in different ways.
- Compute redundancy: backup processing capacity can detect or tolerate faults.
- Fail-operational capability: the vehicle can continue a defined maneuver or reach a safe state after a fault.
These concepts should not be confused with driver fallback. A driver who is expected to resume control is not the same as a driverless safety architecture. Redundancy must cover more than perception: power, braking, steering, communications, monitoring and the procedure for stopping safely all matter.
REM mapping: useful context, not a substitute for perception
Mobileye’s Road Experience Management (REM) is a crowdsourced mapping system. Equipped vehicles contribute road data that can be processed into detailed information about lanes, road geometry and the driving environment. The 2022 interview treated mapping as part of autonomous perception, not merely a navigation convenience.
REM remains central in Mobileye’s current Chauffeur and Drive descriptions. Fleet scale can help a system anticipate road structure and localize itself, but a map is never the whole world. Construction, temporary lane shifts, fallen objects, unusual road users, weather and newly changed signs can make a stored map incomplete or stale. Onboard sensors must still detect what is happening now.
RSS: a formal safety policy, not a guarantee
Responsibility-Sensitive Safety, or RSS, is Mobileye’s mathematical framework for describing safe driving behavior. It addresses questions such as how much distance to keep, when a merge is acceptable, what assumptions can be made about another road user and how a vehicle should respond to an unpredictable maneuver.
Mobileye describes RSS as an open-source, verifiable approach to driving policy on its current product pages (Chauffeur). The value of a formal policy is that safety decisions can be stated, inspected and defended rather than left entirely to an opaque learned model. RSS does not prove that every deployment will avoid crashes. Its effectiveness still depends on accurate perception, correct assumptions, appropriate operating limits and a vehicle that can execute the required action.
Robotaxis and consumer autonomous cars are different products
Shashua’s interview distinguished between a fleet robotaxi and a privately owned autonomous vehicle. A robotaxi can operate in a defined operating domain, use centralized fleet support and receive maintenance on a schedule. A consumer vehicle may be driven to an unfamiliar destination, encounter roads outside its mapped area and be expected to work across a much wider range of conditions.
Robotaxi constraints
- Geofenced or otherwise defined operating domains.
- Remote assistance and escalation procedures.
- Fleet cleaning, charging, sensor inspection and repair.
- Passenger safety, vandalism, violence and emergency response.
- Regulatory approval and economics for every service area.
Consumer-vehicle constraints
- Unpredictable destinations and changing road conditions.
- More varied weather, geography and road coverage.
- Clear responsibility when the system reaches its limits.
- Vehicle price, insurance, servicing and software support over many years.
A car that can drive itself in a constrained domain is not automatically a profitable, broadly available consumer product.
What happened to the 2022 predictions?
The interview’s early-2020s dates should be treated as forecasts. The public evidence available by August 18, 2026 supports a mixed assessment:
| 2022 claim | Status by August 18, 2026 | How to read it |
|---|---|---|
| EyeQ Ultra production around 2025 | Current public product pages emphasize EyeQ6H-based configurations; the supplied public sources do not establish the original mass-production claim. | Historical target, not a confirmed production history. |
| Consumer autonomous-driving option near $10,000 | No current public retail price is established in the cited sources. | Ambitious company projection. |
| Complete Level 4 system below $5,000 | No current public bill of materials or verified market price is established. | Target economics, not a price consumers can rely on. |
| Robotaxis in the early 2020s | Mobileye continues to develop Drive and announced a planned U.S. robotaxi launch for 2027. | The timing changed; the strategic direction remains. |
| Gradual progression from ADAS to autonomy | Mobileye’s current product ladder still runs from conventional ADAS through SuperVision and Chauffeur to Drive. | The core architecture strategy remains visible. |
What aged well
- Purpose-built automotive compute matters more than a headline TOPS figure.
- Safety is a system-architecture problem, not only a perception-model problem.
- Mapping and fleet data can add valuable road context.
- Consumer autonomy must be cost-conscious.
- Robotaxis and consumer vehicles have different technical and commercial requirements.
- Moving incrementally from ADAS toward higher automation is more practical than treating autonomy as one feature switch.
The unresolved part is not whether the concepts are technically plausible, but when they can be validated, approved and operated at useful scale.
Mobileye’s current portfolio
Mobileye now presents autonomy as a staged portfolio rather than one undifferentiated “self-driving” product:
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →| Product | Automation position | Public description |
|---|---|---|
| Mobileye Base ADAS | Hands-on, eyes-on | Single forward-facing camera and EyeQ-based driver assistance. |
| Cloud-Enhanced ADAS | Hands-on, eyes-on | ADAS enhanced with REM mapping. |
| Surround ADAS | Hands-on or hands-off, eyes-on | Surround cameras and radar with EyeQ6H. |
| SuperVision | Hands-off, eyes-on | Advanced assistance; the driver must remain attentive. |
| Chauffeur | Hands-off, eyes-off | Consumer AV technology for specified operating domains. |
| Drive | No driver | Autonomous mobility for robotaxis, public transport, ride-pooling and delivery. |
These categories come from Mobileye’s current product overview. “Hands-off” does not mean driverless. “Eyes-on” means the driver must continue monitoring the road. “Eyes-off” applies only within the approved operating domain. “No driver” describes a service designed to operate without a human driver.
Current hardware descriptions
- SuperVision: two EyeQ5 High or EyeQ6 High SoCs and 11 cameras.
- Chauffeur: three or four EyeQ6H SoCs, depending on operating domain, plus cameras, imaging radar and front lidar.
- Drive: four EyeQ6H SoCs in the current public description, with 360-degree cameras, imaging radar and front lidar.
Mobileye’s SuperVision page explicitly says the driver remains responsible and must keep eyes on the road. It is not a consumer self-driving system.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.The 2026 business-model change
Mobileye has historically been best known as an autonomy and ADAS technology supplier to automakers. On June 16, 2026, it announced plans to establish a vertically integrated robotaxi business, including a planned U.S. launch in 2027 (company announcement). The initiative was described as additive to existing automaker and mobility-partner programs.
That shift matters because operating a robotaxi service adds responsibilities beyond building a vehicle system: fleet operations, charging, cleaning, maintenance, remote assistance, passenger support, insurance, local approvals and unit economics. A technically capable vehicle still has to function as a reliable service.
Scale claims need their denominators
Mobileye’s current website says its technology has powered more than 250 million SoCs, while its Drive page describes experience across more than 150 million vehicles globally. Those are different measures and must not be added together or treated as interchangeable. The 2022 interview separately stated that Mobileye had shipped 100 million chips at that time; that was a historical figure.
Technical trade-offs and failure cases
Camera-centric systems
- Advantages: lower sensor cost, rich semantic information and potentially simpler packaging.
- Risks: glare, darkness, fog, snow, spray, dirty lenses, occlusion and dependence on image quality.
Radar and lidar redundancy
- Advantages: different sensing modalities can compensate for some camera weaknesses and provide stronger range information.
- Costs: additional hardware, calibration, integration, validation, power and maintenance.
Mapping
- Advantages: road context, localization and anticipation supported by fleet-scale data.
- Failure modes: construction, temporary lane changes, stale updates, low-coverage roads and conditions that differ from the mapped baseline.
Centralized compute
- Advantages: fewer components and possible power and integration savings.
- Risks: a central failure can affect multiple functions, so independence and fault containment must be designed in from the beginning.
Any real deployment must also handle emergency vehicles, unusual road layouts, aggressive drivers, pedestrians and cyclists behaving unpredictably, heavy weather, sensor blockage, loss of connectivity, power or steering faults, safe-stop behavior, remote escalation and passenger misconduct.
What readers should ask instead of “How many TOPS?”
- What is the operating domain? Specify geography, road type, speed, weather and other limits.
- What is the driver’s responsibility? Distinguish hands-on, eyes-on, eyes-off and no-driver operation.
- Which failures are independent? Ask whether sensor, software, compute, power, braking and steering paths can fail without a common cause.
- How does the system behave when uncertain? Look for degraded modes, minimal-risk maneuvers and safe-stop procedures.
- How current are the maps? Mapping supports perception but cannot replace real-time observation.
- What evidence exists? Separate company descriptions from independently verified deployment, safety and intervention data.
- Can the service scale? Consider regulation, maintenance, remote assistance, insurance, support and per-vehicle economics.
Who can actually buy Mobileye’s technology?
Mobileye’s relevant products are enterprise platforms, not aftermarket accessories for private cars.
| Platform | Intended customer | Commercial signal | Important limitation |
|---|---|---|---|
| EyeQ Kit | Automakers and technology partners | Official inquiry; no public self-serve price cited | Not an individual developer or consumer retrofit. |
| SuperVision | Automakers integrating production ADAS | Mobileye says nearly 300,000 equipped consumer vehicles are on the road | Requires driver attention and eyes on the road. |
| Chauffeur | Automakers and vehicle platforms | Contact-based engagement; no public consumer MSRP cited | Not currently a direct-to-consumer product. |
| Drive | Automakers, fleets, transit and mobility companies | Enterprise partnership | Requires substantial regulatory and operational capability. |
Neither the approximately $5,000 system target nor the approximately $10,000 consumer-option estimate from 2022 should be used as current pricing. The cited official sources provide no public consumer MSRP, monthly subscription or self-serve plan.
Free tools Windows power users keep installed
One-click scans. No signup required.
The bottom line
Shashua’s interview was not a promise that a 176-TOPS chip would make every car autonomous. It was an argument that autonomy requires an engineered system: efficient compute, independent sensing, mapping, formal safety rules and viable economics. That thesis still appears in Mobileye’s product ladder. What changed is the timetable and the practical emphasis. EyeQ Ultra’s projected dates and prices remain historical targets, while Mobileye’s current public strategy centers on EyeQ6H-based products, clearly separated automation levels and a planned 2027 U.S. robotaxi operation.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
