【The Future Map of Autonomous Driving】 Is it true that "Tesla will win in the end"? The next 3 years as I see them

Having recently analyzed the earnings of TSLA, GOOGL, QCOM, UBER, and others, I wanted to take a moment to update my thoughts on autonomous driving as of now.

Finding small gaps in my schedule, I researched the current state of autonomous driving for myself and tried to forecast the next five years.

To begin with, I feel that the discussion around autonomous driving has tended to be overly simplified.

Will Tesla scale and win in the end?
Or will a robotaxi model like Waymo be implemented in society first?

Looking at X, perhaps because Tesla fans are so vocal, I feel the market holds the view that "Tesla will eventually take it all at once."

The reason is easy to understand.

Tesla already has a massive number of vehicles on the road worldwide, collecting real-world data, improving via OTA, and expanding FSD to existing vehicles.

If this reaches full autonomous driving, it is overwhelmingly easier to scale than a model like Waymo's, which expands steadily city by city.

However, looking at the moves of each company again, this categorization must be quite crude.

Tesla, Waymo, Wayve, NVIDIA, Qualcomm, Zoox, Chinese players, Uber/Lyft.
Each is looking at a different layer.

I don't think full autonomous driving is simply a matter of "whether a car can drive itself."

Can AI understand the physical world, respond to unknown situations, detect failures, stop when dangerous, recover after stopping, and who takes responsibility when an accident occurs?

This is a theme close to the core of Physical AI, where AI enters the real world.

■ Conclusion: Autonomous driving will not be a winner-take-all for one company, but a battle layer by layer

I will write the conclusion first.

At this point, I do not see it as "Tesla will take everything in the end."

Of course, Tesla is strong.
It has the greatest scaling option.

However, Tesla itself clearly states that the current FSD Supervised "requires active driver supervision and does not make the vehicle autonomous."

In other words, the current Tesla FSD is merely advanced driver assistance and is different from L4 unmanned driving. At the very least, it has not been proven as L4 as it stands right now.

First, as a starting point for discussion, we cannot get this wrong.

On the other hand, Waymo is actually running without human drivers, albeit in limited regions. Waymo has disclosed that by December 2025, it had completed 170.7 million miles of rider-only, or driverless, driving. This is a "narrow but real" track record.

What is important here is that while Tesla and Waymo seem to be playing the same game, they are actually different.

Tesla is trying to capture a wide ODD, that is, a wide range of drivable situations and road environments.
Waymo removes the human within a limited ODD, and the company takes responsibility for safety.

Tesla is wide.
Waymo is deep.

Unless you understand this difference, you won't be able to see the future map of autonomous driving.

And now, Wayve, NVIDIA, Qualcomm, Chinese players, and Uber/Lyft have joined in.

In other words, autonomous driving is not "Tesla vs. Waymo."
It is divided into at least the following layers.

The operation layer for robotaxi L4.
The popularization layer for mass-produced vehicle ADAS/L2+/L3.
The software layer that supplies E2E AI Driver to OEMs.
The in-vehicle AI compute layer like NVIDIA and Qualcomm.
The Physical AI development infrastructure layer like Cosmos and Alpamayo.
The verification and simulation layer like Applied Intuition.
The demand aggregation platform layer like Uber and Lyft.

How do I organize this and connect it to my portfolio?
I think this is what I should do as an investor.

■ Tesla: The greatest scaling option. However, the bridge from L2 to L4 is still unconfirmed

Tesla's appeal is easy to understand.

Cars are already running all over the world.
They control the vehicles, software, AI chips, data, OTA, and user touchpoints in-house.
If FSD reaches a level where it can run without human supervision, Tesla has the potential to spread autonomous driving all at once through existing vehicles.

This is a strength Waymo does not have.

Waymo needs to expand its service area city by city, deploy vehicles, and manage operations.
Tesla can theoretically scale by distributing software to vehicles already sold.

That is why the story that "Tesla will win in the end" is very attractive to investors.

However, there is a big leap here.

That is, the data collected through L2 driver assistance does not automatically become safety proof for L4 unmanned driving.

In FSD Supervised, humans are monitoring.
Humans intervene when it is dangerous.
In other words, the human driver is the final safety device.

On the other hand, in L4 full autonomous driving, that human is removed.
Responsibility in the event of an accident also shifts from the driver to the system side.

This difference is very large.

What you should look at as an investment decision is not how smoothly FSD runs.
What you really should look at is, when you remove the human, how often and what kind of failures occur, how those failures are detected, how it stops, how it recovers, and who takes responsibility.

This has not yet been fully confirmed.

That is why, while I evaluate Tesla as the candidate for the largest scale in autonomous driving, I cannot go so far as to say that an "irreversible moat has been confirmed" at this point.

Tesla is huge if it hits.
However, the bridge from L2 to L4 is still unconfirmed.

In this sense, for me, Tesla is closer to a high-volatility satellite candidate than a core holding as an autonomous driving stock.

■ Waymo: Is it true that it "doesn't scale"? From narrow but real to expanding L4

Waymo is often said to be like this.

The technology is amazing.
But it doesn't scale.

Certainly, there is a reason for this view.

Waymo operates in limited areas. Vehicle costs are also high. LiDAR, high-precision maps, operation management, vehicle maintenance, remote support, and coordination with local governments are also required. It is different from the model of distributing software all at once to cars already sold, like Tesla.

However, it has become a bit dangerous to conclude that Waymo "doesn't scale."

Waymo has already accumulated a track record of driving without human drivers. Furthermore, in February 2026, it announced that it had begun general services simultaneously in four cities—Dallas, Houston, San Antonio, and Orlando—expanding its commercial metro areas to 10. This is one counter-evidence to the view that Waymo can only expand one city at a time.

Of course, this alone does not mean Waymo has fully scaled.

The real point of contention for Waymo is not the number of cities.
It is the cost of adding cities.

How much map creation, verification, vehicle deployment, local government coordination, and remote support systems are needed to add one city?
What is the utilization rate per vehicle?
Can it recover vehicle costs, insurance, maintenance, remote support, and operation costs against the ride fare?

Waymo is an AI traffic operation company as well as a technology company.

This is significantly different from Tesla.

Tesla is closer to the idea of "selling cars and spreading software."
Waymo is closer to the idea of "operating unmanned traffic services within cities."

Which one is stronger depends on the time axis.

In the short to medium term, it is Waymo that is actually implementing L4 unmanned driving in society.
In the long term, if Tesla reaches L4, Tesla may have the potential to scale more widely.

In other words, Waymo is "narrow but real."
Tesla is "wide, but proof of unmanned operation is yet to come."

I think this organization is the most accurate at this point.

If you look at it as an investor, Waymo is a huge option within Alphabet.

Alphabet has Search, YouTube, Cloud, TPU, Gemini, and Android. Waymo is there. You cannot buy Waymo alone, but if you view Alphabet as a "global OS company," Waymo is a very important real-world AI option.

For me, Alphabet is not a stock for autonomous driving alone.
However, as a global OS stock in the AI era that includes Waymo, it can be placed as a core candidate.

■ Waymo is not an old company either. It is leaning toward the E2E/multimodal side with EMMA

There is another misunderstanding about Waymo.

That is to view Waymo as an "old modular company."

Waymo has indeed emphasized LiDAR, high-precision maps, sensor redundancy, safety verification, and operation management. It seems different from the philosophy of trying to solve driving End-to-End with cameras, like Tesla.

However, Waymo is also clearly leaning toward the AI model side in research and development.

Waymo has announced EMMA, an End-to-End Multimodal Model for Autonomous Driving. EMMA is explained as a model that directly outputs planner trajectories, perception objects, road graph elements, etc., from raw camera sensor data, based on a multimodal large language model.

This is important.

In other words, while Waymo already has an L4 operation track record, it is also advancing in the direction of E2E, multimodal, and world knowledge in its research.

This is strong.

It is not a simple story that Tesla will suddenly overtake Waymo with AI.
Waymo is also advancing to the AI model side.

On the other hand, Waymo has experienced real L4 operations: unmanned operation, safety verification, accident response, remote support, and city operations.

This cannot be obtained through research alone.

Autonomous driving does not end with just the model.
What to do when the model fails?
How to recover when the vehicle stops?
How to explain when an accident occurs?
How to face local governments and regulatory authorities?

Autonomous driving includes all of this.

Waymo's strength lies not only in its models but in its experience in social implementation of L4 operations.

■ Wayve: A third path that is neither Tesla nor Waymo

Next, let's look at Wayve.

Wayve is unlisted, but I think it is quite important when thinking about the future map of autonomous driving.

The essence of Wayve is an AI Driver for OEMs.

It does not make its own vehicles like Tesla.
It is not a company that operates large-scale city robotaxis itself like Waymo.

Wayve is aiming for an AI Driver that can be deployed to multiple OEMs using End-to-End AI, world models, language-based explanation, and learning loops. Wayve itself also describes its AI Driver as a general-purpose driving intelligence that extends to "any vehicle, anywhere."

Especially important is the partnership with Nissan.

Nissan plans to equip its next-generation ProPILOT, which uses Wayve AI Driver software and next-generation LiDAR, in mass-produced vehicles in Japan in fiscal year 2027. However, Nissan also clearly states that this is SAE Level 2, and the driver must monitor and immediately switch to manual operation as needed.

This is interesting.

Wayve has an AI philosophy close to Tesla.
However, Nissan also uses LiDAR.
And first, it will enter mass-produced vehicles as L2.

In other words, Wayve is aiming for an intermediate solution of "loading a Tesla-like E2E AI Driver philosophy into mass-produced OEM vehicles."

This could be a threat to Tesla.

Because Tesla's strength is not just solving driving with AI.
It is vertically integrating its own vehicles, data, chips, and software.

However, if Wayve teams up with Qualcomm, NVIDIA, Nissan, and Uber and can expand horizontally to OEMs, a path will open for automakers other than Tesla to also obtain an AI Driver.

Wayve, Uber, and Nissan have also announced an MOU to prepare to start a robotaxi pilot in Tokyo in late 2026, equipped with the Wayve AI Driver in a Nissan LEAF. This is not yet full-scale commercialization, but it is worth noting in terms of verification in the Japanese market.

However, the Wayve-style horizontal expansion also has structural weaknesses.

That is the issue of data sovereignty.

Tesla holds its own vehicles, customers, software, and data together. Therefore, it is easy to return driving data to its own model's learning.

On the other hand, when supplying an AI Driver to multiple OEMs like Wayve, the issue of who owns the driving data and whose AI learning it is used for arises.

OEMs should fear that if they let their customer data and driving data be used too much as learning assets for external AI vendors, they will eventually become just "boxes that make car bodies."
On the other hand, if they enclose data too much for each OEM, the AI Driver's learning loop will weaken.

I think this is a difficult point common to the horizontal expansion models of Wayve, Qualcomm, and Mobileye.

In other words, Tesla's vertical integration is not old; in terms of data learning loops, it is still quite strong.
While the horizontal expansion model has the strength of spreading to multiple OEMs, it carries the difficulty of adjusting data sovereignty.

Since Wayve is unlisted at this point, direct investment is difficult.
However, it is necessary to connect and watch the moves of NVIDIA, Qualcomm, Uber, and Nissan.

For me, Wayve is not a stock I can buy, but an important unlisted company that must be observed.

■ NVIDIA: Rather than guessing the winner of autonomous driving, grasp the Physical AI infrastructure

NVIDIA cannot be left out when talking about autonomous driving.

However, it is different if you view NVIDIA as a "company that makes autonomous vehicles."
The essence of NVIDIA is to capture the development infrastructure for Physical AI.

Cosmos and Alpamayo are symbols of that.

NVIDIA Cosmos is explained as a platform for World Foundation Models for Physical AI. Including video data processing, evaluation, and post-training, it becomes the foundation for AI to learn, predict, and simulate the physical world.

On the other hand, Alpamayo is announced as a portfolio of open AI models, simulation, and Physical AI datasets for autonomous driving. NVIDIA positions Alpamayo as something that accelerates reasoning-based autonomous vehicles.

What is important here is that NVIDIA is not in the same layer as Tesla or Waymo.

Tesla is vertical integration with vehicles and software.
Waymo is L4 robotaxi operation.
Wayve is horizontal expansion of AI Driver.
Qualcomm is the reality foundation for mass-produced vehicle ADAS.

NVIDIA is capturing the Physical AI infrastructure below that, including learning, simulation, synthetic data, in-vehicle AI compute, and robotics.

This is very important as an investor.

It is difficult to guess one winner of autonomous driving.
Tesla, Waymo, Wayve, Zoox, Baidu, Pony, Mobileye? This is quite hard to read.

However, no matter which camp wins, a computational foundation will be needed for AI model learning, simulation, in-vehicle inference, data generation, and robotics.

Therefore, NVIDIA is not a "stock to guess the winner" in autonomous driving, but closer to a "foundation stock that is used as more winners increase."

For me, NVIDIA remains a core holding.

Even if autonomous driving is delayed, there is room for recovery in AI factories, robotics, data centers, and Physical AI.
If autonomous driving advances, options will increase further.

I think this is truly strong.

■ Applied Intuition: Hard to see on the surface, but could become the verification OS for autonomous driving

Here, I would also like to touch on Applied Intuition.

What is easy to see in articles about autonomous driving are vehicle and robotaxi companies like Tesla, Waymo, and Zoox.
However, to actually load autonomous driving into mass-produced vehicles, the intelligence of the model alone is not enough.

How do you verify safety?
How do you reproduce rare cases?
How do you create synthetic data?
How do you connect simulation and real-world driving?
How do you match each OEM's vehicle architecture and safety requirements?

Behind this is Applied Intuition.

In March 2026, Applied Intuition announced a collaboration with NVIDIA and was positioned as a software provider for OEMs to develop and deploy L2+ autonomous driving assistance on the NVIDIA platform. In the announcement, it is explained that Applied's L2+ stack is optimized for NVIDIA DRIVE AGX Orin and the upcoming DRIVE AGX Thor, and incorporates Cosmos World Foundation Models into data augmentation and synthetic generation workflows.

This is important.

If NVIDIA has the foundation models and computational infrastructure for Physical AI, Applied Intuition has the potential to capture the verification and development workflows for the automotive industry.

In other words, if NVIDIA is the "shovel of semiconductors and foundation models," Applied Intuition is closer to the "verification OS for safely bringing autonomous driving to mass production."

Since it is unlisted, you cannot buy it directly.
However, in seeing whether the horizontal expansion of Wayve, Qualcomm, NVIDIA, and each OEM will really progress, the existence of verification infrastructure like Applied Intuition becomes quite important.

Autonomous driving is not enough just by being able to run.
It must be loaded into mass-produced vehicles safely, explainably, and while complying with regulations.

In that sense, I think Applied Intuition should be watched as a "company that is hard to see on the surface but supports the scale of the entire industry."

■ Qualcomm: The reality layer of autonomous driving that is not too futuristic

Looking only at NVIDIA, I feel it leans a bit too much into the future.

Cosmos, Alpamayo, VLA, world models, synthetic data.
All are important, but if you look at the reality layer that will actually spread advanced ADAS and L2+ to mass-produced vehicles in the next 3 to 5 years, you must also touch on Qualcomm.

Qualcomm is trying to capture the mass-production foundation for software-defined vehicles by integrating cockpit, ADAS, in-vehicle AI, communication, and cloud connection through Snapdragon Digital Chassis, Snapdragon Ride, and Snapdragon Ride Flex.

Especially important is the Snapdragon Ride Pilot with BMW. Qualcomm and BMW debuted Snapdragon Ride Pilot in the BMW iX3, verified in over 60 countries, and expect to expand to over 100 countries in 2026. The scope of support ranges from basic safety functions for NCAP to highway and urban NOA for L2+.

This is different from Waymo.

Waymo removes humans in a limited ODD.
Qualcomm/BMW widely load advanced ADAS into mass-produced vehicles.

The autonomous driving-like experience that many people will first encounter may not be a robotaxi, but mass-produced vehicle ADAS from BMW, Nissan, Mercedes, Toyota, VW, Chinese OEMs, etc.

Here, Qualcomm is in a very realistic position.

Furthermore, Qualcomm is also teaming up with Wayve. Qualcomm and Wayve have announced a collaboration to provide the Wayve AI Driver as a new AI driving software option for customers using the Snapdragon Ride Platform and Active Safety software.

This is quite big.

Wayve's E2E AI Driver will be loaded onto Qualcomm's Snapdragon Ride.
In other words, the possibility has emerged that the Tesla-like AI Driver philosophy will spread to multiple OEMs via Qualcomm.

Tesla will no longer be the only company solving driving with AI.

Furthermore, the collaboration with Bosch is also important. Using Qualcomm's Snapdragon Ride Platform, it supports everything from entry-level ADAS to advanced autonomous driving systems, and Bosch explains that it has already shipped over 10 million cockpit computers equipped with Snapdragon.

Here, I think another important thing is "inference performance per watt."

When running heavy AI models like E2E or VLA in real-time on the in-vehicle edge, it is not enough for the model to just be smart.
Power consumption, heat generation, cost, and packaging become barriers to mass production.

In the collaboration announcement between Applied Intuition and NVIDIA, it is clearly stated that for mass-produced vehicle ADAS/L2+, cost, power consumption, thermal design, and packaging become realistic constraints.

This is the interesting part of Qualcomm.

Qualcomm has the DNA of low-power, high-efficiency inference derived from smartphones.
While NVIDIA is strong as a future Physical AI foundation, Qualcomm has strengths in the reality layer of edge AI where AI descends to the terminal side, such as mass-produced vehicles, smartphones, PCs, XR, IoT, and in-vehicle AI.

In its recent earnings, Qualcomm disclosed that it achieved record quarterly revenue in QCT Automotive, and the combined revenue of QCT Automotive and IoT increased by 20% year-on-year. This is important when considering expansion into distributed AI, in-vehicle, and IoT other than smartphones.

Qualcomm is not a dream stock for autonomous driving.
It is a realistic standard foundation candidate for spreading ADAS/AD to mass-produced vehicles.

For me, Qualcomm can be placed as a core or semi-core holding.

If NVIDIA is the future Physical AI foundation, Qualcomm is the ADAS/SDV foundation that descends to mass-produced vehicles.

I think this contrast is quite important.

Note that in this reality layer of mass-produced vehicle ADAS, it is also necessary to touch on Mobileye.

Mobileye is different from Tesla or Waymo. It does not vertically integrate up to the vehicle like Tesla, and it is not a company that makes robotaxi operation its main axis like Waymo or Zoox.

Mobileye is a company that supplies ADAS/autonomous driving stacks to OEMs through EyeQ, SuperVision, Chauffeur, Drive, REM, and RSS. Mobileye itself also explains SuperVision as a Premium Driver Assist with eyes-on / hands-off, a bridge to consumer AV.

In other words, Mobileye is in a place quite close to Qualcomm.

If Qualcomm is going to capture the mass-produced vehicle ADAS/SDV foundation with Snapdragon Ride and Digital Chassis, Mobileye is going to capture it with its accumulation as an ADAS specialist and EyeQ, maps, RSS, and safety philosophy.

However, if I were to put it in my portfolio, Qualcomm is easier to place at this point.

The reason is that while Qualcomm expands to smartphones, communication, in-vehicle, SDV, and AI edge, Mobileye has a high purity in automotive ADAS, and therefore is strongly affected by OEM adoption cycles, Chinese demand, Intel control, and Israeli geopolitical risks. Mobileye also disclosed in its 2025 10-K that regional conflicts surrounding Israel and the mobilization of employees as reservists could be business risks.

Mobileye is not a company to ignore.

Rather, I think it is one of the favorites in mass-produced vehicle ADAS.
However, in my portfolio, I view it not as a core, but as an observation target or a satellite candidate when it becomes quite undervalued.

■ Zoox: Another L4 robotaxi option held by Amazon

Zoox cannot be overlooked either.

Zoox is an Amazon-owned robotaxi company and, like Waymo, is a player on the L4 robotaxi operation side. However, unlike Waymo or Tesla, Zoox has been building a dedicated robotaxi vehicle from the ground up.

Amazon announced its acquisition of Zoox in 2020, describing Zoox as a company that "designs autonomous ride-hailing vehicles from the ground up."

Tesla is expanding FSD from its existing mass-produced EV fleet.
Waymo is building up L4 operations with limited ODD based on existing vehicles.
Zoox is attempting to create urban robotaxis using dedicated vehicles that do not assume the presence of a steering wheel or pedals.

I think this is very Amazon-like.

Amazon is not just an e-commerce, AWS, and advertising company; it is also a company involved in logistics, warehouse automation, robotics, and last-mile delivery.

Zoox officially announced that it launched its robotaxi service in Las Vegas in September 2025.

I do not think Zoox will move Amazon's earnings immediately.
However, if viewed as an option for Physical AI and intra-city mobility, the significance of Zoox being within Amazon is substantial.

That said, at this point, it does not show the same level of operational track record, safety data, or depth of urban deployment as Waymo.

Therefore, I think it is appropriate to treat Zoox not as the centerpiece of Amazon's investment strategy, but as a hidden long-term option.

■ Cruise: Autonomous driving dies not from accidents, but from the lack of transparency after accidents

I would also like to touch upon Cruise.

Cruise was once a leading candidate for US robotaxis, alongside Waymo.
However, the current Cruise is no longer a primary candidate for an independent robotaxi.

In February 2025, GM announced that it had made Cruise a wholly-owned subsidiary and that, going forward, Cruise would work with GM to tackle autonomous driving technology and advanced driver assistance technology for consumer vehicles. In other words, at least for now, Cruise is closer to being an entity absorbed into GM's ADAS/personal autonomous vehicle side rather than an independent robotaxi frontrunner.

What is important here is not to dismiss Cruise's retreat as a mere defeat.

The biggest lesson Cruise left behind is that in autonomous driving, it is not just the "accident itself" but the "transparency after the accident" that determines business value.

After the October 2023 accident, Cruise faced heavy criticism regarding its disclosure of information to regulatory authorities. In September 2024, the NHTSA issued a Consent Order stating that several of Cruise's accident reports were incomplete, pointing out that it had omitted critical information from its reports, particularly regarding the behavior after the accident where a pedestrian was dragged.

Autonomous driving will always carry the risk of accidents.

That is precisely why it must be evaluated not only on technology but also on safety culture, record-keeping, disclosure, trust with regulatory authorities, and post-accident operations.

The lesson from Cruise is heavy.
The robotaxi business cannot be sustained by the accuracy of AI models alone. If trust is lost from society, the business will stop even if the technology remains.

■ Chinese players: Baidu and Pony.ai are black boxes, but cannot be ignored

Next, let's look at the Chinese players.

Transparency of information is difficult here.
Chinese players like Baidu, Pony.ai, and WeRide cannot be ignored in terms of technology or operations, but for investors, issues of governance, regulation, geopolitics, and data transparency are significant.

However, looking at the numbers alone, Baidu Apollo Go is quite large.

Baidu disclosed that in the fourth quarter of 2025, Apollo Go provided 3.4 million fully driverless operational rides, with weekly rides exceeding 300,000. It also explained that as of February 2026, the cumulative number of public rides had exceeded 20 million.

This is a scale that cannot be ignored.

Waymo is not the only frontrunner for L4 robotaxis.
Baidu has accumulated a considerable amount of real-world operational data in the Chinese market.

Looking at SEC filings, Pony.ai also discloses autonomous driving mileage, robotaxi fleet size, and driverless mileage, and has relationships with Toyota, GAC, and FAW. However, Pony itself also lists as a risk that its experience with large-scale commercialization is still limited.

It is difficult to assess the Chinese players.

It is not that their technology is weak.
On the contrary, with low-cost vehicles, relationships with governments and cities, rapid demonstrations, and a massive market, they have the potential to be quite strong.

On the other hand, for investors, the comparability of information is low.
It is difficult to know how to view safety data, accident response, regulations, overseas expansion, and geopolitical risks.

If it were me, I would be cautious about Chinese players as an investment target, but I would consider them essential for industry observation.

If Baidu or Pony.ai succeed in overseas expansion, it will put pressure on global robotaxi prices.
At that time, it will also affect Waymo, Tesla, Uber/Lyft, European OEMs, and Japanese OEMs.

Therefore, regardless of whether you buy them or not, you must keep an eye on the Chinese players.



■ Uber / Lyft: Not development companies, but important as a demand aggregation layer

Uber and Lyft are not companies developing core autonomous driving technology.

However, they are not irrelevant in the robotaxi era.
Rather, if multiple autonomous driving camps coexist, Uber and Lyft will become important as demand aggregation platforms.

Robotaxis cannot be sustained by technology alone.

Apps to gather passengers.
Dispatch algorithms.
Pricing.
Utilization rates.
Deadheading.
Payments.
Customer support.
City-specific operations.
Connections with vehicle supply companies.

These are all necessary.

Uber has partnered with Waymo and announced a framework to provide Waymo's fully autonomous vehicles on the Uber app in Austin and Atlanta. According to Uber's announcement, Uber will manage and dispatch Waymo's fully autonomous Jaguar I-PACE fleet, with plans to expand the fleet to hundreds of vehicles.

Furthermore, Uber, Wayve, and Nissan have announced preparations for a robotaxi pilot in Tokyo that will provide Nissan LEAFs equipped with Wayve AI Driver via Uber.

Lyft has also teamed up with Baidu and announced plans to deploy Baidu Apollo Go's RT6 in Germany and the UK starting in 2026.

In other words, Uber/Lyft are not the winners of technology.
However, they could be the winners of distribution, demand, and utilization rates if multiple technologies coexist.

This is an interesting investment point.

If Waymo can capture enough demand with its own app alone, Uber's value will be limited.
If Tesla builds a robotaxi network with its own app, Uber could be bypassed.

However, if Waymo, Baidu, Wayve, Pony, Zoox, May Mobility, and others proliferate, Uber/Lyft could become the "gateway to ride any robotaxi."

For me, Uber/Lyft are not autonomous driving technology stocks, but platform satellites.

■ Will standardization happen?

Here, I would like to think about standardization.

Like communication standards, memory cards, and EV charging sockets, there are worlds that eventually converge on a single standard. In EV charging, the Tesla-style NACS became the de facto standard in North America.

So, will autonomous driving also eventually converge on a single method?

I don't think it will become completely one.

The reason is that autonomous driving is not a single physical standard, but a combination of multiple layers.

Sensor configuration.
In-vehicle compute.
AI models.
Simulation.
Safety verification.
ODD design.
Remote assistance.
Operational management.
Regulatory compliance.
Liability entity.
Ride-hailing apps.
Data sovereignty.

These become autonomous driving when integrated.

Therefore, it is unlikely to be like NACS, where "this connector is the final one."

However, the technical philosophy is converging quite a bit.

It is in the following direction.

Multimodal input.
End-to-End.
Vision-Language-Action.
World Model.
Simulation verification.
Synthetic data.
Safety approval by ODD.
Remote assistance.
Parallel deployment of L2/L3 mass-produced vehicles and L4 robotaxis.

Looking at the moves of Tesla, Waymo, Wayve, NVIDIA, Qualcomm, and the Chinese players, this is not far off.

In other words, autonomous driving is not a winner-take-all for one company.
However, the philosophy of the architecture is beginning to converge.

I think this perspective is important for looking at the next three to five years.

■ What to watch in the next 3 years

I do not think the winners of autonomous driving will be completely decided between 2026 and 2028.

However, the patterns of the winning paths should become quite visible.

What you should look at is not the stories of each company.
It is whether the stories turn into reality.

For Tesla, watch whether FSD Supervised can truly advance from L2 advanced driver assistance to an L4 liability structure. You need to look at safety, intervention rates, regulatory approval, accident liability, insurance, and real-world Robotaxi operations, not just driving videos.

For Waymo, watch whether they can lower the cost of adding cities in multi-city deployments. Vehicle utilization, frequency of remote assistance, recovery when stopped, and handling of airports, highways, and bad weather will be important.

For Zoox, watch whether they can expand urban robotaxis in a different form than Waymo as an Amazon-owned dedicated robotaxi-type L4. While not yet the centerpiece within Amazon, it cannot be ignored as a long-term option for logistics, robotics, and Physical AI.

For Wayve, watch whether the E2E AI Driver truly spreads to OEMs through Nissan's next-generation ProPILOT and robotaxi pilots. However, you also need to look at the issue of data sovereignty. How will OEMs and AI vendors share driving data and learning loops? This could determine the speed of horizontal expansion.

For NVIDIA, watch whether Cosmos and Alpamayo move closer to becoming development foundations for OEMs, Tier 1s, and robotaxi companies, rather than just demos.

For Applied Intuition, watch whether they can enter the standard workflow for autonomous driving verification, simulation, and synthetic data generation through partnerships with NVIDIA and OEMs.

For Qualcomm, watch how far their initiatives with BMW, Bosch, and Wayve spread to mass-produced vehicles. Snapdragon Ride and Digital Chassis, in particular, are important as a realistic layer for near-future ADAS, SDV, and edge AI, rather than fully autonomous driving.

For Mobileye, watch how much SuperVision, Chauffeur, and Drive increase OEM adoption and expand from L2+ to consumer AV and further to Drive for robotaxis. In the realistic layer of mass-produced vehicle ADAS, the competition between Qualcomm and Mobileye is quite important.

For Baidu / Pony.ai, watch their scale expansion within China and overseas expansion. However, you need to look at not just technology, but also transparency, regulation, and geopolitical risks.

For Uber / Lyft, watch whether they can become a demand aggregation layer that bundles multiple robotaxi camps. The issue is whether they will be pushed into the self-app type like Waymo, Tesla, and Zoox.

I think the resolution of the autonomous driving map will increase significantly in these three years.

■ Which layer will I buy?

Finally, I will connect this to investment decisions.

If I were to include the autonomous driving theme in my portfolio, I would not simply buy the "company most likely to win in autonomous driving," but divide it by layer.

・Core: NVIDIA

I think NVIDIA should continue to be a core holding.

The reason is that there is no need to try to guess the winner of autonomous driving alone.

AI factories, GPUs, data centers, robotics, Physical AI, simulation, in-vehicle AI. There are multiple growth axes. It will be a plus if autonomous driving progresses, but the investment thesis for NVIDIA will not collapse even if autonomous driving is delayed.

If viewed as an autonomous driving theme, NVIDIA is not a "company that makes cars," but a company that holds the foundation of Physical AI.

This can be placed in the core.

And I already own NVIDIA.
In this sense, even if I don't own pure autonomous driving stocks, I have covered the foundation layer to some extent when autonomous driving evolves as Physical AI.
This is significant.

・Core candidate: Alphabet

Alphabet cannot be bought for Waymo alone.
However, Waymo is a huge option within Alphabet.

Search, YouTube, Cloud, TPU, Gemini, Android, Maps.
Waymo is added to that.

Google is a company that has held the gateway to the digital world, but if Waymo truly scales, it will also enter the mobility layer of the physical world.
This is very significant.

However, looking at Alphabet's total market capitalization, Waymo's contribution is still hard to see. So, rather than buying it just for Waymo, it is natural to evaluate Alphabet as a world OS company in the AI era and see Waymo as an option within it.

If it were me, I would place Alphabet as a core candidate.

・Core or semi-core: Qualcomm

Qualcomm is not a dream stock for fully autonomous driving.

However, it is important as a realistic layer for spreading ADAS/AD to mass-produced vehicles.

Snapdragon Ride, Digital Chassis, BMW, Bosch, Wayve.
Looking at this lineup, Qualcomm is not just a smartphone semiconductor company, but is going after the mass-production foundation for software-defined vehicles.

If NVIDIA is the foundation for future Physical AI, Qualcomm is the foundation for realistic mass-produced vehicle ADAS.

This role is quite significant.

And Qualcomm can be seen not just as an autonomous driving stock, but as a stock for edge AI as a whole.

AI first exploded in data centers.
Next, it will descend to PCs, smartphones, XR, cars, robots, and industrial equipment.

If you view this flow as edge AI, Qualcomm becomes a very realistic option.

I think Qualcomm's recent rise also includes expectations not just for a smartphone recovery, but for edge AI, in-vehicle AI, IoT, and SDVs.

If you are not trying to guess the final winner of autonomous driving, but are buying the mass-production layer of vehicles and the realistic layer of edge AI, Qualcomm is a sufficient candidate.

・Core candidate/long-term observation: Amazon

I would also like to touch upon Amazon here.

I am not buying Amazon as an autonomous driving stock.
However, it is also wrong to completely ignore Zoox.

Zoox is an L4 robotaxi company under Amazon and is another counter-axis to the Waymo-type urban robotaxi. Moreover, Zoox is designed from the start as a dedicated robotaxi vehicle, not a modification of an existing vehicle. This is different from both Tesla and Waymo.

Tesla aims for L2→L4 from a fleet of existing mass-produced vehicles.
Waymo accumulates L4 operational track records with limited ODD.
Zoox is attempting to create urban robotaxis using dedicated vehicles that do not assume a human driver from the start.

However, buying Amazon just for Zoox is wrong.

Amazon's main business is AWS, advertising, e-commerce, logistics, warehouse automation, robotics, and AI. It is natural to see Zoox as a long-term option for Physical AI within that.

In other words, I view Amazon not as a protagonist of autonomous driving, but as a massive collection of options for the Physical AI era.

In terms of portfolio, it is positioned not as a core candidate, but as a long-term observation stock where Zoox is added on top of the strength of existing businesses.

・Satellite candidate: Tesla

I feel Tesla is difficult.

It has the biggest scale option.
If it hits, it's huge.
If FSD transitions to L4 and Robotaxi goes into full swing, the stock valuation could change significantly again.

However, at this point, FSD Supervised is L2, and it cannot be said that L4 safety liability has been proven. In addition, Tesla involves EV sales, margins, competition, and political/brand risks, among other things, besides autonomous driving.

By my standard of "confirming an irreversible moat," Tesla is currently a satellite rather than a core.

It has the potential to be a big hit.
However, there are still things I need to see before I can comfortably place it as a confirmed irreversible moat.

I already have the foundation layer with NVIDIA.
Therefore, there is no need to force myself to chase Tesla.

I think I should judge it again at the stage where FSD's L4 transition, Robotaxi's real-world operation, safety, regulation, insurance, and monetization become visible.

・Satellite candidate: Uber / Lyft

Uber and Lyft are not autonomous driving development companies.

However, if multiple robotaxi companies coexist, they will have value as a demand aggregation layer.

Uber, in particular, is making contact with multiple camps such as Waymo, Wayve, and Baidu.
This is a strategy of incorporating potential winners into the platform rather than building autonomous driving itself.

However, if companies like Waymo, Tesla, and Zoox complete their services with their own apps, Uber's share will be limited.

So I view this not as a core, but as a platform-type satellite.

Observation targets: Wayve, Applied Intuition, Zoox, Mobileye, Pony.ai, Baidu

Wayve is unlisted but is the most important observation target.
Applied Intuition is also an important company behind verification and simulation.
Zoox is Amazon's autonomous driving option.
Mobileye is one of the frontrunners for mass-produced vehicle ADAS.
Pony.ai and Baidu cannot be ignored as Chinese players.

However, whether to buy them directly is a different issue.

Especially for Chinese players, even if their technology and scale are attractive, the problems of transparency, geopolitics, regulation, and investor protection are heavy.

■ Final conclusion

After organizing this, I thought about it again.

Autonomous driving is not just an additional feature of EVs.
This is the main battlefield of Physical AI.

AI is moving from the stage of writing text on a screen to the stage of seeing, judging, moving, and taking responsibility in the real world. The forefront of that is autonomous driving.

Therefore, this theme should not be taken lightly.
However, as an investor, you should not buy based on dreams alone.

Tesla is the biggest scale option.
Waymo is the realistic solution for L4 implementation.
Zoox is the dedicated robotaxi-type L4 option held by Amazon.
Wayve is the third pole of AI Drivers for OEMs.
NVIDIA is the Physical AI foundation.
Applied Intuition is a candidate for the OS behind verification and simulation.
Qualcomm is the realistic layer for edge AI and mass-produced vehicle ADAS.
Mobileye is one of the frontrunners for mass-produced ADAS.
Baidu/Pony are scale candidates for Chinese-style L4.
Uber/Lyft are demand aggregation platforms.

Among these, if you ask me if I need to buy a pure autonomous driving stock right now, the answer is no.

I already own NVIDIA.
In other words, I have covered the foundation layer to some extent when autonomous driving evolves as Physical AI.

Therefore, I don't need to force myself to hold pure robotaxi options like Tesla or Uber/Lyft right now.

I think the next thing to watch is Qualcomm.

Qualcomm is not just an autonomous driving stock.
However, considering the flow of AI descending to terminals, cars, IoT, XR, and robots, it is a very realistic edge AI stock. Within that, options for autonomous driving, ADAS, and SDVs also come in.

In other words, my conclusion at this point is this.

There is no need to force yourself to guess the final winner of autonomous driving from now.
First, have a foundation with NVIDIA.
Next, monitor Qualcomm as a realistic layer for edge AI and mass-produced vehicle ADAS.
View Alphabet as a world OS company that includes Waymo.
View Amazon as a long-term option for Physical AI that includes Zoox.
Keep Tesla as a satellite candidate until the bridge to L4 becomes visible.

The future of fully autonomous driving has not been decided yet.

However, I think it will become quite visible in the next three to five years.

Will "Tesla win in the end"?
Will "Waymo steadily expand L4"?
Will "Amazon's Zoox show presence with dedicated robotaxis"?
Will "AI Drivers like Wayve spread to OEMs"?
Will "NVIDIA or Qualcomm take the most delicious position as a foundation"?
Will "verification infrastructure like Applied Intuition standardize behind the scenes"?
Will "Chinese players go global with price destruction"?
Will "Uber or Lyft hold the gateway to robotaxis"?

I think this theme is quite important as the next expansion destination for the AI factory market.

I conclude at this point as follows.

Autonomous driving is not a winner-take-all for one company.
However, the technical philosophy is converging.
E2E, VLA, world model, simulation verification, ODD-specific L4, remote assistance.
This combination will be the next main battlefield for autonomous driving.

And as an investor, rather than trying to guess the final winner of dreams, first hold the company that holds the foundation.

In that sense, I have already covered it to a certain extent with NVIDIA.
On top of that, I will monitor Qualcomm as the next realistic option.

Thank you for reading to the end. If you found this helpful, I would appreciate it if you could like and follow.

(Disclaimer)This content is published for the purpose of organizing and recording the author's personal decision-making, etc., and involves hesitation, reconsideration, revision, and hypothesis updating. In addition, the information provided by this writing does not guarantee accuracy, etc. Therefore, since it may not necessarily be appropriate for readers, when investing, please recognize that investing in stocks involves significant risks and make investment decisions based on your own judgment and responsibility. Furthermore, this does not recommend investment in any financial products or individual stocks, nor does it recommend any investment methods.