The Battleground for Autonomous Driving Has Moved to Software | How In-Vehicle OS, SDV, and AI Agents Are Reshaping Mobility in 2026 [SusHi Tech Tokyo 2026]

Hello, this is Hamamoto from TIMEWELL. "Google has been working on autonomous driving for fifteen years. Why is it still not really deployed at scale? The answer is simple — the technical approach itself was wrong." That quiet line from Qasar Younis, CEO of Applied Intuition, set the tone of the autonomous-driving panel at SusHi Tech Tokyo 2026. Of every discussion I have heard in the past several years, this was the one that most clearly exposed the tectonic shift in the industrial structure.

The moderator was Rinko Toshima, editorial committee member at the Yomiuri Shimbun. The speakers were Qasar Younis (CEO, Applied Intuition), Kazumasa Sugimoto from Nissan Motor, and Hiroshi Sato, leader of the commercial vehicle business at Isuzu Motors. Three top players from Japan and the US, speaking openly about where autonomous driving stands today and what comes next.

Summary of this article

• The battlefield in autonomous driving has fully shifted from "hardware performance" to "in-vehicle OS x End-to-End AI x data pipeline."
• The keyword is 2027. Nissan is concentrating Level 2 mass production and robotaxi commercialization there. Isuzu is targeting Level 4 driverless trucks and buses in the same year.
• Applied Intuition is expanding from 18 locations toward more than 20, with a valuation north of USD 6 billion. The structure is heading toward a "winner takes most" platform race in which only a handful of companies survive worldwide.
• Younis's "Physical AI sovereignty" argument — countries that cannot design and operate AI that moves in the physical world inside their own borders will pay an economic and defense price.
• What Japan's auto industry needs is to drop "in-house only" and create a culture in which hardware strength and Silicon Valley-style software development can coexist inside the same organization.

---

About the event — SusHi Tech as the intersection of mobility and AI

SusHi Tech Tokyo 2026 is one of Asia's largest innovation conferences, running April 27 to 29, 2026 at Tokyo Big Sight. Its four focus themes are AI, robotics, resilience, and entertainment, and autonomous driving is treated as a flagship "AI x robotics" domain with a substantial session slot.

At TIMEWELL I work daily with new-business leaders and intrapreneurs. Through that work, I feel the change in mobility hitting management decisions company by company. After this session, my view of Japan's auto industry shifted. The reach and depth of what was said on stage made that inevitable.

For the keynote layer of the same event, I have laid out the city- and nation-strategy context in SusHi Tech Tokyo 2026 Keynote.

Reading the present in numbers — 2027 as the inflection year

The first thing all three speakers shared was the year "2027." Nissan plans to mass-produce and commercialize Level 2 autonomous driving by 2027 and to apply the same software controller to its robotaxis. Isuzu stated explicitly that it will introduce Level 4 driverless trucks and buses before 2027. Applied Intuition is on a high-growth path, expanding from its current 18 locations to more than 20 within 2026.

The most striking number came next. By 2030, more than 30 percent of Japan's freight will become impossible to transport — Sato's estimate. The driver shortage is no longer a "problem"; it has reached the level of an "industrial-survival crisis." The so-called "2024 problem" (driver work-hour regulation) has already strained logistics on the ground, and if it is left unaddressed until 2030, the bloodstream of Japan's economy will narrow.

Player	Key milestones in 2026-2027
Nissan Motor	Mass production of NSOP-based Level 2; common stack with robotaxi
Isuzu Motors	Level 4 driverless trucks and buses (in partnership with Tier IV / Applied Intuition)
Applied Intuition	Expansion from 18 to 20+ locations; valuation above USD 6 billion
Japan logistics	Without action, more than 30 percent of freight will be "unmovable" by 2030

The core of Younis at Applied Intuition — Transformer changed everything

Younis's diagnosis was sharp. "The Transformer architecture of the past five years is the key to pushing autonomous driving into the production stage." The Transformer-based end-to-end learning behind ChatGPT and Gemini has now reached a level where it can be applied to autonomous driving as well.

That means a fundamental shift away from rule-based autonomous driving toward data-driven, learning-based autonomous driving. The end-to-end approach — learning the full path from camera input to driving action — was implemented first by Tesla, picked up afterward by Xpeng and Huawei in China, and is now being adopted by Waymo as well. It is becoming the new industry standard.

The other important point was the structural similarity to the chip industry. Worldwide, only a handful of companies can manufacture chips. Why? Because it requires enormous capital, advanced technology, and vast amounts of data. Autonomous driving will follow the same structure — only a handful of platform providers worldwide will survive. That was Younis's view.

Applied Intuition's move from 18 to more than 20 locations within 2026 is a deliberate front-running of that "winner takes most" structure. Its partnerships with Toyota, Volkswagen, Stellantis, and Isuzu fit the same logic. The 2024 funding round put its valuation above USD 6 billion, and in autonomous-driving simulation and toolchain it now sits in the strongest position in the market.

Younis's framing also helps explain why Tesla's lead in End-to-End learning is so hard to attack. Tesla is not just shipping cars; it is operating the world's largest moving sensor network, generating training data at a rate no other automaker can match. Each Tesla on the road is a labeled data-collection device for edge cases — heavy rain in Bangalore, snowdrift on the Hokkaido coast, a parade route in Lisbon. That data feeds back into the model, which is then redeployed over the air. It is the same flywheel that gave Google search its lead twenty years ago, transposed onto physical roads. Catching up is not impossible, but it requires either a comparable fleet (Toyota and Volkswagen have it) or access to a shared simulation environment that approximates fleet-scale variety. Applied Intuition is selling the second option, which is why every legacy OEM is on the customer list.

Nissan — The architectural bet called NSOP

What Sugimoto from Nissan brought on stage was the Nissan Scalable Open Software Platform (NSOP). It is an ambitious project that marks a major turn in the history of automotive development.

In the traditional development model, autonomous driving, cockpit, and body each lived in siloed domains. For autonomous driving, however, a distributed approach actually delivers more development flexibility. The most important question becomes a strategic one: "What do we integrate, and what do we keep distributed?" Sugimoto put it cleanly: "In-vehicle data must be integrated and seamlessly connected to the cloud. That is what becomes the foundation of AI learning."

What stayed with me was a candid line he added: "I look at the fact that companies like Applied Intuition concentrate the best engineers, and as a Japanese person I feel both envy and a sense of crisis. With our current Japan team alone, we cannot build something like this."

I take that as the honest voice of corporate Japan. The fact that this kind of statement can be said openly on a panel is itself proof that Japan is starting to turn its rudder away from "in-house only" and toward "partnership strategy." Toyota, Subaru, Mazda — each is pursuing its own OS while accelerating moves to bring in outside partners. The structural shift across the industry is happening quietly, but unmistakably.

The architectural question Sugimoto kept circling back to is what I would call the "vertical-integration paradox" of automotive software. If you integrate too much, you build a monolith that no one inside or outside the company can iterate on quickly. If you distribute too aggressively, the latency and safety-critical paths break down because boundary contracts between components leak. NSOP's bet is to centralize the data layer (one canonical bus for sensor and event data, one canonical link to the cloud) while keeping the application layer modular, so that perception, planning, and HMI can be improved by separate teams — or by external partners — without each release blocking the others. That is the same separation-of-concerns logic that Android adopted for the smartphone, and it is the closest thing the automotive industry has to a viable answer to Tesla's vertically integrated stack.

Isuzu — Commercial trucks as the answer to the driver shortage

Sato's story from Isuzu had the strongest commercial impact for me personally. The future of Japan's logistics industry depends almost directly on how fast autonomous driving can be deployed.

Isuzu has been deploying autonomous-driving truck and bus systems in society since 2017, and disclosed plans to introduce Level 4 driverless systems before 2027 in joint development with Tier IV and Applied Intuition. He also described the strategy of plugging existing telematics networks — the "Gate Server" platform, which has been operating since the 1990s, and the vehicle-operations service "Prism" — into the new architecture for the AI era.

The deepest issue Sato pointed to was the redefinition of quality assurance. In conventional automotive development, engineers could guarantee 100 percent of the logic they designed and ship that to the customer. With AI in the loop, that frame collapses at its root. AI models do predictive generation; their outputs cannot be fully reproduced or evaluated. It is impossible to validate every pattern through physical testing — evaluation in digital environments becomes mandatory.

This "redefinition of quality assurance" is not a problem unique to commercial-vehicle makers. It is the fundamental challenge any manufacturer hits when embedding AI into a product. How Japan translates its "quality myth" into the AI era — that is what the next ten years of Japanese manufacturing turn on.

Isuzu's response to that challenge is also worth attention. Sato described a multi-layer assurance model: a deterministic safety controller wraps the AI driving stack and intervenes whenever the AI's predicted action exceeds bounded operating envelopes, while millions of synthetic miles are run nightly in simulation against curated edge-case libraries. The pairing of "deterministic guard rails" with "stochastic AI core" is becoming the de facto pattern across the industry, and it is the only realistic way to ship an AI system into safety-critical hardware without waiting decades for a formal proof. Japanese manufacturers, which historically prefer fully deterministic systems, will need to make peace with the fact that the new norm is "stochastic core plus deterministic guard rails," and that the engineering organization must become fluent in both modes at once.

Younis on "Physical AI sovereignty" — the highlight of the session

The line that hit me hardest came at the very end, from Younis.

"Conventional software (Google, Facebook, etc.) crossed borders easily. Mobile apps localized a little. But Physical AI — autonomous driving, mining automation, robotics — becomes a matter of national security."

In other words, a country that cannot design, manufacture, and operate physical-world AI inside its own borders will pay a fatal price both economically and defensively. Each country must build its own independent engineering capability, infrastructure, data centers, and regulatory frameworks.

This is not a business statement. It is an explicitly national-strategy argument for "AI sovereignty." As I listened, I remembered the urgency I felt at Panasonic during my time in Silicon Valley. Japan has already lost software sovereignty. If we walk the same road in Physical AI, our long-term competitiveness will be hollowed out.

The same word — "sovereignty" — keeps appearing in the food context too. Reading European Agri-Food Innovation Report alongside this piece sketches, in three dimensions, the contour of "the sovereign domains Japan must protect now."

The safety question — humans demand "superhuman safety" from AI

Another important point Younis raised was social acceptance. Statistically and mathematically, autonomous driving is already safer than human driving. Yet humans demand a far higher bar from autonomous driving than from human drivers.

Hundreds of people every day, thousands worldwide, die in traffic accidents. If autonomous driving can reduce that, deployment is morally justified — that argument is short and strong. Citizens in Silicon Valley are already riding Waymo and Tesla autonomous vehicles as part of daily life. "As more people accumulate the experience of riding in one, social acceptance moves naturally," Younis said, with data on his side.

My take — How TIMEWELL sees "careers in the Physical AI era"

At TIMEWELL I work with new-business leaders and intrapreneurs, and over the past one or two years the questions coming from people running new-business projects at automakers and parts suppliers have clearly changed in quality. Earlier, the question was "what can we do with AI?" Now it has become a concrete question: "Inside this NSOP-scale platform consolidation, where does my business stand?"

The strength of Japan's auto industry is precise hardware engineering, supply chains, and on-the-ground craftsmanship. None of that disappears in the Physical AI era. In fact, when bound to a software platform, that value should multiply. The real challenge is fundamentally an organizational one: can "field-level hardware knowledge" and "Silicon Valley-style software development" coexist inside the same organization?

Learning from Sugimoto's candor — the courage to admit "Japan's crisis"

We have arrived at an era where Sugimoto can say on a panel that "we cannot build this with our Japan team alone." That is a major shift. Ten years ago, that statement would not have been allowed inside the company. Acknowledging your own weakness and turning the rudder toward partnerships — that is, in my view, the cultural shift Japan's large enterprises must make to survive.

I want the intrapreneurs TIMEWELL supports to find the same courage. Honestly grasp your own strengths and weaknesses, and complement the weaknesses with outside force. Only the companies that break the spell of "in-house only" will be the winners of the Physical AI era.

There is also a talent dimension that deserves explicit attention. The engineers Younis described as "concentrated at Applied Intuition" are not magical hires; they are the output of a hiring pipeline that pays at the top of the global market, ships software in production within weeks of joining, and rewards individual judgment instead of escalation. Every one of those characteristics runs against the grain of how most Japanese OEMs hire and manage engineers today. Closing the gap is less a recruiting problem than a re-design of the engineering operating model: how performance is reviewed, how decisions are made on a Pull Request, and how disagreements are resolved without flowing up to the section chief. Companies that solve this organizational design question — not the ones that simply raise salaries — will be the ones able to retain the kind of engineers who can ship Physical AI at scale.

Closing — Three messages from three leaders to Japan

I left the session carrying three conclusions. First, 2027 is the year (Level 2 mass production and Level 4 commercialization will become reality). Second, Physical AI sovereignty is national strategy, and Japan urgently needs to build its own independent engineering capability. Third, AI adoption must be accelerated at every layer — corporate leaders embedding AI in operations, manufacturers embedding AI in products, consumers embedding AI in daily life.

The message Younis sent last to Japan stayed with me: "Rapidly embedding AI across every layer of society is the key to Japan's long-term success." Building infrastructure for challengers means making sure that this embedding is not the privilege of "a handful of tech companies" but the default for "every working professional." Three hours that made TIMEWELL's job a bit clearer.

Key points in bullets:

• 2026-2027 is a "compressed inflection," with in-vehicle OS, End-to-End AI, and Level 4 logistics rising in parallel.
• Applied Intuition's expansion pace shows that the autonomous-driving platform is heading toward "a handful of global winners taking most."
• Nissan's NSOP turns on the strategic call of "what to integrate vs. what to keep distributed"; for Isuzu, the largest issue is the redefinition of quality assurance.
• "Physical AI sovereignty" is a national-strategy-level agenda. Japan must avoid repeating the loss of software sovereignty.
• The decisive factor will be an organizational culture that drops "in-house only" and combines hardware strength with external software partners.

---

TIMEWELL also offers individual consultations through our AI consulting service WARP. You can start with a 30-minute online consultation.

---

References

[^1]: YouTube. "The Battleground for Autonomous Driving Shifts to Software." https://www.youtube.com/watch?v=NAL7qYS5OGc [^2]: Applied Intuition. https://www.appliedintuition.com/ [^3]: Nissan Motor. "Autonomous Driving Technology." https://www.nissan-global.com/ [^4]: Isuzu Motors. "Autonomous Driving Commercialization." https://www.isuzu.co.jp/