Autonomous driving boom with open vision and motion models, 81 startups challenge Waymo for under $1 million

The fever of investment in self-driving vehicles has returned, surpassing the 2016 boom that drained billions of dollars in complex early trials. This time, the fundamental shift is directly tied to the technical architecture of the systems themselves. In 2026, the early-stage fund Tracx VC received proposals from 81 new startups in the autonomous-driving sector, a figure that is about two-thirds higher than last year and exceeds the peak deal volume the sector saw a decade earlier.

The decisive transition from first-generation to second-generation autonomous drivingBonith Mirova, a partner at Tracx VC, explains that the autonomous-vehicle sector has been searching for years for its coveted end-to-end architecture, a fully integrated software stack from start to finish. Today, foundational models have provided the technical framework that makes these integrated systems practically deployable. These foundational models are distinguished by their strong ability to generalise across diverse operating environments and conditions, which makes the rollout of autonomous vehicles much faster and far cheaper than earlier systems that were constrained by rigid codebases and pre-defined routes.

This technological shift has lowered entry barriers to unprecedented levels, thanks to open-source models such as Nvidia’s “Alpaca-Yo” and the multi-task “CoE” models that combine vision, language and motion. The fund believes that startups can now launch their vehicles and begin pilot operations within a few months with funding of less than a million dollars, a figure that was deemed impossible in the 2016 cycle when trials cost tens of millions. The most striking surprise has been the emergence of teams offering autonomous-taxi services to compete with giants such as Waymo, expanding beyond the traditional applications limited to warehouses, mines, farms and construction by focusing on specific geographic routes and specialised operating domains.

Manufacturing and infrastructure deals run parallel to a wave of lightweight softwareThe market shows a dual trend that combines the lightness of foundational models with the scale of industrial capital. Pedrock Robotics raised $600 million in a financing round led by Capital G and Emergence Capital, with participation from Nvidia and a group of major funds, valuing the company at $4.1 billion. In the commercial shipping sector, Codiac AI signed a binding contract to supply 75 autonomous trucks to DTL Transport for highway operation by 2028. At the same time, Toyota and Waymo sealed a manufacturing partnership to produce electric vehicles ready for autonomous driving in Japan and distribute them internationally.

What does this shift mean for those who manage transport fleets or develop logistics systems in the Gulf, Egypt and the Levant? The fundamental change here is procedural and economic, affecting decisions and direct budgets. Previously, reliance required waiting for major tech firms to map cities and provide costly sensor equipment that demanded sovereign support or multibillion-dollar investments. Now, with open-source vision-and-action models available, technical teams and warehouse or port operators in cities such as Riyadh, Cairo and Dubai can train specialised navigation and motion systems and test them locally at modest cost, with flexibility that adapts to local temperatures and traffic patterns. The obstacle is no longer limited to possessing a massive budget for hardware and mapping; the priority has shifted to fine-tuning foundational models on actual road data and operating them within the available regulatory scopes.