The Real-World AI Shift: How Smart Machines and Edge Computing Are Rewiring the Market

Artificial intelligence is stepping out of the digital realm and into the physical world. We’re entering a phase where intelligence is embedded directly into machines capable of seeing, reasoning, and acting in their environments. Think robots assembling cars, drones running delivery routes, and autonomous vehicles navigating city streets. Bank of America’s thematic investing team, spearheaded by analyst Martyn Briggs, views this transition as a trillion-dollar shift that is already unfolding right in front of us.

The Next Generation of AI Models

At CES 2026, Nvidia CEO Jensen Huang captured the industry’s momentum, declaring that the “ChatGPT moment for physical AI is here.” Unlike the first wave of AI, which was heavily focused on chatbots and digital content creation, this new era relies on multimodal world models. Trained on a mix of vision and action data, these systems do much more than process text; they predict, plan, and execute complex physical tasks autonomously.

Scaling Autonomy Across Industries

The real-world rollout is accelerating rapidly. Robotaxis are already operational across multiple cities, and advanced driver-assistance systems are on track for massive mass adoption in China by the end of the decade. As hardware costs drop and onboard computing power surges, the economic landscape of ride-hailing and freight is fundamentally shifting.

Meanwhile, humanoid robots are hitting the production lines to tackle industrial labor shortages and hazardous tasks. With over 50 companies currently developing these platforms, annual shipments are projected to skyrocket from the tens of thousands today into the millions over the next decade. Bank of America notes that the dominant players—highlighting infrastructure leaders like Nvidia, Tesla, Qualcomm, AMD, and Ambarella—won’t just be the ones building the hardware. The real value lies in controlling the sensors, actuators, and the underlying AI stack that ties it all together.

Solving the Hardware Bottleneck

However, pushing advanced AI into the physical world and highly secure environments introduces a massive challenge regarding computing power and data sovereignty. Relying entirely on cloud-based, GPU-heavy infrastructure isn’t always a viable option, especially for organizations fiercely protecting sensitive intellectual property. This specific bottleneck is driving a massive surge of interest in optimized AI platforms designed to run efficiently on accessible, localized hardware.

A Breakthrough in Chemistry Synthesis

A clear example of this market pivot is Redwood AI Corp. (CSE: AIRX), whose stock jumped over 27% to 2.78 EUR on February 26 following the announcement of a major performance upgrade to its proprietary chemistry synthesis prediction model. Designed to help researchers map out the absolute best pathways for manufacturing specific chemicals, the newly upgraded platform completely breaks away from the industry’s heavy reliance on specialized GPUs. Redwood’s data science team successfully optimized the model to run full, high-speed studies entirely on standard CPU infrastructure.

Empowering AI Sovereignty and Security

This technical update is a massive deal for biopharma organizations operating under strict data security protocols. By enabling efficient CPU-based execution, Redwood allows clients to deploy powerful AI applications directly on-premise, within specific geographic regions, or even on air-gapped systems that are completely disconnected from the internet for security reasons.

CEO Louis Dron emphasized that technical teams shouldn’t be forced to compromise between speed, control, and cost. The upgrade streamlines code management, keeps the system incredibly lean without sacrificing accuracy, and significantly slashes ongoing hosting expenses. Ultimately, it ensures that powerful predictive AI remains highly accessible to research teams, even as their parent organizations adopt increasingly strict internal guidelines regarding the use of massive, external cloud models.