World of Tomorrow brought founders, investors and government officials to Edinburgh to discuss what it will take to move AI from software promise into industrial reality.

...

Megan Carnegie

By Megan Carnegie

0

Share

Newsweek is a Trust Project member

See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.


On one of the hottest June days on record, global founders, ministers and investors gathered in Edinburgh for the “World of Tomorrow” summit. Inside the cool of The Caves, a former stables on Niddry Street, the conversation kept returning to a familiar tension in AI, robotics and advanced manufacturing: that software is racing ahead, while factories and supply chains still move too slowly.

Across the panels, speakers wrestled with the gap between model capability and real-world deployment, between regulation and innovation, and the need to build sovereign manufacturing and defense capacity. The result was a discussion about who can actually build, power and scale the systems that turn AI from code into machines, products and production lines.

Read More on AI

The “two-speed” AI economy: exponential software meets stubborn physics

For startups trying to build AI in the physical world, the problem is not just model progress but industrial speed. “There’s such high demand for AI, that the bottleneck to growth is in the upstream supply chain given the time needed to build factories and components,” said Roy Raanani, founder of conversation intelligence platform Chorus.ai, who’s now building Meticulous, the world’s first robotic lever espresso machine. Software is moving exponentially fast, but hardware, manufacturing and supply chains cannot keep up.

In a market that rewards velocity, one of the few ways for startups to get ahead is through long-term supply deals with infrastructure providers like Nvidia and Intel. “They can backstop them with financing, which creates a completely different set of capabilities,” said Raanani. Antioch, the New York startup building a cloud simulation platform for physical AI, works with Nebius, for example, to accelerate its business through chips, data centers and software layers, said co-founder and CEO Harry Mellsop.

Factory-floor systems were central to World of Tomorrow’s discussions about what it takes to move AI from software advances into real-world production.

...

There was a cross-panel consensus that digital twins and simulation compress design iteration dramatically and offer a promising route through the bottlenecks. “You can tweak, simulate and refine in a much tighter loop,” said Jon Quick, CEO of Launchpad Build AI. Running thousands of simulations before touching hardware is attractive—yet the models are still imperfect, and still under-trusted.

Hurtling into a trust, liability and policy vacuum

There’s another structural lag at play: AI is speeding up decision-making, but governance is failing to keep pace. For many in the room, the liability gap is a major concern. “When AI goes wrong, there’s huge uncertainty around who is responsible, especially when causation is difficult to establish in a black-box system with no clear explanation for how a decision was reached,” said Alex Sobolev, a partner at Orrick, Herrington & Sutcliffe.

Legal frameworks still draw heavily from traditional software models, which don’t map neatly onto probabilistic, opaque AI systems. Autonomous vehicles arose repeatedly as a cautionary example, including the Tesla Autopilot crash case, which illustrates many of the tensions: immense promise, unclear liability and the need to prove safety through simulation, testing and regulation before wider deployment.

Autonomous-vehicle technology came up at the summit as speakers discussed the challenge of assigning responsibility when AI systems operate in the physical world.

...

Regulation was another fault line. Several speakers debated the tension between protecting users and stifling innovation. Poorly designed rules, rigid labor laws and planning bottlenecks around data centers were all cited as risks that could throttle growth. Yannis Georgas, vice president of artificial intelligence at Launchpad Build AI, pointed out that while the old tech maxim claims the U.S. designs, China builds, and Europe regulates, physical AI completely shatters that old division of labor as both the U.S. and China drive advanced engineering and design natively. However, he argued that Europe’s regulatory power remains absolute: under the EU AI Act, AI models controlling physical machinery are categorized as "High-Risk" and must demonstrate compliance right at the border of entry—meaning systems shouldn’t be pushed into production without robust simulation and strong data foundations to verify safety first.

Stephen Miller, managing partner at Exponential, argued we’re still in a nascent phase where there simply isn’t the language to describe the risk. “Legislation doesn’t correlate with what’s powerful about AI.” In particular, this slows the path to progress in high-stakes domains like defense and intelligence. Juliane Gallina, partner at Lavrock Ventures and former Deputy Director of the CIA (Digital Innovation), saw this first hand. She said: “With LLMs and generative AI, it’s hard to prove to a human analyst if there isn’t provenance and lineage back to the source that trained the model.”

Power shifts down the stack

Across the day, it became clear that the center of gravity has moved away from apps and toward the underlying means of production: compute, energy, data centres, and advanced manufacturing. Raanani emphasized that: “Energy and data processing is the currency that limits growth.” Mellsop summed up the industrial logic as “not just one set of shovels, it’s shovels all the way down.”

Geopolitical pressure has certainly changed the investment mood. Steve Smoot, co-founder and managing partner at Lavrock Ventures, described feeling the need to call his largest investors to seek approval before their first defense investment three years ago, an example of the sensitivities of investing in the sector. The flow of investment and talent into defense and dual-use technologies has evolved since that time and investors have grown more comfortable making bets in the sector. Stephen Bennington, CEO and co-founder of Q5D, which uses robotics and software to automate the manufacture of wiring harnesses argued that “sovereign capability in manufacturing is key,” with supply chains themselves becoming part of that contest. The next phase of AI may not be set only by who builds the best models, but by who controls the energy, computer and manufacturing capacity needed to bring them into the physical world.

The next phase of AI may not be set only by who builds the best models, but by who controls the energy, compute and manufacturing capacity needed to bring them into the physical world.

View 0 Comments

Request Reprint & Licensing

Submit Correction

View Editorial & AI Guidelines