
...
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.
Attendees at the "World of Tomorrow" summit gathered in Edinburgh at the end of June at an inflection point. SpaceX, OpenAI and Anthropic are all expected to hit public markets within the same window, at trillion- and near-trillion-dollar valuations. There’s set to be a gargantuan cluster of IPO capital brought to market this year, which will be the first real test of whether frontier valuations can withstand the scrutiny of disclosed financials. The outcome will be decisive: either a defining IPO cycle for the AI era, or the moment the hype begins to fracture.
Yet it’s the same handful of giants supplying the compute, the cloud and much of the capital, often investing in the same firms that depend on their infrastructure. OpenAI, for example, will be working with Microsoft through 2032, while Anthropic has committed to Google Cloud for at least the next five years. Both rely on Nvidia’s hardware.
Read More on AI

...
Some see this circular, interlocking structure as a modern-day keiretsu, a tightly bound ecosystem of platform power. In this view, startups aren’t independent challenges, but satellites orbiting the pull of “the Magnificent 7”.
Yoav Zingher, founder and president of Launchpad Build AI, voiced unease about a future in which value consolidates into a handful of dominant platforms, warning that it could reproduce the social downsides of the last platform era. “If the future is going to be a few big platforms, I think China will win that game,” he said. “They have a controlled economy, collect as much data as they want, and direct capital into priority sectors. I don't think that's a good outcome.”
Others were less concerned, framing the system as a collaborative stack. Hyperscalers provide the tools, while value accrues to those who deploy them effectively. “It will just make the incumbent stronger,” said Steve Smoot, co-founder and managing partner at Lavrock Ventures. “The winners will be those who can isolate their advantages, define what’s proprietary, and build narrow, custom workflows that maximize the value of their data through AI.”
Build now, or bet big
If the structure of the market is contested, so too is the path to building within it. There was a split between two competing visions: narrow, immediately deployable automation versus the long-term promise of general-purpose robotics.
Jon Quick, CEO of Launchpad Build AI, favored highly specific, production-ready workflows embedded within existing systems. The priority, he suggested, is not to “reinvent everything,” but to automate discrete, high-value tasks and deliver utility now—rather than chase expansive, uncertain visions.
By contrast, capital continues to flow toward humanoid robotics, a trend that Ricky Horwitz, a partner at Exponential, the AI accelerator and early-stage venture arm of Cox Enterprises, laments. “Humanoids are easier to understand,” he said. “Companies see them as a way to remove people from the equation, without having to adjust factory lines.”
He pointed to a growing number of experiments designed to train these systems, including companies placing cameras on factory workers’ heads to record daily workflows. The goal is to build datasets capable of powering future general-purpose robots. “Our competitor here could be the humanoids,” he noted.
Another scenario would be reshoring simpler manufacturing closer to customers. As automation lowers labor dependency, simpler forms of production might move closer to end markets, with reshoring becoming an economic inevitability. “Making a change to your wire harness involves 12 weeks to go through that process, ship parts and test,” said Stephen Bennington, CEO and co-founder of Q5D. “Localizing that and putting it right next door means a prototype can be delivered the next day.”
As well as speeding up lead times, speakers cited the boost it could give to supply chain resilience, operational agility and productivity in local areas.
The talent-data trap
A skills mismatch emerged as a major brake on adoption. Companies lack people who can bridge engineering and AI-driven data workflows, and those hybrid profiles remain difficult to recruit and harder still to build internally.
Timothy Le, who leads the Physical AI Workbench at Nebius, described the shift from software engineering to “forward-deployed engineering”, whereby technical expertise is embedded directly within operations. “We’re focusing on upskilling engineers to work with AI tools, often in partnership with companies like Nvidia,” he said. “The goal is not deep expertise in everything, but enabling systems and agents to navigate tools and physical AI workflows effectively.”
Even then, cultural and cognitive divides persist. Getting the data right ultimately comes down to skills and there’s a crucial difference between hiring engineers or data scientists, said one technologist at a British multinational manufacturer. The U.K.’s long-time skills gap might be improving, but firms are still grappling with how to recruit and integrate the hybrid talent needed to fully exploit their data.

...
That shortfall is slowing what many see as the most urgent priority: data collection. Several speakers stressed that companies must begin capturing structured, workflow-level data now, or risk missing the next wave of value creation. Unlike software, this data cannot be retroactively generated. It is built over time through lived operations. “If you don’t have the data, then by the time that inflection point hits, you’re not going to be able to take advantage of it,” said Roy Raanani, founder of conversation intelligence platform Chorus.ai, who is now building Meticulous, the world’s first robotic lever espresso machine.
Yannis Georgas, vice president of artificial intelligence at Launchpad Build AI, echoed the point, noting that across sectors such as manufacturing, construction, and utilities, he has yet to see a firm that has fully “got its industrial data right.” The implication is stark: much of the sector remains stuck in foundational work and tribal knowledge. Engineering is digitised, but not data-centric, and as the narrative pushes toward advanced AI deployment, organizations are finding their data foundations too weak or simply absent.
In that context, the push for trusted data environments is the biggest hurdle yet to be crossed. Without the right skills to capture and structure data—and without the data itself—many firms risk being locked out of the gains AI promises to deliver.
View 0 Comments
Submit Correction
View Editorial & AI Guidelines