AI Leaders Warn the Real Constraint Is No Longer Compute—It's the Physical World

A panel of senior executives from semiconductor manufacturing, cloud computing, autonomous systems, and artificial intelligence research gathered at the Milken Institute Global Conference in Beverly Hills to assess where the industry faces its most pressing constraints. The session, titled 'Designing at the Speed of AI: Supply and Demand in a Hypergrowth Era,' featured Christophe Fouquet, chief executive of ASML; Francis deSouza, chief operating officer of Google Cloud; Qasar Younis, chief executive of Applied Intuition; Dimitry Shevelenko, chief business officer of Perplexity; and Eve Bodnia, founder of Logical Intelligence. Across the discussion, a consistent theme emerged: the supply of raw computational power, once the dominant scarcity, is now outpacing the physical infrastructure, energy systems, and real-world deployment mechanisms needed to put that power to work.

The Details

ASML occupies a critical position in the global semiconductor supply chain as the leading manufacturer of extreme ultraviolet lithography systems. Christophe Fouquet told the panel that ASML will not be the chip industry's bottleneck, according to panel coverage. That assurance is significant given the company's central role in advanced chipmaking equipment. Even so, demand continues to outstrip supply in the near term, Reuters reported separately in advance of the conference.

On the cloud infrastructure front, Google Cloud has posted growth rates that place it among the fastest-expanding divisions in the technology sector. The unit surpassed $20 billion in quarterly revenue for the first time in the first quarter of 2026, a 63 percent increase compared with the same period a year prior, according to Alphabet's filings with the Securities and Exchange Commission. The financial momentum is underscored by a backlog of committed revenue that swelled to more than $460 billion as of the first quarter of 2026, up from $240 billion in the fourth quarter of 2025, Alphabet Chief Financial Officer Anat Ashkenazi disclosed in the company's 10-Q filing.

As demand for AI infrastructure strains terrestrial power grids and cooling capacity, Google is exploring an unconventional expansion path. Under an initiative called Project Suncatcher, Google Research is designing solar-powered satellite constellations that would carry Google Tensor Processing Units and communicate via free-space optical links. The concept, outlined in a Google Research blog post, envisions arrays of satellites operating as orbital data centers. Google has partnered with Planet Labs on prototype satellites, with a targeted launch by early 2027.

Applied Intuition, which carries an approximate valuation of $15 billion and has raised more than $1 billion while maintaining profitability through long-term enterprise contracts, is staking its position in what it calls 'physical AI.' The company launched Applied Edge, billing it as the first mobile operations center for autonomous systems. The platform is aimed at sectors including vehicles, defense, mining, and agriculture.

Perplexity, known primarily for its AI-native search interface, is pivoting toward autonomous task execution. The company introduced Perplexity Computer alongside Comet Enterprise, a $200-per-month subscription tier. The platform orchestrates 19 distinct AI systems and is positioned as a multi-model AI agent platform rather than a traditional search engine.

Logical Intelligence, co-founded by Eve Bodnia, a former colleague of Yann LeCun at Meta, is pursuing a fundamentally different technical approach. The company is developing energy-based models, or EBMs, as an alternative to the autoregressive large language models that currently dominate the industry. Logical Intelligence claims that EBMs solve tasks more precisely while consuming significantly less energy.

Context

The Milken Institute Global Conference is an annual meeting that draws leaders from finance, business, and technology. The 2026 panel reflected a growing recognition among industry architects that software and hardware advances have raced ahead of the physical systems required to deploy them at scale.

Qasar Younis captured the shift in direct terms during the panel session. 'The bottleneck is no longer compute, it's the physical world,' he said, according to coverage of the event.

Francis deSouza pointed to Google's orbital research program as a direct response to that concern. 'If Earth can't power and cool all these AI data centers, why not put them in orbit?' he said.

Energy-based models represent an architectural departure from the token-prediction approach used by contemporary large language models. Rather than generating text by predicting the next token in a sequence, EBMs use energy functions to represent goals and constraints. Proponents, including Logical Intelligence, argue that this method can achieve greater precision while demanding less computational power.

The panel's composition highlighted the breadth of sectors now grappling with AI deployment at scale. ASML's lithography systems enable the chips that power data centers. Google Cloud provides the infrastructure layer. Applied Intuition focuses on embedding AI into vehicles and heavy machinery. Perplexity is redefining how users interact with multiple AI systems. Logical Intelligence is attempting to redesign the underlying model architecture itself.

What's Next

Several of the initiatives discussed at the panel carry near-term milestones. Google's partnership with Planet Labs on Project Suncatcher is targeting prototype satellite launches by early 2027, a date that will test the feasibility of orbital AI infrastructure. If successful, the approach could provide an alternative to the land, power, and cooling constraints that currently limit terrestrial data center expansion.

Applied Intuition's Applied Edge product will face the challenge of proving that mobile operations centers can operate reliably across defense, mining, and agricultural environments, sectors that have historically adopted automation more slowly than consumer-facing transportation.

Perplexity's shift from search to autonomous task execution via Perplexity Computer and Comet Enterprise enters a competitive arena where enterprise willingness to pay $200 per month for multi-model orchestration remains unproven at scale.

For Logical Intelligence, the challenge is technical validation. Energy-based models must demonstrate not only energy efficiency but also performance parity or superiority against established large language models in real-world applications.

ASML's assurance that it will not become the industry's bottleneck, combined with continued demand pressure, signals that chip supply may ease even as competition for physical deployment infrastructure intensifies.