Eclipse's Lior Susan Wants Collaboration, Not Retreat, on AI

Eclipse chief executive Lior Susan told Bloomberg Tech on Sept. 11, 2026 that the US technology industry should pursue collaboration rather than retreat on advanced AI to avoid ceding its lead to China, and said the sector must better explain the benefits of data centers to local communities while physical AI becomes a driver of US economic growth.

The argument over how fast the United States should build advanced artificial intelligence has split into two camps that rarely talk to each other: one focused on the risks of increasingly capable models and the physical footprint of the data centers training them, the other focused on what happens if China gets there first. Lior Susan, chief executive of the venture firm Eclipse, used an appearance on Bloomberg Tech on Friday to argue that the choice is false — and that a retreat would be the more dangerous option.

Speaking with Ed Ludlow on Bloomberg Technology, Susan said the industry needs collaboration rather than withdrawal if it wants to keep the American technological lead intact. He also made two claims that sit slightly outside the usual venture-capital script: that the sector has done a poor job explaining what a data center actually does for the community that hosts it, and that physical AI — software embodied in machines that move, sense and manufacture — could become a major driver of US economic growth.

The political problem is local, not federal

Most of the public debate about AI risk is conducted at national scale: export controls, model evaluations, compute thresholds, frontier-safety commitments. Susan's point about local communities cuts against that framing. The binding constraint on the buildout is increasingly not capital or chips but permits, grid interconnection queues, water allocations and county-level politics. Those decisions are made by people who will never read a model card.

That is where the explanation gap bites. A data center is an unusual industrial neighbor: capital-intensive, tax-generative, and comparatively light on permanent jobs once construction crews leave. It consumes power and, depending on the cooling design, water. The construction phase is loud and temporary; the electricity demand is quiet and permanent. Industry advocates have tended to lead with abstractions about national competitiveness. Residents ask about utility bills and traffic. Susan's contention is that the sector has not bridged that distance, and that the cost of failing to is measured in projects that never get built.

The practical consequence is that opposition is becoming a sequencing problem rather than an ideological one. A rejected siting application does not stop AI development; it relocates it, often to another state or another country with a more accommodating grid. For anyone tracking capital expenditure guidance from the large cloud operators, the risk is not that the money disappears but that it moves slower than the announcements imply.

Why 'retreat' is the argument Susan is answering

The case for slowing down rests on the idea that capability is advancing faster than the ability to govern it. The counterargument Susan is making is a strategic one: unilateral deceleration in an open economy does not pause the technology, it changes who sets the defaults. Standards, model architectures, chip designs and the commercial norms attached to them tend to be written by whoever ships first and at scale.

Collaboration, in his framing, is the alternative to both a race with no guardrails and a retreat with no leverage. In practice that means shared safety research, common evaluation methods and industry coordination that does not require every firm to unilaterally disarm. It is a position that lets the buildout continue while giving the risk conversation somewhere to live — which is also, conveniently, the position most favorable to venture portfolios exposed to the buildout. That does not make it wrong, but readers should register the interest.

Physical AI is the growth claim worth testing

The third element of Susan's argument is the most economically consequential and the hardest to verify in advance. Software AI, however capable, largely reallocates activity within the information economy. Physical AI — robotics, autonomous industrial systems, machine-operated manufacturing and logistics — touches sectors where measured productivity growth has been stubbornly slow: construction, warehousing, fabrication, agriculture, defense manufacturing.

If that transfer works, it shows up in output per hour rather than in software revenue lines, and it shows up over years rather than quarters. It is also the part of the AI thesis most dependent on the same physical inputs as the data centers: electricity, industrial real estate, skilled trades and supply chains that are not trivially relocatable. That is the underappreciated link between Susan's two points. The community consent problem and the physical-AI growth story draw on the same scarce resources, and both are decided at the level of permits and power contracts.

What the market was saying while he spoke

Equities were not registering any anxiety about the debate. As of the last trade at 18:48:49 GMT on Friday, Sept. 11, 2026, the S&P 500 proxy (NYSEARCA: SPY) was at $765.29, up 0.98% from the prior close of $757.83, with a day range of $763.60 to $766.38. The Nasdaq 100 tracker (NASDAQ: QQQ) stood at $716.27, up 1.07% from $708.69, trading between $713.63 and $717.63. The Dow proxy (NYSEARCA: DIA) was at $526.11, up 1.03% from $520.75.

The tech-heavy Nasdaq 100 gauge was ahead of the broad market by roughly 0.09 percentage points on the day — an illustrative gap derived from the two published moves, and far too small to read as a verdict on anything. All three benchmarks were up around a point and trading near the top of their intraday ranges. Whatever the policy debate looks like on television, the tape on Friday reflected a market still comfortable funding the buildout.

What to watch from here

Three markers will tell you whether Susan's collaboration framing is gaining traction or being overtaken. The first is siting outcomes: county and state decisions on new data center campuses, and whether operators start pairing applications with concrete local commitments on power costs and infrastructure rather than national-security rhetoric. The second is whether coordinated safety work becomes an industry structure with teeth or stays a set of voluntary letters. The third is order books in physical AI — industrial robotics, autonomy stacks, automated manufacturing — because that is where the growth claim either materializes as revenue and productivity or remains a pitch.

Susan is arguing, in effect, that the US can carry the risk conversation and the buildout at the same time, and that the binding constraint is persuasion rather than technology. The harder test is not whether the argument is coherent. It is whether it works in a planning meeting.

Key facts

• Speaker: Lior Susan, CEO of venture firm Eclipse, on 'Bloomberg Tech' with Ed Ludlow, Sept. 11, 2026
• Core argument: Collaboration, not retreat, on advanced AI to avoid ceding the US lead to China
• Nasdaq 100 (QQQ): $716.27, +1.07%, as of 18:48:49 GMT Sept. 11, 2026
• S&P 500 (SPY): $765.29, +0.98% from a prior close of $757.83, same timestamp

Frequently asked questions

What did Lior Susan actually argue?

Susan, chief executive of the venture firm Eclipse, told Bloomberg Tech on Sept. 11, 2026 that the technology industry should pursue collaboration rather than retreat on advanced AI, on the grounds that pulling back would risk ceding the US technological lead to China. He also said the industry must explain data center benefits better to local communities.

What is 'physical AI'?

Physical AI refers to artificial intelligence embodied in machines that sense and act in the real world — robotics, autonomous vehicles and industrial systems, automated manufacturing and logistics. Susan said it could become a major driver of US economic growth, because it reaches sectors such as construction, warehousing and fabrication where productivity gains have been slow.

Why do data centers face local opposition?

Data centers are capital-intensive but employ relatively few permanent staff once construction ends, while drawing substantial electricity and, depending on cooling design, water. Residents typically raise utility costs, grid capacity, noise and traffic. Susan's argument is that the industry has not adequately explained the local benefits, which slows or blocks siting decisions.

How were markets trading when Susan spoke?

As of the last trade at 18:48:49 GMT on Sept. 11, 2026, the S&P 500 proxy SPY was at $765.29, up 0.98%; the Nasdaq 100 tracker QQQ was at $716.27, up 1.07%; and the Dow proxy DIA was at $526.11, up 1.03%. All three were near the upper end of their intraday ranges.

Is Eclipse a publicly traded company?

No. Eclipse is a private venture capital firm, so there is no exchange ticker or share price attached to it. Its interest in the AI buildout is through portfolio investments, which is worth keeping in mind when weighing its chief executive's public advocacy for continued expansion rather than a slowdown.

What should investors watch next on this theme?

Three things: local and state siting decisions on new data center campuses and whether operators attach concrete community commitments; whether industry safety coordination becomes a structure with enforcement rather than voluntary pledges; and order flow in industrial robotics and autonomy, which is where the physical-AI growth claim gets tested against revenue.

Sources

• Balancing AI Risks With the Race to Stay Ahead of China — Bloomberg Technology

Photo: Prosper Buka · Pexels Licence — source