The Infrastructure Pivot: Why AI Future Is Built in Orbit and Under the Sea

The current AI frenzy is moving from speculative software to the hard realities of physical infrastructure. While markets focus on stock prices and valuations, the real competitive advantage is shifting toward those who solve bottleneck problems like networking fabrics, orbital heat rejection, and sovereign compute. We are entering a phase where the biggest risk is not regulatory, but operational. For investors and leaders, the advantage lies in looking past the hype of model capabilities and focusing on the unglamorous, high-friction work of building the systems that AI requires to function at scale. Those who bet on the plumbing, such as networking, power, and thermal management, are positioning themselves for long-term dominance, while those chasing short-term model performance risk being left behind by physical constraints.

The Hidden Cost of Fast Innovation

While the market celebrates the rapid deployment of AI, the underlying system is hitting physical and regulatory walls. The US government move to restrict foreign access to Anthropic models, using dual-use authorities, signals that AI is no longer just a commercial product. It is a national security asset.

As Anthropic CEO Dario Amadei noted, there is a tension between the desire to deploy powerful models for defense and the reality of counterintelligence risks. The obvious solution of opening models to defenders to patch vulnerabilities is being slowed by the government because it creates a secondary, uncontrollable surface for adversaries.

I think all serious people here understand that there is real trade-offs here. We see a lot of sniping from people on Twitter and from other AI companies, you look at what they are saying and the inconsistency with what they are doing. It is not, they are not serious people.

-- Dario Amadei, CEO of Anthropic

This reveals a systems dynamic: the faster you release a powerful model to fix security, the more you increase the attack surface for everyone else. The system is responding by forcing a shift toward trusted partner regimes. This will likely fragment the global AI market, favoring companies that can navigate these geopolitical constraints over those that simply move fast.

Orbital Compute: The New Thermal Frontier

SpaceX push toward orbital data centers is a lesson in shifting from software-defined goals to physics-defined constraints. While the vision of hyperscale AI in space is compelling, the transcript highlights a major misconception: space is not just a cold environment for compute.

As Mackenzie Lister, former director at NASA Goddard Space Flight Center, explains, an orbital data center is a heat engine first and foremost. Without air for convection, heat rejection becomes a primary economic variable.

Space is not cold sometimes in the way that people think. And I would say that in orbit, a data center is a heat engine first and foremost and a compute platform second.

-- Mackenzie Lister, Principal Consultant at Paradot Services

This is where conventional wisdom fails. The infinite energy of the sun is useless if you cannot manage the thermal load or the radiation-tolerant hardware required to process it. The competitive advantage here is not just having the rocket; it is mastering the unstacking of the technology layer, moving from a world where one company does everything to a specialized infrastructure market where companies like Observable Space provide turnkey optical communication solutions.

The Networking Bottleneck

If compute is the engine, networking is the transmission. HPE acquisition of Juniper highlights a pivot away from the GPU-first mindset. Antonio Neri argues that as we scale, the GPUs themselves are not the primary constraint; the connectivity between them is.

This creates a feedback loop: as companies add more compute, they create more demand for networking fabrics. This is a delayed payoff investment. Most enterprises are currently struggling to prove the ROI of AI, but those who build the underlying networking foundation now will be the ones who capture the value when the digestive phase of AI adoption ends and productivity gains hit the bottom line.

Key Action Items

• Audit your AI ROI: Stop measuring success by pilots or proofs of concept. Shift focus to cost management and sales generation. If you cannot trace AI adoption to profit growth, you are likely over-investing in the wrong layer of the stack.
• Prioritize Infrastructure over Models: If you are building for scale, stop optimizing for the latest model. Focus on the networking and data architecture that allows your current models to communicate efficiently. The bottleneck is rarely the model; it is the fabric.
• Prepare for Sovereign Tech Constraints: Assume that the regulatory environment for AI will become more like the defense industry. If your business relies on cross-border data or model access, start diversifying your infrastructure to be sovereign-ready to avoid sudden compliance shocks.
• Look for the Unstacking Opportunities: Watch the space economy. As it matures, the companies that provide off-the-shelf infrastructure for orbital compute, such as laser communication or thermal management, will become the critical suppliers for the next generation of hyperscalers.
• Factor in Hidden Operational Costs: When evaluating new tech, specifically in high-performance or edge computing, calculate the cost of heat rejection and maintenance as primary variables. If the vendor has not accounted for the thermal and power reality of their solution, the project will likely fail at scale.