The Decade of Full-Stack Capture

← Back to Insights

How Companies and Nations Are Re-Architecting Power

We are entering the decade of full-stack capture, a period in which the most consequential actors in tech and geopolitics are racing to control the foundational layers of tomorrow's economy: energy, compute, data, intelligence, and physical infrastructure.

This is no longer vertical integration in the classical sense. It is auto-sufficiency at scale, the deliberate pursuit of control over every critical input required to build and deploy intelligence systems. The largest tech companies are not optimizing for asset-light models; they are optimizing for strategic independence from suppliers, grids, and external constraints that could bottleneck their ambitions.

What they are securing:

  • Energy at the source: owning or co-locating power generation rather than competing for grid capacity
  • Custom compute hardware: silicon designed for specific workloads, bypassing the margins of third-party chipmakers
  • Proprietary data ecosystems: feedback loops that continuously fuel and improve their intelligence systems
  • Distribution and infrastructure platforms: direct connections to users that cannot be intermediated

The Evidence Is Already Here

SpaceX Acquires xAI: The Most Ambitious Vertical Integration in History

In February 2026, SpaceX announced its acquisition of xAI for $250 billion, creating a combined entity valued at approximately $1.25 trillion—the world's most valuable private company. Elon Musk described the merger as creating "the most ambitious, vertically-integrated innovation engine on (and off) Earth, encompassing AI, rockets, space-based internet, direct-to-mobile communications, and the leading platform for real-time information."

This is not about owning a social media platform. It is about integrating:

  • Data (X's real-time information streams)
  • Compute (xAI's Grok models and infrastructure)
  • Connectivity (Starlink's satellite network)
  • Launch capacity (SpaceX's rockets and Starship)
  • A path to space-based data centers (the only scalable solution to AI's energy constraints, per Musk)

SpaceX has already filed with the FCC to launch up to 1 million solar-powered satellites engineered as orbital data centers. Musk's thesis: terrestrial grids cannot sustain the exponential compute demands of frontier AI. Space offers near-constant solar energy, no land or freshwater constraints, and freedom from grid bottlenecks.

Alphabet Buys Intersect Power: Vertical Integration to the Grid Level

In December 2025, Alphabet agreed to acquire Intersect Power for $4.75 billion, a developer of co-located data center and energy infrastructure. This is not portfolio diversification. It is a direct response to what Intersect's CEO called the fundamental constraint on AI: "AI today is stuck behind one of the slowest, oldest industries… there isn't enough electricity for all the racks full of GPUs."

The acquisition gives Google:

  • Multiple gigawatts of energy and data center projects under development or construction
  • The ability to build power generation in lockstep with compute deployment, bypassing years-long grid interconnection queues
  • A proven model (the Haskell County, Texas site) for co-locating solar, storage, flexible gas, and data centers

When Google wants a new AI facility in 2027, it can now orchestrate the build of the matching power plant on its own timetable. This is vertical integration at a level that redefines what "tech company" means.

The Stargate Project: Infrastructure at Civilizational Scale

OpenAI, Oracle, and SoftBank have committed up to $500 billion over four years through the Stargate Project to build the world's largest AI infrastructure platform. The target: 10 gigawatts of data center capacity, enough to power approximately 7.5 million homes.

As of late 2025, Stargate had already secured nearly 7 gigawatts of planned capacity and over $400b in committed investment, putting the project ahead of schedule. New sites span Texas, New Mexico, and the Midwest, with over 2 million chips to be deployed.

This is not a bet on better models. It is a bet on owning the physical substrate on which all future intelligence runs.

Custom Silicon: The Great Decoupling from Nvidia

The hyperscalers are no longer content to pay what analysts call the "Nvidia tax"... the 70–80% gross margins commanded by the dominant chipmaker. By late 2025:

  • Custom accelerators now handle over 50% of internal inference workloads at Google, Amazon, Meta, and Microsoft
  • Nvidia's share of data center compute within this tier has dropped from approximately 90% to 75%
  • Microsoft is deploying its Maia 200 chip across data centers; Google's TPU Ironwood delivers 4.6 PetaFLOPS; Meta's MTIA v2 powers its recommendation algorithms
  • Tesla is developing the AI5 chip, targeting 1.2 million chips quarterly by 2027 at costs 10x lower than Nvidia's H100 for inference workloads

This is not just cost optimization. It is the construction of vertically integrated AI stacks... where the software, the model, and the silicon are co-designed for maximum efficiency and minimum external dependency.

The Exponential Infrastructure Thesis

Despite efficiency gains in algorithms and chips, the world will need every data center currently planned and more. The AI 2027 forecast from the AI Futures Project models global compute growing 2.25x per year, with leading AI companies capturing 15–20% of global compute by decade's end. The returns to scale are so powerful that the authors describe "winner-take-all dynamics" intensifying as infrastructure investment compounds.

The Geopolitical Dimension: Sovereignty Requires Infrastructure

This dynamic is not confined to corporations.

Technological sovereignty has become a core pillar of national strategy. The Tony Blair Institute's 2025 analysis frames it bluntly: "Sovereignty is shaped by how well countries configure and negotiate their position within an inherently interdependent technological system." No state can maximize control, access, and coherence simultaneously, the task is to manage trade-offs across the AI stack in ways that preserve strategic autonomy.

The geopolitical competition for AI supremacy is not just about algorithms. It is about:

  • Energy infrastructure: who controls the power required to train and run frontier models
  • Semiconductor supply chains: who has secure access to advanced fabrication
  • Critical minerals: who can source the inputs for chips, batteries, and next-generation hardware

This explains the Trump administration's intensified interest in Greenland, framed as "strategic necessity" for national security, with its rare earth minerals, Arctic positioning, and freedom from Chinese supply chain dependencies. It explains the broader pattern of US engagement in resource-rich regions, from Venezuela to the Indo-Pacific.

China, for its part, leads in 66 of 74 critical technologies tracked by the Australian Strategic Policy Institute. The US leads clearly in one domain: frontier AI models. But as one analyst noted, China is not trying to beat the US to AGI... China is building "the infrastructure layer that makes AGI, whenever it arrives, irrelevant to most of the world if it remains locked behind American APIs, energy costs, and capital intensity."

Beyond Earth: The Final Frontier of Infrastructure

The logic of full-stack capture does not stop at the edge of the atmosphere.

SpaceX's orbital data center ambitions are not science fiction. Google's Project Suncatcher aims to deploy solar-powered satellites housing tensor processing units, with prototype launches targeted for early 2027. The rationale is straightforward:

  • Near-constant solar energy (no day-night cycles, no weather)
  • No land or freshwater constraints
  • Freedom from terrestrial grid bottlenecks
  • Potential for lunar manufacturing at scale (Musk envisions factories on the Moon producing data center satellites via electromagnetic mass drivers)

Economic viability for space-based compute is most commonly estimated for the early-to-mid 2030s, but SpaceX's Starship could accelerate that timeline dramatically by reducing launch costs by an order of magnitude.

This is a frontier with no clear turf lines and few legacy rules. Everything truly is up for grabs.

The Thesis

The theme running through all of this is simple but profound:

The next ten years are less about incremental improvement and more about re-architecting the systems that shape the world.

Companies and nations are relentlessly expanding control over the layers that determine long-term value creation, from electrons and energy to intelligence and infrastructure. The actors that secure auto-sufficiency across the full stack will define the trajectory of the AI-driven era. Those that remain dependent on external suppliers, grids, and intermediaries will find their optionality constrained at precisely the moment when optionality matters most.

Calling these "software companies" is no longer adequate. What they are building is much bigger than code.

This is the era of full-stack capture, technological sovereignty, and exponential infrastructure reimagining. The rules are being rewritten, the stakes are global, and the opportunities and the risks are enormous.

written by

Elias Mufarech
https://www.linkedin.com/in/elias-mufarech/https://x.com/eliasmufa