How physical infrastructure, platforms, and AI agents are restructuring who controls the gateway to the digital world — and who captures the value.
From raw silicon to user experience — each layer concentrates control in fewer hands. Click any layer to explore who dominates and why.
The physical foundation of all digital power. In the AI era, control of compute hardware and energy supply has become the primary bottleneck — and the ultimate source of leverage.
TSMC fabricates ~92% of the world's most advanced semiconductors (sub-5nm). A single point of failure for global AI compute.
NVIDIA's H100/H200/B100/B200 GPUs power the vast majority of AI training worldwide. Competitors (AMD, Intel, custom chips) still trail significantly.
Surpassed $3T in 2025, now world's second most valuable company. Revenue ~$130B in 2025 fiscal year, up from $27B in 2023.
Global data center electricity consumption projected at ~75 GW by end of 2026, roughly 3% of global electricity. AI workloads driving exponential growth.
The three hyperscalers (AWS, Azure, GCP) collectively control ~67% of global cloud infrastructure. Their GPU clusters are the training grounds for frontier AI models.
Amazon Web Services remains the largest cloud provider. Invested $100B+ in AI/data center infrastructure through 2025–2026.
Azure's OpenAI integration drives enterprise AI adoption. Microsoft committed $80B to AI data centers in FY2025 alone.
GCP leveraging custom TPU chips and Gemini integration. Google pledged $75B in AI capital expenditure for 2025.
The "Magnificent Seven" collectively spending over $300B on AI infrastructure — an unprecedented capital allocation cycle.
Frontier models require billions in training compute. Only a handful of organizations can build them — creating a new oligopoly of intelligence. Open-source challengers (Meta's Llama, DeepSeek) add competitive pressure but haven't broken the concentration.
GPT-5 and o3 models. $300B+ valuation. ~300M monthly active users on ChatGPT. $13B+ annual revenue run rate. Backed by Microsoft ($13B invested).
Claude Opus 4 series. Known for safety-first approach. Valued at ~$60B. Backed by Google ($2B+) and Amazon ($4B+). Growing enterprise adoption.
Gemini Ultra/Pro models integrated across Google products. Custom TPU infrastructure. Massive data advantage from Search, YouTube, Gmail.
Open-weight Llama 3+ models. Strategy: commoditize models to protect social media dominance. Hundreds of thousands of GPUs deployed.
R1 model demonstrated that frontier performance is achievable at lower cost. Shook markets in Jan 2025. Challenged Western compute-maximalist paradigm.
Elon Musk's AI lab. Grok models trained on X (Twitter) data. Massive Memphis supercomputer cluster. Valued at ~$50B.
This is the contested frontier. The "Interface Reset Hypothesis" suggests every tech epoch ends when the primary interface between human intent and machine execution changes. AI agents may be that next shift — displacing app stores, search engines, and traditional UIs.
OpenAI's consumer interface and autonomous agent. First-mover advantage in conversational AI. Operator browses the web and executes tasks on behalf of users.
Google transitioning from "search" to "assist" — embedding Gemini into Chrome, Search, and Android. AI Mode in the browser represents self-disruption.
Siri enhanced with on-device AI + cloud models. Apple's bet: the screen persists, AI augments apps rather than replacing them. Late but integrated.
Embedded across Office 365, Windows, Edge, and Bing. Enterprise-first distribution. Copilot+ PCs with dedicated NPUs shipping since mid-2024.
Integrated across WhatsApp, Instagram, Facebook. 600M+ monthly users. Ray-Ban Meta smart glasses as new form factor — voice + vision interface.
Claude Code, Claude for Enterprise, API-first distribution. Claude's computer use and agentic capabilities targeting developer and enterprise workflows.
Power is exercised through defaults, fees, and friction — not explicit restriction. Apple's 30% App Store tax, Google's $20B+ default search deal with Apple, and platform rules that can eliminate billions in market value overnight. The AI era is restructuring these economic relationships.
Google paying Apple an estimated $26B/year (2026) to remain default search on Safari. A court ruled Google holds monopoly power but allowed the deal to continue, citing AI competition.
Apple's platform tax on digital transactions. The "gateway toll" that enabled $1.1T in App Store ecosystem commerce in 2025. Under pressure from DMA and global regulators.
Services revenue (App Store, iCloud, Apple Music, AppleCare) now ~$100B/year — the highest-margin segment. Ecosystem lock-in generates recurring revenue.
The "next App Store" won't be an app marketplace but an agent layer. Whoever orchestrates task routing, user trust, and monetization across devices owns the new digital tollbooth.
February 2026: Seven companies control ~$22 trillion in market value. AI has reshaped the hierarchy.
Combined AI capital expenditure exceeded $300B in 2025. The 2026 trajectory is even steeper. "Mountains of cash being burned to buy up computing power."
AI's power appetite is reshaping energy markets, grid planning, and even geopolitics. Data centers are becoming the largest new source of electricity demand worldwide.
Data center electricity demand projected to reach ~75 GW globally by end of 2026 — up from ~50 GW in 2023. Equivalent to powering France and Belgium combined.
New hyperscale AI campuses being planned at 1–5 GW each. A single campus consuming more power than some mid-sized cities. Grid operators scrambling to accommodate.
Microsoft signed a deal to restart Three Mile Island. Google and Amazon investing in small modular reactors (SMRs). Nuclear is the only carbon-free baseload option at scale.
Data centers projected to consume 3–4% of global electricity by 2027, up from ~1.5% in 2022. IEA warns of grid strain in key regions (Virginia, Dublin, Singapore).
Each technological epoch ends when the primary interface between human intent and machine execution changes. Mainframes to PCs. PCs to Mobile. Mobile to AI. Who controls the next gateway?
Tim Cook argues the screen remains central. AI enhances apps, Siri becomes intelligent, but the interaction model persists: user → screen → app → action. Vision Pro as the next screen, not the end of screens.
Zuckerberg bets against the screen. AR glasses overlay AI on the physical world via voice and augmented vision. Ray-Ban Meta glasses as early proof. The endgame: seamless digital-physical fusion.
No dedicated hardware. No screen requirement. No app store. Always-on AI agents that understand context, execute tasks, learn preferences. Platform-agnostic intelligence accessed through voice on any device.
A map of dominance across the five layers of digital power. Hover over any company to see their strategic position.
A historically unprecedented environment: regulators declining intervention on the basis that AI itself provides sufficient competitive disruption. But the EU is taking a different path.
Judge rules Google holds monopoly power in search. However, the court declines to block the Google-Apple default search deal, arguing AI competition (ChatGPT, Gemini, Claude) provides sufficient market disruption.
The EU AI Act enters full enforcement. Frontier model providers must comply with transparency, safety testing, and risk classification requirements. Fines up to 7% of global revenue for violations.
The Trump administration issued executive orders focused on AI dominance and deregulation, rolling back Biden-era AI safety orders. Focus on maintaining US AI leadership over China through reduced regulatory burden.
Apple forced to allow third-party app stores and alternative payment systems in the EU. "Gatekeepers" (Apple, Google, Meta, Microsoft, Amazon, ByteDance) face ongoing compliance obligations.
Expanded restrictions on AI chip exports to China. NVIDIA H20 chips restricted. ASML EUV machines banned for Chinese fabs. DeepSeek's success raises questions about whether controls are working.
A historical pattern repeats with a twist: IBM distracted by antitrust gave rise to Microsoft. Microsoft's antitrust enabled Google/Apple. Now regulators say AI disrupts incumbents naturally — so they decline to intervene. The incumbents use this window to consolidate AI power.
"We don't know. That's the gamble. Who's going to control the gateway?" — The uncertainty three years into the AI agent era is itself the signal.
Screens remain central. Phones persist as primary computing devices. AI enhances existing apps and interfaces but doesn't fundamentally alter the distribution architecture. Apple's ecosystem advantage transfers to AI-enhanced experiences.
New form factors — AR glasses, ambient devices, always-on wearables — replace smartphones as primary interfaces. The "embodied AI" vision wins. A new hardware cycle creates new gatekeepers.
The interface becomes software-defined and hardware-agnostic. AI agents operate across devices seamlessly. Platform control shifts to whoever owns the "agent layer" — the orchestration of task routing, user trust, and monetization. Hardware commoditizes.
Lessons from the architecture of digital power. Distribution architecture determines power more than technology capability.
Distribution control migrates to new interface layers during technological transitions. Position for where interaction will happen, not where it currently happens. The "next App Store" is an agent layer, not an app marketplace.
The App Store taught that reach trumps autonomy — developers accepted 30% commissions for access. AI may invert this: whoever provides best intelligence captures attention, regardless of hardware manufacturer.
Uncertainty about which company will dominate AI interfaces = opportunity. Multiple architectural approaches remain viable. Market structure hasn't crystallized. The $300B+ annual CapEx cycle is a bet on different outcomes.
Platform power manifests through defaults and friction, not explicit restriction. The Google-Apple deal exemplifies this: legal, voluntary, yet creating nearly insurmountable competitive barriers. AI concentration needs proactive frameworks.
Interface transitions create brief windows when switching costs decrease. The shift from apps to AI agents represents one such moment — before new lock-in mechanisms establish themselves. Choose wisely.
Every platform eventually becomes a bottleneck. Every technology eventually centralizes around a dominant architecture. The economic forces — network effects, scale economies, data compounding — reassert themselves in new forms. Whoever solves AI agent distribution becomes the new gatekeeper.