#44 | Who's really running the AI world

TL;DR: Who’s actually building our AI future? The companies that really run the AI world. My AI ecosystem map.

👋 Hello AI enthusiast,

If AI is reshaping everything from your work to your world, shouldn’t you know who’s actually behind the wheel?

We’re eight months into 2025, and the AI landscape isn’t the Wild West anymore.

It’s crystallized into clear power structures, dangerous bottlenecks, and ecosystem dependencies that will determine your options for the next decade.

While everyone debates AGI timelines, the real action is happening in supply chains, regulatory frameworks, political disruptions, and market consolidation.

I aim to better understand who controls the infrastructure we’ll depend on, which tools will survive the shake-out, and where the critical vulnerabilities lie.

The list is not comprehensive, and you might have a different opinion. Here is my AI ecosystem map.

The AI Learning Guy newsletter 🤖 🧠💡

AI learning hacks and mega prompts delivered to your inbox.

The Infrastructure & Software Builders (they control the plumbing)

These companies own the infrastructure everyone else builds on. (No infrastructure, no AI party.)

NVIDIA — 92% of the data center GPU market share. Locked up 60% of TSMC’s advanced packaging capacity through 2025, creating the industry’s biggest bottleneck. When an earthquake in Taiwan damaged 30,000 wafers earlier this year, every AI company’s deployment timeline shifted.

TSMC — The single point of failure for the entire AI ecosystem. Their CoWoS packaging technology is required for every advanced AI chip, but they can only meet 80% of demand despite massive expansion efforts.

Microsoft & Google — Each hold ~20% of enterprise AI infrastructure. Microsoft leverages Azure’s OpenAI integration, Google uses cloud infrastructure plus custom TPU chips. Both are diversifying beyond foundation model partnerships.

AMD & Intel — Challenging NVIDIA’s dominance with MI300X chips and Gaudi processors, respectively. Competition is heating up as companies seek alternatives to the GPU monopoly.

Oracle — Major infrastructure partner in OpenAI’s $100 billion Stargate initiative, providing the backbone for massive AI computing needs.

BlackRock AI Infrastructure Partnership — $100 billion investment with Microsoft, MGX, and NVIDIA, reshaping AI’s physical layer. Shows how traditional finance is entering the infrastructure game. (Wall Street wants a piece of the AI pie.)

The Tool Makers (what you actually use daily)

The AI applications people use for real work, not just demos.

OpenAI — 400 million weekly ChatGPT users. Launched ChatGPT Pro at $200/month for professionals. Still the consumer AI leader despite enterprise losses.

Anthropic — Flipped from 12% to 32% enterprise market share while OpenAI dropped to 25%. Claude’s superior code generation (42% market share) and agent-first architecture won over enterprises.

Cursor — Breakout development tool reaching $500 million ARR in two years. Created an AI-native coding environment that developers prefer over GitHub Copilot.

GitHub Copilot — Despite Cursor’s rise, still massive in enterprise development environments. Microsoft’s integration across development workflows maintains a substantial market presence.

Perplexity — 10 million monthly users challenging Google’s search monopoly with conversational search and direct citations. (Google’s having some competition for the first time in decades.)

Salesforce Agentforce 2.0 — Positioning the CRM leader as the enterprise AI platform. Connects AI directly to business workflows without requiring technical expertise.

xAI (Grok) — Elon’s uncensored AI integrated with X. Less filtered than other models and able to generate images of public figures—controversial, but growing fast.

DeepSeek R1 — Chinese model achieving state-of-the-art performance with significantly fewer resources, challenging computational requirements assumptions and affecting NVIDIA stock prices.

Character.AI — Conversational AI with personality. Users build ongoing relationships with AI personas that feel surprisingly real.

The Problem Solvers (industry-specific champions)

Vertical AI solutions proving economic value in specific industries.

Harvey — Raised $300 million at $5 billion valuation for legal AI. Law firms report significant efficiency gains in research and document drafting.

Databricks — Fastest-growing major enterprise software company with 60% YoY growth, serving 10,000+ enterprise customers. Unified data and AI platform becoming the standard for AI-powered analytics.

UiPath — Leading RPA platform offering business automation tools across API integration, intelligent text processing, and low-code app development. Acquired Peak AI in 2025 for industry-specific decision intelligence.

PathAI, Abridge, ClinicalKey — Healthcare AI addressing physician burnout through disease detection, clinical documentation, and medical decision support. Adoption despite longer sales cycles proves necessary.

ServiceNow (acquired Moveworks) — Enterprise IT automation with AI-powered features including workflows, analytics, knowledge management, and ticket automation. Leading copilot assistive AI technology.

Hebbia — Financial services due diligence and complex data room analysis. Portrait Analytics provides conversational access to real-time market data.

Palantir AIP — Enterprise data foundations with machine learning acceleration for both commercial and government applications. Worth noting: the same technology helping enterprises optimize supply chains also powers surveillance and military operations.

Anduril — (#1 on CNBC’s Disruptor 50) AI-powered defense systems, including autonomous underwater vehicles and air defense systems. Acquired Numerica’s radar and command control businesses, expanding into space technology.

Hyperexponential — Insurance industry pricing models. Even traditionally conservative industries implement AI when it solves real problems.

Bittensor (TAO) — Decentralized AI network where contributors earn TAO tokens based on the value they provide to machine learning models. It represents the convergence of AI and cryptocurrency.

Isomorphic Labs — Google’s AI drug discovery arm, using AlphaFold3 to revolutionize how medicines are developed.

Insilico Medicine — First company to advance an entirely AI-designed drug into Phase 2 clinical trials.

Triomics — AI platform matching cancer patients to clinical trials by parsing unstructured medical data.

Sana — an AI-powered knowledge management and enablement platform that acts like an embedded organizational brain, enabling employees to query and interact with internal knowledge through conversational interfaces.

The Creative Catalysts (redefining expression)

AI tools changing how we create and communicate.

Midjourney — Set the quality standard for AI image generation—professional workflows beyond novelty applications.

RunwayML — Hyper-realistic video generation challenging traditional production methods. (Your next favorite movie might be AI-generated.)

Synthesia — Enterprise customers with 230+ AI avatars in 140+ languages, replacing expensive video production for training content.

ElevenLabs — Voice cloning powering podcasts, audiobooks, and virtual assistants with near-human realism.

Suno — Complete 4-minute songs from text prompts. Democratizing music creation like Instagram democratized photography.

The Access Democratizers (AI for everyone else)

No-code platforms eliminating technical barriers to AI adoption.

Bubble — AI-powered application development. Make workflow automation connecting 1,000+ services. Zapier integration platform.

Google Vertex AI & DataRobot — Automated machine learning enabling business users to build predictive models without coding expertise. No-code platforms are projected to account for 65% of application development by 2027.

Consumer adoption numbers — 1.7-1.8 billion global AI users. This isn’t early adopter enthusiasm—it’s mainstream integration across demographics and industries.

The Wild Cards (disrupting everything)

Breakthrough companies challenging assumptions about AI development.

Safe Superintelligence — $2 billion raise at $5 billion valuation. Ilya Sutskever’s focus on AI safety represents the largest bet on alternative AGI approaches. (Because apparently we need a backup plan for the backup plan.)

xAI — 100,000 GPU Colossus cluster targeting 200-300,000 GPU capacity. Elon Musk’s X platform integration creates unique training data advantages. “No-filter” AI positioning appeals to users frustrated with existing model restrictions.

World Labs — Unicorn status developing Large World Models for 3D spatial intelligence, targeting AR/VR and robotics. Figure humanoid robots and Cerebras wafer-scale AI chips represent the “future is physical” trend.

11x — AI sales agents delivering full-cycle automation from prospecting to closing deals, already producing measurable revenue gains.

H Company — Paris-based startup making AI agents for real operational tasks. Runner H is outperforming leading benchmarks.

Beam AI — Enterprise-grade AI agents designed for deployment in complex business environments.

Windsurf — New AI coding assistant competing head-to-head with Cursor for daily developer workflows.

AutoGen, LangChain, CrewAI, Multi-Agent Systems, Relevance AI, Agent.so, n8n — The emerging AI agent and orchestration layer. These tools coordinate multiple AI models, automate workflows, and execute tasks end-to-end with minimal human intervention—pushing the frontier of autonomous operations.

The Political Reality Check

Regulatory frameworks affecting every AI company’s strategic planning.

US Policy Reversal — Trump rescinded Biden’s AI Executive Order within hours of inauguration, eliminating safety-focused regulations in favor of promoting “AI systems free from ideological bias.” Complete policy reversal creates uncertainty for companies that invested in safety compliance.

EU AI Act Operational — World’s first comprehensive AI governance framework. Companies with models trained using more than 10^25 FLOPS must notify the European Commission and conduct evaluations. Non-compliance risks market access restrictions. (Europe goes full regulatory while the US goes full deregulation.)

Export Control System — A three-tier global system forcing countries to choose between the US and Chinese AI ecosystems. Tier 1 countries get unrestricted access, Tier 2 receive limited quotas, Tier 3 face effective bans. (Cold War 2.0, now with semiconductors.)

The AGI Timeline Reality Check

Corporate executives predict AGI within 2-5 years, but evidence suggests complexity.

OpenAI’s o3 achieved 87.5% on ARC-AGI benchmark compared to humans at ~85%. Mathematical performance reached International Mathematical Olympiad gold-medal levels. Genuine progress in pattern recognition.

But controlled studies reveal limitations. METR’s rigorous trial found experienced developers worked 19% slower when using AI tools despite believing the technology helped them. Enterprise failure rates increased from 17% to 42% as companies abandoned initiatives that failed to deliver promised benefits.

76% of AI experts believe scaling current approaches is unlikely to lead to AGI, with median forecasts placing 50% probability around 2047. Current systems still struggle with basic reasoning about time, causality, and logical relationships. (Turns out intelligence is harder than we thought.)

Understanding Ecosystem Dependencies

The AI ecosystem’s interconnectedness creates both efficiencies and vulnerabilities.

TSMC’s packaging bottleneck represents the most critical single point of failure. Power supply emerging as the next constraint after GPU shortages gradually improve. AI Infrastructure Partnership’s $100 billion investment includes energy infrastructure because data centers require gigawatts of power.

Foundation model market concentration around three providers creates platform dependencies, though competitive dynamics remain intense. Anthropic’s rapid market share gain demonstrates how quickly leadership can shift.

Competitive cooperation patterns reveal strategic thinking beyond zero-sum competition. OpenAI adopted Anthropic’s Model Context Protocol despite being competitors, showing companies prioritize ecosystem growth over proprietary lock-in.

The Bottom Line

The AI ecosystem has evolved beyond the experimental phase into identifiable power structures with clear winners, critical vulnerabilities, and practical applications affecting how ambitious professionals work and compete.

The companies building your AI future aren’t necessarily generating headlines—they’re solving real problems, controlling critical infrastructure, and navigating complex regulatory environments.

Understanding these dynamics matters more than tracking model benchmarks or AGI predictions.

Focus on which AI tools provide genuine productivity gains, understand supply chain constraints affecting tool availability, and navigate geopolitical factors determining market access.

The transformation is accelerating, but it’s following predictable patterns of enterprise adoption, infrastructure development, and regulatory response.

Stay sharp,

Mark
The AI Learning Guy
👋⚡😎

The AI Learning Guy newsletter 🤖 🧠💡

AI learning hacks and mega prompts delivered to your inbox.

Sources and books

Note: No single website has all the answers. This list serves as a starting point for those who want to explore or satisfy their curiosity about AI.
Links: Links with * are affiliate links. See disclosure below.

Affiliate disclosure: To cover the cost of my email software and the time I spend writing these emails, I sometimes link to products. Please assume these links are affiliate links. If you choose to buy through my links, a big THANK YOU – it will make it possible for me to keep doing this.