What is Grey Machines?

Grey Machines is a decision intelligence platform powered by system dynamics and Monte Carlo simulation. Users describe any strategic question in natural language and receive causal loop diagrams, probability distributions, and ranked scenarios that quantify uncertainty and reveal what drives outcomes.

How does Grey Machines work?

Grey Machines uses a multi-stage AI pipeline: question analysis and domain detection, data retrieval from connected sources, baseline extraction, system modeling to build causal loop diagrams, validation, agent identification, intervention design, lever analysis, equation generation, Monte Carlo simulation, and results aggregation with ranked scenarios and actionable insights.

What types of decisions can Grey Machines simulate?

Grey Machines can simulate any strategic, policy, or business decision including market expansion, pricing strategy, supply chain resilience, urban policy design, investment decisions, workforce planning, energy transition, healthcare policy, and more. The platform works across any domain where feedback loops and systemic effects matter.

What is a causal loop diagram?

A causal loop diagram (CLD) is a visual representation of how variables in a system are interconnected through feedback loops. Grey Machines automatically generates CLDs from natural language questions, identifying reinforcing loops (which amplify change) and balancing loops (which resist change). These diagrams reveal the hidden dynamics that determine whether a decision succeeds or fails.

The AI Funding Singularity: $297B, Four Companies, and the Feedback Loop That Ate Venture Capital

Q1 2026 shattered every venture capital record. 81% went to AI. Four companies took 65%. We map the three reinforcing loops driving concentration - and what the dot-com crash teaches us about what breaks next.

Startups raised $297 billion last quarter. Four companies took $188 billion of it.

Q1 2026 was not just a record. It was a structural break in how capital flows through the global innovation economy. According to Crunchbase, the quarter's total exceeds every full-year venture total before 2018 and represents nearly 70% of all VC deployed across the entirety of 2025. Compressed into 90 days.

But the headline number obscures the real story. This is not a boom. It is a singularity - a point where the gravitational pull of a few companies becomes so strong that the rest of the ecosystem warps around them.

The Numbers

OpenAI closed at an $852 billion valuation, with Amazon ($50B), Nvidia ($30B), and SoftBank ($30B) competing to get in. Anthropic hit a $380 billion valuation with $14 billion in annualized revenue. These are not speculative bets on vaporware. They are the largest private capital deployments in history, concentrated into a handful of companies building the same category of technology.

The Reinforcing Loops: Success to the Successful

System dynamics has a name for what is happening: success to the successful. It is one of twelve canonical system archetypes identified by Donella Meadows in her foundational work on systems thinking. The structure is simple, and devastating.

Two or more efforts compete for the same finite pool of resources. The more successful effort gets a disproportionately larger allocation. That allocation widens the performance gap. The wider gap justifies even more allocation. The loop reinforces until one player dominates and everyone else starves.

Three distinct reinforcing loops are operating simultaneously in AI venture capital:

R1: The Capital Vortex

Mega-round announced -> Top talent migrates to funded company -> Capability gap widens -> More investors pile in -> Even larger round raised -> Repeat. Each cycle tightens the loop. OpenAI's $122B round had Amazon, Nvidia, and SoftBank competing to participate.

Sebastian Mallaby documented this pattern in The Power Law: venture capital follows a power-law distribution where a single investment can yield returns larger than all other investments combined. But the current concentration exceeds anything in venture history. Four rounds - four - captured 65% of all global startup investment in a quarter.

R2: The Compute Arms Race

Capital raised -> Massive GPU/data center buildout -> Better models trained -> Revenue grows (OpenAI: $2B/month) -> Higher valuation justifies more capital -> Repeat. Unlike dot-com companies, frontier AI labs have real revenue. OpenAI's enterprise revenue is 40%+ of total and growing.

This loop is self-sustaining as long as model capability continues to scale with compute. The $1.2 trillion in committed data center infrastructure spending suggests investors believe it will. But the relationship between compute investment and capability improvement is not guaranteed to hold indefinitely.

R3: The Starvation Loop

AI absorbs 81% of VC -> Non-AI sectors cannot raise -> Talent leaves non-AI for funded AI companies -> Non-AI innovation slows -> Investors see AI as only high-return sector -> More capital flows to AI.

Fintech, health, climate, and enterprise SaaS split $55 billion. That is less than half of what one company raised alone. Seed funding dropped while round sizes climbed at every stage, meaning fewer companies getting funded but survivors raising bigger rounds. PitchBook noted that for AI-native companies, "the winners will emerge and the number 3 to 8 players in categories will really struggle to raise and likely seek M&A."

These three loops do not operate independently. They reinforce each other. R1 concentrates capital, R2 converts capital into competitive advantage, and R3 eliminates alternative investment destinations. Together, they form a self-amplifying system that accelerates concentration with every quarter.

The Dot-Com Parallel: Same Structure, Different Variables

In 1999, internet companies absorbed roughly 80% of all venture capital deployed. Annual VC investment in the US surged from $7 billion in 1995 to nearly $100 billion in 2000 - a fourteen-fold increase in five years. The NASDAQ surged 85.6% in 1999 alone.

The feedback loops driving the dot-com bubble were structurally identical to what we see today. But with one critical difference.

Dot-Com R1: The Hype Spiral (1997-2000)

IPO pops 200%+ on day one -> Media amplifies success stories -> Retail investors pile in -> VCs fund anything with ".com" -> More IPOs with bigger pops -> Repeat.

Pre-revenue companies with no business model went public and doubled. Pets.com raised $82.5 million in its IPO with $619K in revenue. A study of 57 venture capital investors during the 1998-2001 period, published in the journal Venture Capital, found that accepted VC decision-making practices were systematically bypassed as investors competed in an unfamiliar sector.

Dot-Com R2: The Overcapacity Trap (1998-2000)

Cheap capital available -> Massive infrastructure buildout (dark fiber, servers) -> Overcapacity emerges -> Competition drives prices to zero -> Revenue models collapse -> But capital keeps flowing because "internet changes everything."

$1.7 trillion was spent on fiber optic cables. Only 2.7% of the installed fiber was ever lit. The overcapacity took a decade to absorb. Today, $1.2 trillion is committed to AI data center infrastructure. The GPU chips filling those data centers have an effective economic life of roughly one year before the next generation makes them obsolete. The parallel is uncomfortable.

Dot-Com B1: The Balancing Force (2000-2002)

Fed raises rates -> Cost of capital rises -> Investors demand profitability -> Unprofitable companies cannot raise -> Layoffs and shutdowns cascade -> Confidence collapses -> Reinforcing loops reverse direction.

The crash was triggered when balancing loops overpowered the reinforcing ones. The Fed raised interest rates six times between 1999 and early 2000. Capital costs rose. Investors demanded profitability. Companies that had been burning cash with no revenue path could not raise again. The NASDAQ fell 78% between March 2000 and October 2002, wiping out $5 trillion in value. Over 50% of dot-com companies that went public during the bubble had failed entirely by 2004.

The same "success to the successful" archetype works in reverse. Failure begets failure. Capital flight begets more capital flight. The reinforcing loop that built the bubble is the same one that destroys it - just spinning in the other direction.

What Is Different This Time (and What Is Not)

The bulls argue this cycle is fundamentally different. They have data to support it. OpenAI generates $2 billion per month in revenue. Anthropic's run-rate exceeds $14 billion with 10x annual growth. Enterprise AI spending hit $37 billion in 2025, up from $11.5 billion in 2024. OpenAI has 800 million weekly active users. These are not Pets.com numbers.

But system dynamics teaches us to look at the structure, not just the variables. And the structure is disturbingly similar.

What is genuinely different: real revenue, real users, real enterprise adoption, and infrastructure investments that create tangible assets rather than vaporware. The presence of actual business models means the reinforcing loops can sustain themselves longer before a balancing force intervenes.

But longer is not the same as forever.

What Breaks the Loop?

Every reinforcing loop eventually encounters a balancing force. In 1999, it was interest rate hikes and the evaporation of IPO demand. In 2026, the candidate balancing loops include:

The system dynamics model does not predict which balancing loop will dominate, or when. It predicts that one will. Reinforcing loops in finite systems always encounter limits. The only question is whether the correction is gradual or catastrophic.

The Takeaway for Decision-Makers

If you are allocating capital, setting strategy, or making policy, the system dynamics lens offers three non-obvious insights.

The dot-com era left us with transformative infrastructure - the internet itself, fiber optic networks, cloud computing foundations - alongside $5 trillion in destroyed value. The AI era will likely follow the same pattern: transformative technology, concentrated capital, and inevitable correction.

The companies and institutions that survive both the boom and the correction will be the ones that modeled the full range of outcomes - not the ones that assumed the reinforcing loop would spin forever.