A Tiny Island’s Metric for Our Moment

There’s a small Caribbean island called Anguilla with a population of about 16,000 people. For decades, it has administered the .ai domain extension, a modest footnote in the grand architecture of the internet. Through the 2010s, registrations trickled in. In 2020, there were roughly 40,000 .ai domains worldwide. The extension earned Anguilla’s government about $6 million annually: not nothing, but hardly transformative.

Then ChatGPT launched in November 2022.

Within twelve months, .ai domain registrations surged 230%. By 2024, the growth hit 300%. In January 2026, the total crossed one million registered domains. Anguilla’s revenue from .ai domains jumped from $6 million to $70 million, nearly half of the entire government’s revenue, up from under 1% just five years earlier.

Year	Estimated .ai Revenue (USD)	% of Government Revenue	Estimated GDP (USD)	% of GDP
2020	~$7M	~5–6%	~$300M	~2%
2021	~$12M	~9–10%	~$300M	~4%
2022	~$20M	~15%	~$310M	~6–7%
2023	~$30–32M	~20–22%	~$320M	~9–10%
2024	~$40–45M	~23–25%	~$320–330M	~13–14%
2025 (proj.)	~$90–95M	~45–50%	—	—

This isn’t just a quirky economic story. It’s a seismograph reading for the pace of innovation itself. And here’s why: it is an accident. It has nothing to do with Anguilla itself except an alphabetic coincidence. Anguilla can’t control it in any way. And yet, it is impacting the tiny island like nothing else in the country’s history. 

The Numbers Behind the Acceleration

The .ai domain explosion is one visible metric in a broader pattern. Consider what happened in 2024 alone:

In academic research: ArXiv, the primary repository for AI papers, saw submissions in its AI category nearly double from 1,742 papers in 2023 to 3,242 in 2024. That’s not linear growth. That’s exponential.

In model development: The U.S. produced 40 notable AI models in 2024, while the performance gap between the top-ranked model and the 10th-ranked model shrank from 11.9% to just 5.4% in a single year. The top two models are now separated by only 0.7%.

In business formation: Venture capital funding into AI reached $107 billion globally in 2025, up 28% year-over-year, with AI startups capturing 26% of all global VC funding.

In patents: AI-related patent applications in the U.S. grew from 34,544 in 2020 to 54,022 in 2024: four consecutive years of growth.

These are big numbers. They’re also accelerating numbers. The rate of change is itself increasing.

The Threshold We Just Crossed

For most of the 2010s and early 2020s, those of us paying attention to technology could maintain a mental map of the landscape. We could identify the major players, understand the breakthrough papers, track which startups mattered, and discern meaningful signals from noise.

That changed sometime in the past few months. We can even pinpoint it with some accuracy to November 2025. Or three short months ago. In just over 30 days:

• November 12: OpenAI – GPT-5.1
  
  
• November 19: OpenAI – GPT-5.1-Codex-Max (specialized coding version)
  
  
• Mid-November: xAI – Grok 4.1
  
  
• Mid-November: Google – Gemini 3.0
  
  
• Late November: Anthropic – Claude Opus 4.5
  
  
• December 11: OpenAI – GPT-5.2
  
  
• December 18: OpenAI – GPT-5.2-Codex
  
  

The volume, velocity, and interconnectedness of developments have exceeded our cognitive bandwidth. We’ve passed a threshold where human attention, even expert attention, can no longer track the frontier. This is partly due to scale of funding, partly due to consequent concentration, and partly due to the explosion of creativity founded in users driving this innovation.

Consider what this looks like in practice:

• Biomedical information now doubles every two months, and that doubling rate is itself expected to double by 2025.
  
  
• The world produces over 403 million terabytes of data each day, about 147 zettabytes per year.
  
  
• By 2028, AI-generated scientific papers are projected to outnumber human-authored papers.
  
  
• Stanford’s AI Index notes that nearly 90% of notable AI models in 2024 came from industry, up from 60% in 2023, while model training compute doubles every five months.
  
  

The problem isn’t separating wheat from chaff anymore. It’s that there’s so much wheat we can’t possibly mill it all, let alone bake it into anything comprehensively and collectively digestible.

The Paradox: AI Creating Its Own Information Crisis

Here’s the strange loop we’re in: AI is both the cause of this information explosion and potentially the only solution to managing it. Research shows that large language models themselves suffer from “information overload”: when given too much context, their performance actually degrades. Like humans experiencing bounded rationality, AI systems have structural limitations in processing excess input.

We’re not just drowning in information. We’re drowning in relevant information. The challenge isn’t filtering out spam or obvious nonsense. It’s that there are legitimately important developments happening simultaneously across:

• Multiple competing AI architectures (transformers, mixture-of-experts, sparse attention mechanisms)
  
  
• Breakthrough applications in dozens of domains (drug discovery, climate modeling, materials science, code generation)
  
  
• Fundamental shifts in training methodologies (from supervised learning to reinforcement learning with verifiable rewards)
  
  
• New hardware paradigms (quantum computing, neuromorphic chips, custom AI accelerators)
  
  
• Regulatory frameworks being written in real time across different jurisdictions
  
  
• Ethical debates about safety, alignment, and existential risk
  
  

Each of these threads involves hundreds of papers, dozens of companies, and competing technical approaches. Even researchers specializing in narrow subfields struggle to keep current.

What It Means to Lose the Macro View

The loss of macro-level comprehension has real consequences:

For researchers: You can no longer be confident you’ve found the best approach to a problem. There might be a superior method published last week that you haven’t seen yet, and won’t see, because it’s buried in the avalanche.

For businesses: Strategic planning becomes nearly impossible when the technological landscape shifts fundamentally every quarter. The startups you track in January might be obsolete by June, overtaken by techniques that didn’t exist in March.

For policymakers: How do you regulate a technology when you can’t track what it’s capable of? By the time you understand version N, we’re already at version N+3.

For individuals: The feeling that you’re falling behind is no longer anxiety, it’s accurate. We are all falling behind. The frontier of human knowledge is advancing faster than any human can track it.

The Historical Precedent We Don’t Have

Past technological revolutions (printing, electricity, computing, the internet) all accelerated information flow. But they unfolded over decades, giving society time to develop new institutions, literacies, and filters. The academic journal system evolved to manage the explosion of scientific publishing. Media criticism developed to help navigate information sources. Digital literacy became a skill set.

The AI acceleration is different in kind, not just degree. The interval between breakthrough and obsolescence has compressed from years to months to weeks. The number of meaningful developments happening in parallel has exploded. And the technology itself is recursive: AI systems are now being used to design better AI systems, creating a feedback loop that has no historical precedent, nor an information management or meaningful cognition framework.

Living at the Threshold

We’re experiencing something genuinely new: the moment when collective innovation velocity exceeds human cognitive capacity to track it. The .ai domain surge, from sleepy country code to million-domain gold rush in 39 months, is just one concrete measurement of this phase transition.

When productivity becomes so productive that the job becomes not focusing internally, but scanning emerging signals externally, keeping up with the data is no longer a person’s full-time job. It’s not a full-time job at all. It’s a company-level full-time job. That means work shifts from departments to signal detection, responding to those signals, protecting intellectual property, developing intellectual property, and reacting to changes in the intellectual property environment, when innovation can happen literally in the blink of an eye and eclipse years of negotiation, code, and infrastructure.

I can personally feel this in many different ways. The mutual aid work I do is increasingly overwhelming as people connect to ask for help, but individual voices are getting drowned out by volume and tuned out because of platform changes and attention fragmentation. Twitter aka X seems to shift algorithmically every day, blind (and unconcerned) by how many people are getting lost in the pace of societal change and depend on its mercurial algorithm for food and shelter. People take it personally, because it is personal. Nothing is changing, but nothing is the same, day in and day out, and it is far beyond anyone’s ability to control. 

What I, others, and Anguilla are experiencing isn’t unique or random. German sociologist Hartmut Rosa predicted this exact moment two decades ago.

Hartmut Rosa’s Social Acceleration Theory

The most directly applicable framework is Social Acceleration by German sociologist Hartmut Rosa. He identified three dimensions of acceleration that create a self-reinforcing spiral:

1. Technological Acceleration

The speed of transportation, communication, and production increases. This is what we just witnessed with GPT-5.1, 5.2, Grok 4.1, Claude Opus 4.5, and Gemini 3 all dropping within 60 days.

2. Acceleration of Social Change

Cultural knowledge, social institutions, and personal relationships change faster. What you know today is obsolete tomorrow. The job you trained for doesn’t exist. The company you work for pivots quarterly.

3. Acceleration of the Pace of Life

Despite technology supposedly “saving time,” we experience increasing time pressure. You should have more free time with AI tools, but instead you’re drowning trying to keep up with AI developments.

The critical insight: These three forms of acceleration reinforce each other in a self-propelling feedback loop. Faster technology = faster social change = faster pace of life = demand for faster technology.

The Concept That Explains Everything: “Frenetic Standstill”

Rosa coined the term “frenetic standstill” to describe the paradox we’re living through:

“Nothing remains the way it is while at the same time nothing essentially changes.”

What this means in practice:

• Everything is constantly moving at breakneck speed
  
  
• Yet nothing “really” ever changes in a meaningful direction
  
  
• We’re running faster and faster just to stay in the same place
  
  
• All activity, no progress
  
  
• Constant motion, no direction
  
  

Why November 2025 Was the Frenetic Standstill Made Visible

When GPT-5.2 dropped 29 days after GPT-5.1, anyone trying to evaluate models experienced this viscerally:

• You couldn’t finish testing 5.1 before 5.2 rendered it obsolete
  
  
• But did anything fundamentally change? Or just marginal improvements?
  
  
• You’re frantically updating, but to what end?
  
  
• Motion without destination
  
  

This is Rosa’s frenetic standstill at the individual level. At the societal level, it’s even worse.

The “Shrinking of the Present”

Rosa argues that modern society is characterized by the “shrinking of the present”, a decreasing time period during which expectations based on past experience reliably match the future.

• 2004: You could assume technologies would work similarly for 2–3 years. A computer purchased in 2004 was usable until 2007–2008.
  
  
• 2024: A model released in November might be obsolete by December. The “present” where your knowledge is valid has shrunk to weeks.
  
  
• 2025: After November–December 2025, the present has shrunk to days or hours. By the time you integrate GPT-5.1, GPT-5.2 is here.
  
  

The Consequence: Social Cohesion Breaks Down

When the present shrinks this much:

1. Shared reference points disappear
   
   
   • You’re using Claude Opus 4.5, I’m using GPT-5.2, your colleague is still on GPT-4o
     
     
   • We’re not even working with the same capabilities anymore
     
     
   • How do we have shared conversations about “what AI can do”?
     
     
2. Generational divides become intragenerational
   
   
   • It’s not Boomers vs. Millennials anymore
     
     
   • It’s “people who adopted in November” vs. “people who adopted in December”
     
     
   • The knowledge gap opens within months, not decades
     
     
3. Experience becomes devalued
   
   
   • Your expertise from six months ago is worthless
     
     
   • Elders can’t pass down wisdom because it’s obsolete
     
     
   • The mechanism for cultural transmission breaks
     
     

Dynamic Stabilization: Running Faster to Stay Still

Rosa identifies dynamic stabilization as a defining feature of modern capitalist societies:

“Capitalist societies require (material) growth, (technological) dynamization and high rates of (cultural) innovation in order to reproduce their structure and to preserve the socioeconomic and political status quo.”

In plain English: We must accelerate just to maintain what we have.

The AI version:

• OpenAI must release GPT-5.2 29 days after GPT-5.1 just to maintain market position
  
  
• Google must release Gemini 3 to avoid falling behind
  
  
• Anthropic must release Claude Opus 4.5 to stay competitive
  
  
• Nobody wants this pace, but everyone is trapped in it
  
  

This is the “race to the bottom” of innovation cycles. No company can unilaterally slow down without losing.

This isn’t necessarily catastrophic. Humanity has always operated with incomplete information. Scientists have always specialized. But we’re entering an era where even aggregated human expertise can’t maintain a coherent picture of the whole.

The question isn’t whether we can slow this down; the economic, competitive, and scientific incentives are too powerful, too unstoppable. The question is what intellectual infrastructure we need to navigate a world where innovation outpaces comprehension.

Because right now, we’re all standing on that tiny island in the Caribbean, watching the registration numbers tick upward faster than we ever imagined possible, knowing this is measuring something profound but not quite able to grasp what comes next.

The wheat is piling up faster than we can mill it. And the harvest is still accelerating.

Innovation Metrics: September 2004 vs 2024/2025

Metric	2004	2024/2025	Change	Growth Factor
US Patent Grants	164,290 (annual)	368,597 (Dec 2023–Nov 2024)	+204,307	2.2x
AI-Related Patents (US)	Negligible	54,022 (2024)	N/A	Exponential
Scientific Publications (MEDLINE)	579,041 (indexed citations)	1,000,000+ (2024 est.)	+420,959+	1.7x+
ArXiv AI Papers	~100–200 (est.)	3,242 (2024 alone)	+3,000+	16–30x
Global Venture Capital Investment	$20.9B (US only)	$340B+ (global, 2024)	+$320B	16x (US baseline)
R&D Spending (OECD, % of GDP)	~2.2%	2.7%	+0.5 pp	23% increase
Global R&D Spending (Total)	~$800B (est.)	~$3.1T (2022)	+$2.3T	~4x
.ai Domain Registrations	Few thousand	1,000,000+ (Jan 2026)	+~1,000,000	100x+

What These Numbers Really Mean

The Absolute Scale Has Exploded

In 2004, the entire US venture capital industry invested $20.9 billion. In 2024, just AI companies globally captured over $100 billion. The scale of innovation investment has grown 10-20x depending on the metric.

But the Velocity Has Increased Even More

Look beyond the totals to the rate of change:

– 2004 context: ArXiv AI papers were published steadily, maybe 100-200 per year

– 2024 reality: 3,242 AI papers on ArXiv in a single year, and that’s accelerating

  – This means roughly 9 AI papers per day in 2024 vs less than 1 per day in 2004

The Concentration Is Unprecedented

2004 .ai domains: A few thousand registrations, mostly tech companies and AI researchers  

2024-2026 .ai domains: Over 1 million — growing at 20,000+ per month

This represents a 100-fold increase in just ~20 years, with most of the growth happening in the last 2-3 years.

Patent Growth Masks the Real Story

Patents grew “only” 2.2x from 2004 to 2024. But:

– AI patents grew from nearly zero to 54,022 in 2024 alone

– Traditional patent categories are flat or declining

– The composition of patents has fundamentally shifted toward AI and software

*Disclosure statement: Claude Sonnet 4.5 helped with sourcing of the data for both tables, and editing. ChatGPT 5.2 helped with verification of the data in both tables, the production of the image used, and proofreading/editing.