Then, in 2019, the oracle went silent.

The Return of the Oracle

For six years, whilst the world tumbled through a pandemic, crypto mania, and the birth of generative AI, Mary Meeker—the woman once dubbed "Queen of the Internet"—remained conspicuously quiet. No trend reports. No grand pronouncements.

Just the steady work of a Bond Capital general partner, watching, waiting, and perhaps most importantly, thinking.

I remember refreshing browsers in 2007, waiting for Meeker's latest insights to drop. Would mobile really matter? Was social media sustainable? Were we building something meaningful or just inflating another bubble? Her reports didn't just predict the future—they gave us permission to believe in it.

Then, silence. Six years of silence.

This month, the oracle returned with a vengeance: 340 pages titled "Trends – Artificial Intelligence." Not a gentle re-emergence, but a full-throated declaration that everything—and I mean everything—has changed whilst we weren't paying attention.

But here's what makes this different from every other AI report cluttering your inbox: Mary Meeker doesn't chase trends, she spots the tectonic plates shifting beneath them. When someone who correctly predicted Google, Amazon, and Apple's trajectories in the 1990s dedicates their first comprehensive analysis in six years exclusively to AI, you don't skim it. You study it.

And what that study reveals is both exhilarating and we're not witnessing another technology cycle. We're watching the complete reimagining of how intelligence itself works, at a speed that makes previous revolutions look like geological epochs.

This isn't just the next big thing. It's the end of everything we know.

The Five Forces That Make AI Different

Meeker's report reveals that with AI, we're not just witnessing innovation—we're watching the fundamental rules of technological adoption rewrite themselves. Five forces are converging to create something we've genuinely never seen before.

Force One: The Velocity Paradox

Mary Meeker doesn't use words carelessly. So when she deploys "unprecedented" fifty-one times across her presentation, it's not hyperbole—it's an admission that her historical models have finally broken.

Consider this: it took Google eleven years to accumulate 365 billion annual searches. ChatGPT hit that milestone in seventeen months—5.5 times faster than the search giant that defined the internet era. That's not incremental improvement; that's physics-defying acceleration.

But the ChatGPT numbers are just the visible surface of something deeper. Beneath the consumer adoption headlines, Meeker reveals the infrastructure story that explains why everything is moving so fast. Since 2010, the data available to train AI models has grown at 260% annually, whilst compute capacity has exploded at 360% per year. The result? A 167% growth rate in powerful AI models over just the past four years.

Moore's Law on Steroids

The developer ecosystem tells an even more startling story. Between May 2024 and May 2025, the number of developers in Google's AI ecosystem jumped from 1.4 million to 7 million. That's a five-fold increase in twelve months. To put that in perspective, it suggests the entire AI development community doubled every three months for a full year.

Meeker traces a pattern that reads like technological compression: personal computers took twenty years to reach 50% adoption in US households, desktop internet needed twelve years, mobile internet required six years, and AI appears poised to achieve the same penetration in just three years.

Force Two: The Titans' Gambit

There's something different about this technology wave, and it's not just the speed. It's the participants. Whilst previous revolutions began with scrappy startups carving out niches before the establishment took notice, AI has achieved something remarkable: it's made giants sprint from day one.

The earnings call revolution tells the story in stark numbers. When ChatGPT launched, roughly 10% of S&P 500 companies mentioned AI during their quarterly updates. Today, that figure exceeds 50%. But here's what's fascinating: this isn't reactive scrambling—it's proactive transformation.

A 2024 Morgan Stanley survey revealed that 75% of global Chief Marketing Officers were already running AI experiments, with virtually all others planning to begin within twelve months. Whilst old, the data still points out these aren't cautious pilot programmes buried in R&D departments; they're board-level initiatives with real budgets and measurable outcomes.

Even more telling is where companies are focusing their AI investments. Rather than the expected efficiency plays—cost reduction, headcount optimisation, administrative streamlining—early enterprise adopters are prioritising growth: production output, customer service enhancement, sales acceleration, and revenue expansion. The message is clear: this isn't about doing the same things cheaper; it's about doing entirely new things better.

The capital markets are responding accordingly. The "Big Six" US technology companies—Apple, NVIDIA, Microsoft, Alphabet, Amazon, and Meta—spent $212 billion on capital expenditure in 2024, a 63% increase from the previous year. That's not just growth; it's the largest cash commitment to a single computing platform in history.

But here's where Meeker's analysis becomes deliciously contrarian. Whilst infrastructure companies like NVIDIA are printing money—78% average revenue growth over five years—the foundation model companies themselves are burning cash faster than they can raise it.

Three of the four big tech firms producing AI models have seen significant drops in free cash flow, with only Meta managing to maintain steady margins.

Translation: 
the current AI boom is a customer's paradise and an investor's roulette wheel. 
The fierce competition between trillion-dollar companies attacking the same opportunity is creating unprecedented value for users whilst leaving the ultimate financial winners far from certain.

Force Two.5: When Dollars Become Pennies

If you want to understand why AI feels different from every previous technology wave, follow the cost curve. It's not just declining—it's collapsing so rapidly that it's rewriting the economics of intelligence itself.

Meeker's research reveals that inference costs—the expense of actually using AI models—have dropped 99% over the past two years. Let that sink in: what cost a dollar in 2022 costs a penny today. And the trajectory suggests we're nowhere near the bottom.

The infrastructure advances are staggering. A billion-dollar data centre equipped with NVIDIA's H100 chips can generate 58 trillion inference tokens annually. The next generation Blackwell chips will produce 24 times more tokens for the same billion-dollar investment. Meeker projects that such a facility could generate $7 billion in annual revenue at current costs—or, more likely, will drive costs so low that entirely new categories of AI applications become economically viable.

But here's the strategic insight that separates this revolution from previous technology cycles: we've never before witnessed multiple trillion-dollar companies with 50%+ gross margins attacking the same opportunity simultaneously in a relatively transparent world.

The internet era was largely about scrappy startups finding their niches whilst established players watched from the sidelines. The AI era is the clash of titans, with the world's largest companies deploying essentially all their resources into the same competitive arena. The result isn't just innovation—it's total war, with cost deflation as the primary battlefield.

This isn't gradual infrastructure build-out; it's an exponential arms race where yesterday's cutting-edge quickly becomes tomorrow's commodity.

The implications extend far beyond Silicon Valley balance sheets. When the cost of intelligence approaches zero, the constraint shifts from affordability to imagination. We're transitioning from asking "Can we afford to use AI for this?" to

"What becomes possible when intelligence is essentially free?"

This isn't just market dynamics—it's the fundamental restructuring of how we think about the value and scarcity of intelligence itself.

Force Three: The Global Awakening

The third force shatters Silicon Valley's comfortable assumption that innovation starts in California and slowly spreads outward. Meeker's data reveals that AI adoption is happening everywhere, all at once. It took 23 years for 90% of internet users to be outside North America. With ChatGPT, it took three years.

Here's where Mary Meeker's latest report reveals something that should fundamentally change how we think about technology adoption: AI didn't spread globally—it started globally.

The internet took twenty-three years for 90% of its users to be outside North America. ChatGPT achieved the same global distribution in just three years. But the numbers tell an even more remarkable story. The top user country for ChatGPT isn't the United States—it's India, representing 13.5% of usage, whilst the US accounts for just 8.9%. Indonesia and Brazil each claim over 5%, Egypt nearly 4%, and the pattern continues across every continent.

This isn't just impressive adoption metrics—it's a completely different model of how transformative technologies reach the world. Previous revolutions followed a predictable pattern: innovation in Silicon Valley, adoption in America, expansion to Europe, and eventual global reach. AI has inverted this entirely.

Think about the implications. When these users first experience the internet, they won't encounter browsers and search bars. They'll start with AI—conversational, intuitive, speaking their native language.

This leapfrog phenomenon isn't theoretical. It's happening now, and it's redefining what we mean by technological progress. Rather than upgrading existing paradigms, entire populations are skipping directly to AI-first interfaces.

They're not learning to search; they're learning to converse. They're not adapting to western-designed user interfaces; they're experiencing technology that adapts to them.

The strategic implications are profound. Companies built on traditional internet assumptions—search, browse, click—may find themselves speaking a language that half the world's AI users never learned. Meanwhile, experiences designed for conversational AI from the ground up have an immediate global audience that thinks in their native paradigm.

Even where ChatGPT faces restrictions, alternatives are rapidly filling the void.

DeepSeek, the Chinese model, is quickly infiltrating markets across Asia, Africa, and beyond. The result is a fragmented but comprehensive global AI ecosystem where geographical boundaries matter far less than user preference and government policy.

We're witnessing the birth of a truly global technology for the first time in history—one that doesn't require users to adapt to Silicon Valley's vision of how computing should work.

Force Four: The New Cold War

Andrew Bosworth, Meta's CTO, described the current state of AI as "our space race."

Andrew Bosworth, Meta's CTO, recently described the current state of AI with a metaphor that should chill every strategic planner.

This isn't competition—it's an arms race where the weapons are algorithms and the battleground is economic dominance.

Meeker's data reveals the scope of this contest with uncomfortable clarity. The total number of large-scale AI systems developed in the US and China absolutely dwarfs the rest of the world combined.

But the gap between the superpowers themselves is narrowing rapidly. Chinese AI models are achieving performance parity with their American counterparts whilst doing so with significantly lower training costs.

More concerning for Western strategists is China's advantage in embodied AI. As we transition from chatbots to robots, China has positioned itself with overwhelming infrastructure dominance. The country currently has more industrial robots installed than the rest of the world combined. The United States, meanwhile, ranks embarrassingly low in this critical category.

This isn't just about manufacturing efficiency—it's about the fundamental architecture of the future economy. When AI systems can manipulate the physical world at scale, the nations that control that capability will shape global commerce, logistics, and production.

But perhaps the most telling indicator isn't technological—it's psychological. Meeker's research reveals a stark attitude gap between Chinese and American citizens regarding AI's benefits. Between 2022 and 2024, 70% of Chinese citizens agreed that AI products and services offer more benefits than drawbacks. In the United States, that figure languished in the 30-40% range.

This optimism gap matters more than you might think. In a democracy, public scepticism creates political pressure for restriction and regulation. In China's system, public optimism provides political cover for acceleration and investment. The result is an asymmetric competition where one side operates with cultural tailwinds whilst the other fights cultural headwinds.

As Meeker warns with characteristic precision: "AI leadership could beget geopolitical leadership, not vice versa."

The message is clear: this isn't just about who builds the best models or deploys the most compute. It's about which societies, economic systems, and value frameworks can most effectively harness artificial intelligence to augment human capability. The winners won't just dominate technology—they'll define the next century of global power.

Force Five: Everything, Everywhere, All at Once

For all its comprehensiveness, Meeker's report reveals something fascinating through its omissions. Despite 340 pages of analysis, there's almost nothing about AI agents—the autonomous systems that represent the cutting edge of current development.

This isn't an oversight; it's a feature of macro analysis in hyperspeed times. Meeker's report captures the assistant era of AI brilliantly, documenting the rise of ChatGPT and similar conversational interfaces. But the technology has already moved beyond that paradigm whilst the report was being compiled.

The evidence is hiding in plain sight. Meeker notes that Google searches for "AI agent" have increased 1,100% over the past sixteen months. Yet this explosive interest merits barely a mention in her analysis. Similarly, when the report asks ChatGPT to predict the top ten capabilities that AI will have in five years, many of those predictions—particularly "generate human-level text, code, and logic"—are arguably already here.

This temporal displacement reveals something crucial about the current moment: we're moving so fast that even the most sophisticated analytical frameworks struggle to stay current. By the time comprehensive research is conducted, compiled, and published, the landscape has shifted fundamentally.

For executives, this creates a peculiar challenge. The most thorough analyses of AI trends are simultaneously the most useful and the most outdated. Meeker's report provides essential context for understanding how we reached this point, but offers limited guidance for navigating what comes next.

The agent revolution is already underway, driven by companies like Anthropic, OpenAI, and countless startups building autonomous systems that don't just respond to prompts but actively pursue goals. These systems represent the difference between AI as a sophisticated search engine and AI as a digital workforce.

Yet because agents are still emerging—and because their implications are so profound they resist easy categorisation—they remain largely invisible to macro analysis. It's a reminder that in periods of genuine transformation, the most important developments often happen in the spaces between established categories.

AI Agents Not Factored Yet

This transition represents a qualitative shift, not just quantitative improvement. Assistants augment human capability; agents potentially replace human processes entirely. Assistants require human oversight; agents operate autonomously within defined parameters. Assistants are tools; agents are colleagues.

The implications cascade through every insight in Meeker's report. If ChatGPT's adoption curve looks unprecedented, wait until agents that can independently research, plan, and execute tasks become ubiquitous. If enterprise adoption of AI assistants has been rapid, consider the velocity when agents can run entire workflows without human intervention. If geopolitical competition around AI is intense now, imagine when autonomous agents can conduct cyber operations, manage supply chains, and coordinate economic activities at superhuman scale.

Applying different thinking to decode what everyone else is missing

Eight Counter-Intuitive Truths About AI's Real Impact

The Adoption Paradox: Mass Usage Creates Mass Resistance

What if widespread AI adoption is actually proof of underlying rejection? Like a medieval monk observing that the more people attend mandatory church services, the less genuinely faithful they become, mass AI usage might signal cultural antibodies forming against the technology.

Instead of measuring adoption success, we should measure "adoption quality"—the depth of integration, not breadth of usage. The most successful AI implementations might be the ones with the lowest user counts but highest per-user value creation.

The Infrastructure Inverse: Building Roads Doesn't Create Destinations

An outsider would notice that humans are building the world's most expensive highways (AI infrastructure) to reach places that don't exist yet (profitable AI applications). It's like Victorian railway mania—spectacular engineering serving speculative destinations.

You cannot eat the plough, only what it helps you grow." The $212 billion in infrastructure spending resembles building magnificent cathedrals before anyone knows what religion they're for.

The telecoms bubble of 2001 already solved this—laying fibre optic cables created real value, but not for the companies that laid them. Today's AI infrastructure builders might be tomorrow's valuable assets owned by entirely different companies.

The Benchmark Delusion: Perfect Tests, Imperfect Reality

A Victorian engineer would recognise this immediately—steam engines that work perfectly on test tracks often derail on actual railways with real weather, real cargo, and real time pressures. We're making the same category error with AI.

The more sophisticated our AI becomes at artificial tasks, the less equipped it is for human ones. We're optimising for an imaginary world that doesn't exist.

Models and reality exist in different states simultaneously—brilliant and useless, powerful and fragile, revolutionary and evolutionary. The key is understanding which state applies when.

The Agent Fallacy: Autonomy Is Just Elaborate Dependency

What if "autonomous" agents are actually making us more dependent, not less? Like GPS making us unable to read maps, AI agents might be atrophying our ability to think through complex workflows independently.

"If it's so smart, why does it need so many rules?" Children intuitively understand that true intelligence doesn't require constant supervision—which suggests our "intelligent" agents aren't actually intelligent yet.

We're not building autonomous agents; we're building sophisticated puppets that require increasingly skilled puppeteers. The real scarce resource isn't AI capability—it's human puppet-mastery.

The Tokenisation Trap: Digital Legibility vs. Human Illegibility

We're witnessing the collision between two incompatible knowledge systems—digital (explicit, tokenised, transferable) and human (tacit, contextual, embodied). Like trying to explain the colour blue to someone born blind, some knowledge simply cannot cross the digital divide.

Ecological science already knows this—you can measure every chemical component of soil, but the "life" in living soil emerges from relationships that resist measurement. Organisational life follows the same pattern.

The more we succeed at tokenising knowledge, the more valuable untokenisable knowledge becomes. We're accidentally creating a new scarcity economy around human intuition.

The Human Persistence Premium: Marathons vs. Sprints

Humans and AI exist in complementary quantum states—AI for acceleration, humans for navigation. Neither can do the other's job, and attempts to force convergence break both systems.

Future historians might view this era as "The Great Specialisation"—when humans finally stopped trying to be faster computers and computers stopped trying to be artificial humans. Both species found their unique evolutionary niches.

A child would observe that humans are the only species that gets better at decision-making as tasks get harder and longer. This isn't a bug—it's our core competitive advantage in an age of infinite compute.

The Regulation Racing Paradox: Speed Creates Its Own Obstacles

What if slow regulation is actually accelerating AI development? Like water flowing around rocks, regulatory resistance might be forcing innovation into more valuable channels—toward privacy-preserving AI, explainable systems, and human-centric design.

A monk would recognise this as the eternal tension between ecclesiastical law and lived practice. The gap between official rules and actual behaviour often produces the most interesting innovations.

Evolutionary biology shows that constraints accelerate adaptation. The regulatory "friction" might be creating stronger, more resilient AI systems—like how muscles grow stronger under resistance.

The Altitude Illusion: The View from 30,000 Feet Hides the Terrain

What if we're suffering from "macro myopia"—the inability to see the forest for the trees has been replaced by the inability to see the trees for the forest? Meeker's altitude gives perfect visibility of trends but zero visibility of the rocks that will sink individual ships.

An outsider would notice that humans have developed two completely incompatible languages for describing the same phenomenon—"AI Revolution" (macro) and "AI Implementation" (micro)—with no translation layer between them.

We're simultaneously living in the most documented technological transition in history (macro) and the most mysterious one (micro). Everyone knows what's happening at scale; no one knows what will work at their specific scale.

"Why are all the maps about the same mountain, but everyone gets lost hiking it differently?" The disconnect between pattern and practice suggests we need new kinds of maps—not just topographical, but experiential.

Here are the five strategic frameworks that emerge from Meeker's analysis:

The Speed Imperative: Traditional planning cycles are now too slow. Organisations must develop the capability to make strategic bets with incomplete information and adjust rapidly based on market feedback.

The Platform Play: Rather than building specific AI applications, focus on creating platforms that can incorporate new AI capabilities as they emerge. The winners will be those who can continuously upgrade their intelligence layer without rebuilding their entire system.

The Global Mindset: AI is native global from day one. Products and strategies designed for single markets will lose to those built for immediate global deployment.

The Talent War: The constraint isn't capital or technology—it's human expertise. Organisations must prioritise AI education and capability development as a core strategic initiative.

The Optionality Principle: In a world moving this fast, maintaining multiple strategic options may be more valuable than committing fully to any single approach.

The Inflection Point

There's a moment in every revolution when the world divides into two groups: those who see what's coming and those who don't. Mary Meeker's return after six years of silence isn't just analysis—it's a clarion call that we've reached that moment for artificial intelligence.

When someone who correctly predicted the internet's trajectory declares that AI represents a fundamentally different phenomenon—unprecedented in scope, speed, and global impact—it's time to stop asking whether the revolution is real and start asking whether we're building the future or being built by it.

The data is overwhelming. The trends are undeniable. The convergence of cheap intelligence, global adoption, and geopolitical competition has created a technological perfect storm that will reshape every industry, every organisation, and every career over the next decade.

But here's the paradox that Steve Jobs would have appreciated: in a world where everything changes faster than ever, the greatest risk isn't moving too fast. It's the dangerous illusion that you have time to think about it.

In Silicon Valley, they often speak of inflection points and paradigm shifts. But sometimes, if you listen carefully to the data, you can hear something deeper—the sound of an old world ending and a new one beginning. In 340 pages of charts and statistics, Mary Meeker has documented that sound.

The oracle has spoken. The intelligence revolution is upon us. And this time, unprecedented isn't hyperbole—it's simply accurate.

The future that Meeker describes isn't coming—it's already here. The only question is whether you're ready to step into it.

Think different. The revolution is now.

WATCH OUR COMMUNITY CONVERSATIONS

AI For LUNCH Webinar

A brilliant discussion about AI and how it is impacting enterprise, including:

• How human-in-the-loop is going to change, if at all, over time?

• Data cleanliness

• How LLM's work

• AI Agents

• and much much more

Follow us for more