The 99.7% Solution: Mary Meeker Previews a Future Powered by Commoditized AI
- David Golub
- 15 minutes ago
- 5 min read
Over the course of the world's industrial eras, technologies that increase the efficiency of resource consumption tend to increase overall usage of those same resources, opening new pathways for unprecedented growth.
In her seminal AI report, Mary Meeker looks back at this seeming paradox from the first Industrial Age (steam power) to underscore how the enormous productivity potential of artificial intelligence is poised to unleash entirely new categories of human activity.
Reading the hefty 340-slide presentation, one might easily get lost in debates about whether AI will replace jobs or which model performs best, and miss a key takeaway: AI inference costs (the price of generative outputs) have plummeted 99.7% in just two years.
This isn't incremental improvement. This is the kind of exponential cost reduction that historically awakens entirely new categories of economic production.
Meeker and her BOND colleagues look back at 19th Century British Economist William Jevons, who observed that increasingly efficient coal engines led to increases in coal consumption, to argue that the growing affordability of inference also presages an era of widespread, if bifurcated, AI adoption:
"For some, the evolution of AI will create a race to the bottom; for others, it will create a race to the top."
Like Meeker's well-known Internet Trends report, the AI presentation features an impressive bricolage of industry graphs and statistics, but the key insight about compute's rapid commodification stands out as a key signal for business planning.
For business leaders today, the urgency is envisioning new commercial models and new products powered by ever-cheaper AI capacity.
A Paradox Plays Out in Real Time
What cost dollars two years ago now costs pennies. What cost pennies now costs fractions of cents. Meeker's team documents how NVIDIA's latest GPUs require 105,000x less energy per AI operation than their chips from just a decade ago — a compression of computational economics that makes the PC revolution look glacial.
“Inference represents a new cost curve, and – unlike training costs – it's arcing down, not up,” Meeker’s report reads. “As inference becomes cheaper and more efficient, the competitive pressure amongst LLM providers increases – not on accuracy alone, but also on latency, uptime, and cost-per-token.”
The resource consumption paradox is already visible:
Google now processes 50x more AI tokens monthly than a year ago (from 9.7 trillion to 480 trillion). Microsoft's Azure AI platform handles 5x more quarterly volume. Meta's open-source Llama models have been downloaded over 1.2 billion times.
When OpenAI's CEO notes that "older people use ChatGPT as a Google replacement, but people in their 20s and 30s use it like a life advisor," he's describing the early signs of the personalized AI that cheap inference makes economically viable at massive scale.
“For users (and developers), this shift is a gift: dramatically lower unit costs to access powerful Al. And as end-user costs decline, creation of new products and services is flourishing, and user and usage adoption is rising,” the report argues.
From Conversation to Action
Meeker identifies a crucial shift from what she calls "chat responses" to "doing work" — AI systems that don't just answer questions but execute multi-step workflows autonomously.
This transformation from reactive AI to proactive agents represents what we've called the agentic AI revolution - where AI systems don't just respond to queries but autonomously orchestrate complex business processes.
The examples are no longer experimental: Bank of America's Erica AI assistant has handled over 2 billion customer interactions. JP Morgan reports $1.5 billion in value from AI modernization. Kaiser Permanente's AI has processed 10 million patient visits.
As Amazon's CEO Andy Jassy is quoted observing, "AI does not have to be as expensive as it is today, and it won't be in the future." That future is arriving faster than most anticipated, enabled by the economic fundamentals Meeker documents.
The Strategic Divergence
Here's where the "race to the bottom vs. race to the top" dynamic becomes the defining strategic question of our time. The cost collapse creates two distinct paths:
The Bottom: Companies using cheap AI inference merely to trim costs and optimize existing workflows—treating AI as a better spreadsheet or faster search engine.
The Top: Organizations building entirely new value propositions around always-on intelligence. Meeker highlights companies like Anduril (2x revenue growth in defense AI), Tesla (100x increase in autonomous driving miles), and Applied Intuition (serving 18 of the top global automakers) that are constructing AI-native business models from first principles.
The report captures this urgency in corporate messaging: Shopify's CEO now mandates "reflexive AI usage as baseline expectation." Duolingo's CEO declares his company "AI-first," stating they'd "rather move with urgency and take occasional small hits on quality than move slowly and miss the moment."
Key Questions Every Leader Must Answer
Meeker's data doesn't just diagnose the transformation — it demands immediate strategic decisions. The companies winning the "race to the top" are those answering these questions with clarity and speed:
Value Creation or Cost Cutting? Are you using AI primarily to reduce expenses, or building new revenue streams? The report shows companies pursuing "growth & revenue" strategies with AI significantly outperforming those focused on "cost reduction."
Platform or Feature? Will you build comprehensive AI capabilities, or integrate AI into existing workflows? Meeker documents OpenAI challenging Microsoft's 34-year Office suite dominance with a single application serving 20MM business users, while enterprise incumbents like Salesforce embed AI as features.
Speed or Perfection? Given that AI job postings surged 448% while traditional IT roles declined 9%, can you afford to wait for "perfect" solutions? Successful companies in the report mandate immediate adoption over lengthy pilots.
Data Advantage or Generic Tools? Do you have proprietary datasets or domain expertise that could create defensible AI applications? The report notes "proprietary information will become even more critical" as external data commoditizes.
Agent-Ready Infrastructure? Can your systems support autonomous AI workflows that "book meetings, submit reports, and orchestrate workflows," or are you limited to human-in-the-loop applications? For organizations still building their agent capabilities, we've outlined a practical framework for developing AI agents that can handle these autonomous workflows while maintaining proper governance and control.
The leaders who can answer these questions with specificity and urgency will be the ones building sustainable competitive advantages in the AI era. Organizations looking to accelerate their transition to agent-first business models can explore how Agentic Foundry helps companies build and deploy autonomous AI systems that turn these strategic questions into competitive advantages.
The Opportunity Window Narrows
Meeker's analysis shows that we're quickly reaching an AI inflection point similar to when internet bandwidth became essentially free, or when cloud computing made enterprise software accessible to startups.
The companies moving fastest to build agent-first business models aren't just getting early advantages — they're potentially establishing the same kind of lasting moats that early cloud and mobile companies achieved.
As NVIDIA's CEO Jensen Huang famously warned with characteristic bluntness: "You're not going to lose your job to an AI, but you're going to lose your job to somebody who uses AI.
The infrastructure is ready. The costs have collapsed. The technology works reliably at scale. The question isn't whether AI will transform how business gets done — the economics now make it inevitable.
The question is which leaders will recognize this moment in Meeker's data and move decisively toward building new categories of value creation. Those who do may look back on 2025 as the year they built lasting competitive advantage.
Those who don't may spend the next decade wondering how they missed the signals hidden in plain sight.
Analysis based on BOND "Trends – Artificial Intelligence" Report, May 2025
Agentic Foundry: AI For Real-World Results
Learn how agentic AI boosts productivity, speeds decisions and drives growth
— while always keeping you in the loop.