AI Is Scaling Faster Than Anyone Expected

Key Insights

• AI market expansion is unprecedented: With over $1 trillion in planned investments from 2023-2027 and major tech companies spending ~$400 billion annually on AI infrastructure, the opportunity for AI companies is vastly larger than previous technology cycles
• Adoption speed is revolutionizing distribution: ChatGPT reached 365 billion searches in 2 years—5.5x faster than Google's 11-year timeline—demonstrating how AI built on existing internet and cloud infrastructure enables immediate global distribution
• Input costs are plummeting while capabilities soar: AI model costs have dropped by over 99% in two years (exceeding Moore's Law), while state-of-the-art capabilities double every seven months, creating an ideal environment for building new products
• Massive monetization opportunity exists: With ~2 billion free users of AI products but only ~40 million paying customers, there's enormous untapped potential—comparable to Facebook and Google's 8x monetization increase over a decade
• Private companies are staying private longer: The aggregate market cap of unicorn-valued private companies reached $3.5 trillion (10-12% of NASDAQ value), up from $500 billion a decade ago, as high-growth opportunities increasingly exist in private markets
• Business model innovation is accelerating: Companies can now implement price discrimination strategies previously impossible, with subscription tiers ranging from $3-4/month in developing markets to $200-300/month for premium offerings

The AI Infrastructure Revolution: Building at Unprecedented Scale

The emergence of AI represents the most significant market shift since the growth fund's inception. This isn't merely an incremental advancement—it's a transformational technology that's expanding addressable markets in ways we haven't witnessed before. The infrastructure being built for AI is fundamentally different and vastly larger than previous technology cycles, with estimated five-year investments from 2023 to 2027 exceeding $1 trillion. However, this figure likely underestimates the true scale of the opportunity ahead.

Major technology companies are directing unprecedented capital toward AI infrastructure and data centers, with annual expenditures around $400 billion. This massive investment by established players creates an enormous tailwind for companies building applications on top of this foundation. The beautiful aspect of this dynamic is that large tech companies are bearing the burden of infrastructure buildout and possess sufficient financial strength to withstand potential capacity overbuild. This means portfolio companies leveraging AI infrastructure benefit from a stable, continuously expanding foundation without having to invest heavily in foundational infrastructure themselves.

Simultaneously, the economics of AI models are improving at breathtaking speeds. Input costs for accessing state-of-the-art models have plummeted by more than 99% over just two years—a rate that far exceeds Moore's Law. At the same time, the capabilities of these models are doubling approximately every seven months. This combination of rapidly declining costs coupled with dramatically improving quality creates an unprecedented environment for building new products and capabilities. We believe AI will eventually become as ubiquitous and accessible as electricity or Wi-Fi, fundamentally changing how businesses and consumers approach problem-solving.

The market opportunity represented by AI dwarfs even the massive opportunities created by previous technology cycles. The mobile and cloud computing era generated approximately $10 trillion in new market value across software and internet companies. However, AI's potential impact appears substantially larger due to its economy-wide applicability. Consider this: US enterprise software spending represents only about 1% of GDP, but US white-collar payroll represents approximately 20% of GDP. This dramatic difference illustrates AI's enormous addressable market for augmentation, cost savings, efficiency improvements, and task automation. Even if companies capture only 10% of the value generated by these advancements, with such a vast market, this percentage translates into massive market capitalization growth.

The Monetization Paradox: Capturing Value in an AI-Driven Economy

Understanding how value will be captured in the AI era requires examining how technology companies have monetized users historically. When Google and Facebook began building their empires, few recognized the monetization potential of their user bases. Today, platforms like Google monetize users at approximately $200 per year, while Apple captures significant value through hardware sales and ecosystem lock-in. Yet when we examine the actual value delivered, the surplus is staggering.

Consider Google's services: users receive world-class search, email through Gmail, and increasingly sophisticated AI capabilities—all free. The company monetizes each user at roughly $200 annually, yet the value delivered far exceeds this amount. This massive surplus creation represents the core story of technology advancement. The key insight is that historically, approximately 90% of value generated by technological breakthroughs flows to end customers, while companies providing services capture the remaining 10%.

This principle applies powerfully to AI. While the value created will be enormous, most of it will benefit end customers through cost savings, time savings, and improved capabilities. However, the 10% captured by service providers in such a massive market still represents unprecedented opportunity. Imagine if Google had recognized a decade ago that users would willingly pay $100-200 monthly for a magical, delightful product—the company's market capitalization would likely be substantially higher than today, with even greater market dominance than it currently enjoys.

AI business models are now structured in ways that enable greater success through sophisticated monetization strategies. OpenAI's international expansion demonstrates this clearly: their India subscription product costs approximately $3-4 monthly, reflecting local purchasing power, while US offerings priced at $200-300 monthly are selling exceptionally well. This price discrimination capability—previously impossible to implement cleanly at scale—represents a significant competitive advantage. Companies can address their entire user base with appropriately priced offerings, from freemium tiers monetized through advertising or affiliate systems to premium subscriptions capturing high-value users willing to pay substantially more.

To understand this opportunity concretely, consider using advanced AI research products to perform sophisticated shopping research. When tasked with finding the optimal baseball bat with specific performance specifications and year-over-year value comparisons, these tools deliver extraordinarily accurate, helpful results—far superior to traditional Google searches cluttered with sponsored links. This superior experience justifies premium pricing for power users while free tiers serve broader audiences.

The current monetization landscape demonstrates the scale of opportunity. OpenAI currently has approximately 30-40 million paying users, with other platforms adding another 10 million, totaling around 40 million paying customers globally. In contrast, approximately 2 billion people currently use AI products for free. When compared to platform monetization rates—Facebook and Google monetize US users at $150-200 annually—the opportunity becomes obvious. With 2 billion free users and current premium pricing reaching only 40 million paying customers, the monetization potential is immense.

Consumer engagement metrics further validate the value being created. Daily active users of ChatGPT spend approximately 28-29 minutes per day on the product, comparable to Instagram's 50 minutes and TikTok's 70 minutes. This level of engagement indicates both significant real-time value creation and the potential for substantial monetization through various pricing models. The combination of massive user bases, high engagement, and minimal current monetization creates what may be the largest untapped monetization opportunity in technology history.

Distribution Speed: AI's Unfair Advantage Over Previous Technology Cycles

Previous technology cycles—the internet era and mobile revolution—required years to build adoption momentum. However, AI's distribution dynamics are fundamentally different, creating speed advantages that have surprised even seasoned observers. The comparison between ChatGPT and Google illustrates this vividly: ChatGPT reached 365 billion searches in two years, while Google required eleven years to reach the same milestone. This represents 5.5x faster adoption—a staggering difference with profound implications.

This acceleration stems from AI's structural advantages relative to prior technology waves. Unlike the internet, which required global broadband buildout, or mobile phones, which necessitated manufacturing hundreds of millions of devices and convincing consumers to adopt new hardware, AI builds directly on existing infrastructure. The technology leverages the globally distributed internet and cloud computing infrastructure already in place. Because of this foundation, the entire world immediately gains access to AI capabilities through simple internet connections, without requiring new hardware or infrastructure investments.

The result is unprecedented global adoption velocity. Recent statistics reveal that approximately 1-2 billion people actively use AI products, with another billion having tried AI tools at least once. Across multiple AI platforms, well over half of the global internet population has already engaged with AI tools in some form. This rapid penetration demonstrates how quickly adoption spreads when products are accessible, valuable, and require no hardware investments or infrastructure buildout.

This distribution advantage has profound implications for the demand side of the AI infrastructure equation. Previous infrastructure buildouts—particularly broadband in the early internet era—proceeded without clear signals about user demand. Companies built capacity hoping demand would materialize. In contrast, AI's immediate global distribution provides clear, real-time demand signals. We can already observe which applications users find valuable, how much time they spend using them, and what they're willing to pay. This dramatically reduces execution risk for companies building AI applications and infrastructure.

The speed of adoption also reduces the timeline to demonstrating market viability. Rather than waiting years for manufacturing and distribution channels to develop, or for network effects to gradually build, AI companies can reach billions of users in months. This compressed timeline enables rapid iteration, faster feedback loops, and quicker pivots when necessary. For companies building on AI infrastructure, this means faster paths to product-market fit and dramatically accelerated revenue growth trajectories compared to previous technology generations.

The AI Stack: Where Value Accumulates and How Companies Win

Understanding where value accumulates in the AI technology stack is essential for investors and entrepreneurs. The stack comprises multiple layers: foundation models (like GPT-4 or Claude), infrastructure enabling model training and deployment, and applications built on top of these models. Each layer presents distinct business model characteristics, competitive dynamics, and value capture opportunities.

Foundation model companies like OpenAI, Anthropic, and xAI occupy a unique position. These companies require massive capital investment in research, development, and infrastructure, but they've built direct relationships with end users and businesses willing to pay premium prices for access. OpenAI's ability to command $200-300 monthly subscription prices from power users demonstrates this value capture capability. These companies benefit from strong network effects—as more developers build on their platforms, the ecosystems become more valuable to all participants.

However, foundation model providers face distinct challenges. The B2B developer market shows relatively low switching costs when superior alternatives emerge. If Anthropic releases a superior coding model, developers using OpenAI's API can switch with minimal friction—often a single line of code change. This dynamic means foundation model companies must continuously innovate to maintain market position. They're transitioning from research-first organizations to capital-efficient businesses requiring clear financial returns on R&D investments. No longer can they invest in research without demonstrating tangible business payback.

For applications built on foundation models, success depends critically on how deeply integrated they become into customer workflows. Some applications demonstrate exceptional stickiness. Medical scribing tools, for instance, become deeply embedded in doctor workflows, with proprietary integrations and customizations creating high switching costs. Customer support solutions and high-end financial analysis tools similarly demonstrate strong stickiness, particularly when they incorporate company-specific rules, workflows, and data integrations.

Conversely, many AI applications lack stickiness because they leverage only the emergent capabilities of models without deep integration. Tools used for prototyping, internal experimentation, or simple content generation face intense competition and low customer loyalty. The market will likely bifurcate: some tools serve prototyping and experimentation purposes, while others form durable businesses by becoming embedded in critical workflows.

A crucial insight for evaluating AI application companies involves understanding the nature of their value proposition. Medical scribing doesn't succeed because it uses the latest model—it succeeds because it deeply integrates with doctor workflows, handles regulatory compliance, manages data security, and provides domain-specific customization. A startup cannot simply recreate Salesforce.com by wrapping it in AI; they need fundamental innovations across three dimensions: reimagined user interfaces and workflows, innovative access to new data sources, and novel business models. Only companies achieving innovation in all three areas can successfully disrupt established enterprise software companies.

When Companies Exit: Private Markets, DPI, and the Timing Question

The trend of companies remaining private significantly longer than historical precedent creates both opportunities and challenges. Historically, companies would pursue public offerings within five to ten years of inception. Today's most successful companies—despite growing much faster and improving more rapidly—are remaining private for extended periods, often reaching fourteen years or beyond before considering IPOs. This trend shows signs of continuing and potentially accelerating.

The aggregate market capitalization of private companies valued above $1 billion has reached approximately $3.5 trillion—equivalent to 10-12% of the NASDAQ's total value. Just ten years ago, the entire private market capitalization in this category was merely $500 billion. This sevenfold growth directly reflects the reality that top companies are choosing to remain private for extended periods, concentrating high-growth opportunities in private markets rather than distributing them across public exchanges.

This development is logically connected to broader market dynamics. Currently, only about 5% of public software and internet companies project more than 25% growth over the next twelve months. This means 95% of the publicly traded software and internet market grows at slower rates. Consequently, anyone seeking exposure to high-growth technology companies must increasingly look to private markets, as this is where the most exciting opportunities concentrate. This trend shows no signs of reversing in the foreseeable future, though exceptional companies like Figma may eventually pursue public listings.

For growth-stage investors, extended private company timelines create the challenge of DPI (Distributions to Paid-In capital), a crucial metric for LP compensation. However, this challenge has spurred innovation in private market mechanisms. Tender offers in private markets now enable employees and early investors to realize liquidity without requiring full exits. Furthermore, leading companies have adapted compensation structures to compete for talent despite longer paths to liquidity, implementing regular tender offers and creating structures more similar to public company equity compensation.

The strategic approach involves balancing investor returns with company interests. When companies have legitimate strategic reasons to remain private—such as avoiding public market pressures, maintaining focus on long-term research, or operating in industries with complex regulatory environments—supporting that choice can ultimately maximize value creation. The goal isn't forcing unnatural exits but rather creating structures that enable talented companies to operate optimally while generating appropriate investor returns through various mechanisms.

Evaluating AI Businesses: The Metrics That Matter Most

When assessing AI application companies, certain metrics provide clearer signals of sustainable competitive advantage than others. The primary focus must be the value proposition to customers: Do users genuinely love the product, and is that affection enduring? This question is foundational to everything else.

Gross retention rate serves as the key metric for evaluating core value proposition strength. This metric answers: of 100 customers, how many continue using the product? The target is over 90% retention, indicating customers find sufficient ongoing value to retain. While net retention rate (accounting for expanded usage and upsells) is also important, gross retention provides the clearest signal about core product value. When customers voluntarily continue paying for products, it indicates genuine value creation rather than forced switching costs or lock-in.

Another critical dimension is ease of customer acquisition. Strong products with genuine user demand typically demonstrate organic customer generation, word-of-mouth expansion, and sales efficiency that generates substantial customer lifetime value relative to acquisition costs. When customers actively seek out products and are willing to pay premium prices, it signals a robust business model with strong top-line quality and pricing power.

Regarding gross margins in AI-native companies, the current debate in the market requires nuanced analysis. Our fundamental hypothesis is that as multiple model providers offer comparable quality, input costs will decrease significantly over time. We've already observed a remarkable 100x decline in input costs over two years, and this trend should continue as healthy competition persists among model providers. Companies like OpenAI face competition from Anthropic's Claude, Google's Gemini models, and others—competition that exerts pricing pressure and drives innovation.

Because we believe input costs will continue declining substantially, we evaluate AI application companies differently than mature SaaS businesses. Rather than penalizing companies for currently moderate gross margins, we focus primarily on gross retention and customer acquisition efficiency, giving companies the benefit of the doubt regarding margin expansion. As models improve and companies can deliver more value without raising prices, margins will naturally improve while simultaneously increasing customer value.

This approach requires confidence in several assumptions: that multiple competitive model providers will persist, that input costs will continue declining, and that companies will effectively translate improving models into better products and stickier customer relationships. Recent developments validate these assumptions. GPT-5 serves as a credible alternative to Anthropic, Google's Gemini models demonstrate promising progress in areas like coding, and emerging providers continue entering the market. With this competition persisting, input costs should continue their downward trajectory, validating our approach.

Portfolio Construction: Balancing Certainty and Possibility

Building an effective portfolio in the current environment requires navigating the tension between two distinct opportunity categories: established high-momentum companies and exceptionally talented early-stage teams. The first category includes companies like Cursor, Decagon, and others demonstrating undeniable momentum—"flying off the page" with adoption, revenue growth, and market impact. These companies offer clearer near-term opportunity but potentially more limited upside scenarios.

The second category represents asymmetric bets on exceptional teams—specifically targeting the absolute best teams globally, regardless of current business metrics. This approach recognizes that during periods of rapid technological change, exceptional founders can create massive value even if initial business concepts evolve. These teams raise capital for growth earlier than typical early-stage companies, creating higher business variance, but the inherent quality of the team provides meaningful downside protection.

This distinction between capital risk and business risk is crucial. While business outcomes may be highly uncertain for exceptional teams working with emerging technology, the team's fundamental quality, track record, and judgment provide confidence that capital won't be permanently destroyed. If the initial business direction doesn't succeed, these teams will pivot effectively, potentially creating value in entirely different directions. This asymmetric profile—high business variance combined with lower capital risk—creates attractive return potential.

Portfolio composition also reflects beliefs about future opportunity location. The largest anticipated areas of focus remain AI infrastructure and AI applications, where the market opportunity and talent concentration are highest. American Dynamism represents another significant opportunity area, where we're observing both substantial market demand and a surge of talented founders departing from companies like SpaceX and Palantir. These founders bring deep domain expertise, sophisticated go-to-market capabilities, and understanding of complex sales dynamics, positioning them well for success in defense and national security markets.

AI-enabled healthcare applications are beginning to show interesting developments, though this area currently represents somewhat less focus than AI infrastructure and applications. Early successes in medical scribing and clinical documentation demonstrate how domain-specific AI integration can create substantial value. We anticipate continued activity in this space as more founders recognize healthcare's unique characteristics: massive TAM, regulatory moats, high customer switching costs, and strong customer willingness to pay for efficiency improvements and improved outcomes.

The portfolio approach avoids rigid allocation targets around new investments versus follow-ons, recognizing instead that "best ideas win." When we identify exceptional opportunities offering attractive risk-adjusted returns, we pursue them. If the early-stage team's sourcing generates compelling follow-on investment opportunities, we pursue those. The measure of success isn't hitting predetermined allocation percentages but rather deploying capital into the most attractive risk-adjusted return opportunities available.

The Path Forward: What Success Looks Like in AI Markets

Success in AI markets looks fundamentally different from previous technology cycles. The compressed timelines mean companies now reach $10 million ARR and $100 million ARR approximately four times faster than historical precedent. Evaluating company performance requires real-time market context—understanding how peer companies are performing, what metrics indicate exceptional versus adequate progress, and what growth rates constitute "great" in the current environment.

This real-time calibration requires immersion in the market, seeing companies across different categories and meeting founders building in the space. Without this deep engagement, it's impossible to accurately assess what "great" looks like at any given moment. The metrics that indicated exceptional performance in 2023 may represent merely adequate performance in 2024 as benchmarks shift based on overall market dynamics and the achievement of peer companies.

For growth investors, the opportunity lies in identifying companies that will become category leaders across the expanding AI landscape. This requires evaluating not just current metrics but trajectory, team quality, product quality, and market conditions. Companies demonstrating sustained momentum across multiple quarters, strong customer retention, and expanding total addressable market show clear signs of building durable, valuable businesses.

The business model questions that will define success differ from previous eras. Rather than debating whether pricing should be seat-based versus usage-based, the more fundamental question involves how companies will monetize the value created by AI-driven automation and augmentation. For customer support automation, the answer is becoming clearer—companies can monetize based on completed interactions or resolved tickets. In other domains, the optimal monetization model remains emerging.

What's clear is that value captured won't depend on complex pricing mechanisms but rather on the fundamental value delivered to customers. Companies that create genuine customer love, substantial time savings, cost reductions, or capability improvements will find effective monetization paths. Those relying on feature-based bundling or complex negotiation will struggle. The companies that win will be those that create so much value that customers enthusiastically recommend them and willingly pay for continued access.

Conclusion

The AI revolution represents an unprecedented opportunity not just for technology companies but for the entire economy. The convergence of massive infrastructure investment, rapidly declining input costs, dramatically improving model capabilities, and immediate global distribution creates conditions unlike anything we've witnessed in previous technology cycles. With an estimated $3.5 trillion in private market capitalization concentrated in high-growth companies, and billions of users already engaging with AI products, we're in the early innings of a transformation that will define the next decade of technology innovation and value creation.

For investors, founders, and everyone navigating this transformation, the key insight is simple yet profound: AI is scaling faster than anyone expected, and the opportunity set is larger than previous generations imagined. The companies that recognize this, invest accordingly, and position themselves to capture value in this new landscape will define the next era of technology dominance.

---

Original source: AI Is Scaling Faster Than Anyone Expected

powered by osmu.app