AI Is Scaling Faster Than Anyone Expected

The premise of the current technology market is simple yet staggering: tech markets are larger than ever, companies are staying private longer, and artificial intelligence is expanding the opportunity set at a pace previously unseen. While the dot-com era and the mobile boom offer historical parallels, the current AI cycle is distinct in its velocity and its foundation.

We are witnessing a unique convergence where infrastructure is being overbuilt by the largest companies in the world, effectively subsidizing innovation for startups. At the same time, consumer adoption has outpaced every historical benchmark. For investors and founders alike, understanding the nuances of this scaling—from infrastructure bottlenecks to business model evolution—is critical to capturing the value being created over the next decade.

Key Takeaways

• Unprecedented Infrastructure Build-out: Big Tech companies are spending roughly $400 billion annually on Capex, primarily for AI infrastructure. This creates a massive tailwind for startups building on top of this capacity.
• Deflationary Input Costs: The cost of accessing models has declined by over 99% in two years, outpacing Moore’s Law, while model quality continues to double roughly every seven months.
• Immediate Global Distribution: Unlike the mobile or broadband cycles, AI rides on existing internet and cloud rails, allowing products like ChatGPT to scale 5.5x faster than Google Search.
• The Shift to Private Markets: High-growth opportunities have migrated away from public markets. Today, 95% of public software companies forecast less than 25% growth, creating a mandate for private market investment.
• Business Model Evolution: While seat-based pricing remains dominant, the long-term opportunity lies in capturing consumer surplus and potentially shifting toward outcome-based monetization.

The Infrastructure Subsidy and Declining Costs

The foundation of the AI boom is being laid in a way that fundamentally differs from previous technological cycles. In the broadband era, telecommunications companies bore the burden of infrastructure costs, often leading to financial ruin before the market could catch up. Today, the wealthiest companies in history—Google, Microsoft, Meta, and Amazon—are shouldering the cost of the build-out.

If you run-rate the capital expenditures from these major tech players, the industry is seeing approximately $400 billion in annual spending, mostly directed toward AI infrastructure and data centers. For startups, this is an incredible advantage: the capacity for training and inference is being built at scale, effectively subsidizing the application layer.

The infrastructure is going to get built for all of the training and inference needs that the market is going to need. And this is great for all the companies that are building on top of this.

Input Costs vs. Model Quality

Simultaneous with this massive build-out is a dramatic improvement in unit economics. We are observing a trend faster than Moore’s Law: the cost of input—accessing these frontier models—has declined by more than 99% over the last two years. Essentially, we have seen a 100x decrease in cost.

At the same time, frontier capabilities are improving by a double factor roughly every seven months. This divergence creates a potent environment for builders. As costs plummet and quality skyrockets, AI is on a trajectory to become a utility akin to electricity or Wi-Fi—an omnipresent resource that powers applications without the end-user constantly calculating the cost of usage.

Energy and Cooling Bottlenecks

While chip production capacity generally scales to meet demand, the physical constraints of this build-out are shifting. Currently, energy availability is the primary bottleneck. This has driven a renewed interest in nuclear power, with major tech companies looking to site data centers near nuclear plants to secure reliable baseload power.

Looking five years out, once energy generation is solved, the bottleneck will likely shift to cooling. The thermal density of next-generation compute clusters will require massive innovation in cooling technologies to prevent hardware meltdowns.

Demand Signals and Consumer Surplus

The demand side of the equation offers arguably the strongest signal that this cycle is durable. AI is built on the back of the internet and cloud computing, which allows for immediate global distribution. There is no need to wait for consumers to buy a new hardware device or for cables to be laid in the ground.

Consider the speed of adoption: It took ChatGPT two years to reach 365 billion searches. It took Google 11 years to reach that same milestone. This accelerated adoption curve is possible because over half the global population already has internet access and smartphones.

The speed at which they got to distribution is unlike anything we've ever seen before. And so that is heartening to me that the supply build out will be utilized maybe in a more predictable way than broadband.

The Monetization Gap

Despite massive usage—estimated between 1.5 to 2 billion active users of AI products globally—direct monetization is still in its infancy. OpenAI, for example, has roughly 30 to 40 million paying users against a billion-user base. This discrepancy highlights a massive "consumer surplus," where the value derived by the user far exceeds the price paid.

This suggests significant pricing power upside. Just as search engines and social networks drastically increased their revenue per user over the last decade, AI platforms have immense room to evolve their business models. We are already seeing early signs of price discrimination, such as lower-cost subscriptions in markets like India, alongside high-end enterprise tiers.

Rethinking Business Models and Margins

Investors are currently navigating a complex debate regarding the gross margins of AI-native companies. Critics argue that heavy reliance on third-party models (like those from OpenAI or Anthropic) depresses margins compared to traditional SaaS.

However, smart capital is taking a lenient view on current gross margins based on the hypothesis of continued cost deflation. Because model providers are in fierce competition, input costs will inevitably trend downward. Therefore, a lower gross margin today is acceptable if the company demonstrates two specific traits:

1. High Gross Retention: Are customers sticking around? Top-tier companies show 90%+ gross retention, indicating deep product-market fit.
2. Ease of Customer Acquisition: Is there organic demand? "Flying off the shelves" demand suggests a value proposition strong enough to weather early margin pressure.

The Future of Pricing

We are in a transition period regarding pricing models. The industry moved from perpetual licenses to seat-based SaaS, and then partially to usage-based cloud models. AI offers the potential for outcome-based or task-based pricing—charging for the work done rather than the software access.

However, this shift is in the early stages. Currently, most monetization still fits into seat-based or consumption frameworks. True task-based pricing requires objective measurement of task completion, which is currently most viable in areas like customer support but harder to implement in open-ended creative or analytical workflows.

Moats, Stickiness, and Disruption

A common question facing founders is the durability of revenue. If a product is just a "wrapper" around an LLM, is it defensible? The answer lies in integration and workflow.

Where Stickiness Exists:

• Medical Scribing: deeply integrated into doctor workflows and EHR systems.
• Customer Support: embedded into brand voice guidelines and escalation rules.
• Enterprise Rules Engines: complex integrations that define how a company operates.

Where Stickiness is Low:

• Code Generation (Raw): Developers can easily swap an API call to switch from one model to another if a better one is released.
• Basic Prototyping: Low-end content generation with no workflow integration.

Disrupting the Incumbents

For a startup to dethrone a massive incumbent like Salesforce, it isn't enough to just add AI. True disruption typically requires a combination of three factors:

1. UI/UX Reimagination: Moving from forms and databases to proactive, agentic interfaces.
2. Data Access Innovation: Leveraging unstructured data rather than just structured records.
3. Business Model Innovation: Changing how value is captured (e.g., selling outcomes vs. seats).

The Great Migration to Private Markets

One of the most profound shifts in the financial landscape is the migration of growth to the private markets. Companies are staying private significantly longer—often 14 years or more—compared to the 5-10 year timelines of the past.

This has left the public markets starved of high growth. Approximately 95% of public software and internet companies are forecasting growth of less than 25%. Consequently, the high-growth, high-alpha opportunities are concentrated almost exclusively in the private sector.

If you want access to the high growth segment of new technology companies, it is all living in the private markets now.

This dynamic forces a change in investment strategy. It is no longer viable to wait for an IPO to capture the bulk of a company's value appreciation. Accessing these companies early—often through relationships established at the seed or Series A stage—is essential for shaping the capitalization table and securing allocation in later growth rounds.

Conclusion

The AI scaling cycle is characterized by a unique combination of massive infrastructure subsidies, rapid cost deflation, and immediate global demand. While risks regarding energy consumption and business model maturation remain, the underlying signals suggest a durable transformation of the economy.

For investors and operators, the focus must remain on identifying applications that leverage this new infrastructure to create deep, sticky workflows. The companies that successfully transition from novel usage to integrated utility will capture the immense surplus value currently flooding the market.