Last year, in our Battery Ventures “State of the OpenCloud Report,” we predicted a coming AI “supercycle”—a profound technology platform shift akin to the Internet and cloud-computing revolutions that could unseat large tech incumbents and unlock trillions in market value.

As 2025 draws to a close, we have the benefit of hindsight. We can now definitively say the AI supercycle is here. And it’s so profound that we’ve renamed our annual report to reflect this new reality.

Our new Battery “State of AI Report,” attached below, finds that AI is no longer just a trend. It is the dominant force in the public markets, a massive driver of capital investments and the core infrastructure and foundation of a new class of companies growing at unprecedented rates. But as we look towards 2026, it is clear we are still in the early innings of this transformation. AI scaling laws across pre-training and post-training are holding; AI agents are completing human-level tasks and workflows across industries like software development, healthcare and legal; and a massive wave of computing is just now coming online that will continue to support these tailwinds.

According to our analysis, companies that are AI beneficiaries in the public markets, like Nvidia, Google, and Microsoft, now represent, incredibly, roughly half the value of the S&P 500. Since Nov. 2022, when ChatGPT publicly launched, these innovative companies have added roughly $18 trillion in market capitalization, accounting for around 75% of all the S&P’s gains during that period.

Three years in: Is AI just a bubble or something more?

While there is constant chatter about an “AI bubble,” the fundamentals suggest the current state of play is a rational response to a supply-demand imbalance we haven’t seen since the early internet era more than 25 years ago. While many AI-native company valuations are sky-high, and not every market entrant will make it, we believe much of the hype is warranted. Consider:

• AI is helping reverse the macro-driven slowdown in cloud-provider growth that took hold in 2023, and the three major cloud providers remain a leading indicator of broader AI adoption. Combined run-rate revenue for AWS, GCP and Azure reached $285 billion in Q3’25, growing 29% year-over-year. This compares with 27% growth in the prior quarter and 26% in the prior year, highlighting a clear re-acceleration as AI-driven workloads begin to scale.
• Demand for AI technology continues to outpace available infrastructure capacity, which helps explain the recent surge in capex by large cloud providers. Based on recent earnings, the big cloud providers have $1.2 trillion in revenue backlog, driven by demand for AI applications and infrastructure. This demand for intelligence is outpacing the physical ability to build data centers, suggesting this cycle has a long runway ahead.

2026 Outlook: AI design & usage patterns and the agentic revolution

We are moving past the experimental phase and into the industrialization phase of AI adoption. Distinct architectural patterns for AI-native companies are forming, and we are witnessing a shift from tools that assist humans to autonomous agents that complete work. To drill down:

• While 2024 was dominated by capital-intensive model training, the center of gravity is shifting to inference. As agentic applications—which autonomously plan and execute complex workflows—come online, they will consume vastly more compute power at runtime. We believe this development will continue to drive revenue as a myriad of new agentic applications hit the market. We are also seeing the walled gardens of closed models challenged and believe open models (like that of Reflection AI* and DeepSeek) will grow in adoption, fueling a more diverse AI ecosystem.
• One of the most critical trends we are watching is the change in AI-company profit margin structures, which are slowly shifting. Today, value capture is concentrated in hardware (GPUs) with ~75% gross margins, while AI applications often see compressed margins due to inference costs. However, as the cost of intelligence falls and supply catches up, we expect this to invert. Future value will accrue to the application layer, specifically to those companies that can leverage outcome-based pricing to capture the value of the labor they replace.
• The result is a Cambrian explosion of AI applications we are already seeing in the market, many made possible by the foundation models that power them. This is giving rise to a new stack. In developer tooling, AI-native challengers such as Cursor, Cognition and Lovable are redefining how engineers build and collaborate. In data and analytics, vendors like Databricks*, ClickHouse* and Supabase are becoming the data backbone for intelligent applications. In cloud infrastructure, companies including Baseten, Fireworks and Fal are emerging to serve AI workloads with new abstractions. And at the networking and silicon layer, startups such as Nexthop AI*, Groq, and SambaNova are pushing the boundaries of performance and threatening to disrupt long-standing incumbents.
• In all these cases, this new stack is unlocking AI applications that do the work, creating a much bigger market opportunity than the SaaS 1.0 ever did.

The new playbook for AI founders

These developments and others are also driving our company-building advice for AI founders in the market today. The SaaS 1.0 playbook is officially obsolete and everyone needs an AI story to win funding and succeed in the broader market now. Some of our key recommendations for founders building in the world of AI:

• Start with a focused wedge, land quickly and expand. A targeted use case gets you in the door, and horizontal plus vertical expansion builds the product depth and breadth required for durable moats. Cursor reflects this pattern by landing with code generation and editing, then expanding horizontally across the developer workflow and vertically to own more of the infrastructure layer (including their own model, Composer).
• AI companies should be run against a new set of metrics compared to those of traditional SaaS companies. AI businesses are growing faster as immediate productivity gains compress adoption cycles, but growth alone is not enough. Gross retention and usage patterns are the clearest indicators of whether a use case is delivering durable value or is still experimental. As companies scale, monitoring gross margin also becomes essential to ensure pricing and unit economics remain sustainable in a compute-intensive model.
• Similarly, the old SaaS model of seat-based pricing, and even pure consumption-based pricing, should give way to value-based pricing models. As software shifts from aiding human productivity to autonomously completing work, value-based pricing captures the impact of AI, aligns pricing with successful customer outcomes and allows revenue to naturally scale as customers mature.
• In the SaaS 1.0 era, infrastructure sat behind the scenes while application logic shaped the user experience. In an AI-native world, the opposite is true. Infrastructure is now front and center, and choices around model selection, runtime environments, evals, routing and data access determine system accuracy, responsiveness and the overall product and user experience.

In sum: It’s AI’s world right now, and we’re just living in it. The opportunity has never been larger, and urgency has never been higher. AI founders must act nimbly and adapt to the current market, where innovation is happening more quickly than we’ve ever seen. But those strong enough to build differentiated products and create competitive moats have the potential to realize substantial value.

Loading...

Taking too long?

Reload document

| Open in new tab

Download [4.40 MB]