Table of Contents

• 1. AI-first companies scale faster and generate real revenue
• 2. Enterprise adoption is rising sharply, with growing investment
• 3. Improved efficiency: more capability at lower costs
• 4. AI competition moves into new arenas: search, coding, and browsers
• 5. Infrastructure under strain: power is the new bottleneck
• 6. Talent and labor: the impact is most visible at the entry level
• 7. New AI architectures: modularity, efficiency, and open standards
• In summary

The new State of AI Report, published by Nathan Benaich and Air Street Capital, is now available. Since 2018, it has been one of the most authoritative tools for understanding the evolution of artificial intelligence—not only from a technological perspective but also in terms of its economic, industrial, and societal impacts.

The 2025 edition marks a turning point: artificial intelligence has moved beyond the experimental phase and has become strategic infrastructure for enterprises. It generates significant revenue, accelerates the transformation of industry sectors, and redefines established operating models.

Here are the seven key trends highlighted in the report that every innovation-oriented company should be aware of.

1. AI-first companies scale faster and generate real revenue

AI is delivering tangible business results.

A selected group of 16 AI-first companies reached $18.5 billion in annualized revenue as of August 2025. Companies founded in or after 2022 are reaching $5 million in Annual Recurring Revenue (ARR) 4.5 times faster than those founded before 2020.

Companies earning between $1 and $20 million in revenue are growing at a quarterly rate of 60%; those exceeding $20 million are growing at 30%, both about 1.5 times faster than the sector average.

Verticals such as audio, avatar, and image generation show particular maturity: ElevenLabs, Synthesia, and Black Forest Labs have surpassed $100 million in annual revenue. ElevenLabs, in particular, has doubled its revenue in just nine months, reaching $200 million.

2. Enterprise adoption is rising sharply, with growing investment

AI is rapidly becoming a stable component in business operations.

In the U.S., the percentage of companies using paid AI solutions rose from 5% (January 2023) to 43.8% (September 2025). The tech sector leads with 73% adoption, followed by finance at 58%.

Contracts are also growing in size and duration: retention rates increased from 50% in 2022 to 80% in 2024, while Average Contract Value (ACV) jumped from $39,000 to $530,000 in just two years.

Over 95% of professionals use generative AI in their personal lives, and 92% report significant productivity gains. The most common applications include content generation, coding, research, and data analysis.

3. Improved efficiency: more capability at lower costs

Model capabilities are increasing rapidly, while costs are declining.

The capability-to-cost ratio doubles every six months, giving labs with efficient scaling a strong competitive edge. OpenAI leads in “intelligence per dollar,” followed by Google and other top-tier players.

This trend makes advanced AI applications more accessible even for mid-sized companies, expanding the innovation potential across industries.

4. AI competition moves into new arenas: search, coding, and browsers

Artificial intelligence is reshaping the balance across major digital sectors:

• Search: Tools like ChatGPT have become the first stop for complex queries, research, and coding tasks. ChatGPT holds 60% of the AI search market with around 755 million monthly users.
  
• Google’s decline: Global search traffic on Google fell 7.9% year-over-year—a historic shift.
  
• User acquisition: AI-driven search is emerging as a highly effective customer acquisition channel, with conversion rates increasing from 6% to 11% year-over-year.
  
• Coding: Models like GPT-5 are solving full ICPC-level programming problems. However, startups relying on third-party APIs struggle with fragile margins due to high operational costs.
  
• Intelligent browsers: Assistants like ChatGPT Search and Gemini are turning the browser into a smart operating system, capable of independently navigating and acting on behalf of the user.
  

5. Infrastructure under strain: power is the new bottleneck

AI’s energy demands are growing exponentially.

New frontier models could require data centers with up to 5 GW of power capacity by 2028. Power availability is now a strategic constraint for AI development.

NVIDIA maintains its dominant position in AI hardware, but new players such as CoreWeave and Lambda, as well as custom ASIC chips, are introducing viable alternatives.

Circular mega-deals are also emerging: NVIDIA sells GPUs to AI labs and neoclouds, which reinvest that capital to purchase more compute capacity… from NVIDIA. This self-reinforcing loop may introduce systemic financial risks in the long term.

6. Talent and labor: the impact is most visible at the entry level

AI is already transforming the labor market, especially in the most exposed roles:

• Entry-level positions, such as customer support and junior software development, are experiencing declining hiring rates.
  
• Experienced roles, especially those involving tacit knowledge, remain resilient to automation and are sometimes even expanding.
  
• Talent competition is intensifying: leading AI R&D departments are fighting for top researchers, while restrictive immigration policies risk weakening U.S. leadership and strengthening other regions, notably China.
  

7. New AI architectures: modularity, efficiency, and open standards

The way AI systems are built and deployed is undergoing a major shift:

• The Model Context Protocol (MCP) is emerging as a de facto interoperability standard—akin to USB-C for hardware, enabling seamless integration between models, tools, and data sources.
  
• Small Language Models (SLMs) (1–9 billion parameters) are sufficient for most enterprise tasks, offering up to 30x lower operational costs than large LLMs.
  
• The “SLM-first, LLM-only when needed” strategy is gaining ground as a best practice in AI workflows.
  
• The landscape of AI agent frameworks is rapidly evolving: tools like LangGraph, AutoGen, and MetaGPT are specializing in niche use cases, reflecting a dynamic and still-fragmented ecosystem.
  

In summary

2025 marks a clear transition: AI is no longer an emerging technology; it’s a core driver of growth, competitiveness, and digital transformation.
Understanding these trends and aligning business strategies accordingly is no longer optional for those who aim to lead innovation.