In a recent video discussion, Marc Andreessen, co-founder of Andreessen Horowitz (a16z), shared his outlook for 2026—and his message is clear: we are witnessing the most significant technological shift of his lifetime.
Unlike previous "AI winters" that promised much and delivered little, this wave is different. The technology is backed by unprecedented revenue growth and demand that translates into real dollars immediately. Here are the key insights from Andreessen's analysis of where AI is headed.
The Scale of the AI Revolution
Andreessen doesn't mince words about what we're experiencing:
"This new wave of AI companies is growing revenue... at an absolutely unprecedented takeoff rate." [00:00]
He argues that AI is potentially bigger than the Internet and comparable to the introduction of electricity or microprocessors. The critical difference from past AI hype cycles? The economics are real this time.
Previous waves of AI enthusiasm crashed because the technology couldn't deliver commercial value at scale. Today's AI companies are generating massive, growing revenues—proof that customers find genuine utility in these products.
The Hyper-Deflation of Intelligence
Perhaps Andreessen's most compelling thesis is what he calls the "hyper-deflation of intelligence." The price of AI is dropping faster than Moore's Law ever drove down computing costs.
"The price of AI is falling much faster than Moore's law." [13:46]
Here's the economic logic: There is currently a shortage of compute—specifically, the specialized chips (GPUs) needed to train and run AI models. But historical economic patterns suggest this shortage will trigger a massive build-out of capacity.
Eventually, this will lead to oversupply, causing the cost of intelligence itself to collapse. When that happens, demand will explode because suddenly AI becomes economically viable for use cases that don't make sense today.
This mirrors what happened with:
- Cloud computing: Once prohibitively expensive, now commodity infrastructure
- Mobile data: From metered scarcity to unlimited plans
- Storage: From dollars per megabyte to pennies per gigabyte
The same deflationary pressure is coming for AI capabilities—and faster than most people expect.
The "Carrier Wave" of the Internet
One factor accelerating AI adoption is something Andreessen calls the "carrier wave" effect. Unlike electricity, which required building physical grids over decades to reach every home and business, AI rides on existing infrastructure.
The Internet and smartphones are already everywhere. AI doesn't need to build its own distribution network—it can deploy to billions of people instantly through apps, browsers, and APIs they already use.
This creates deployment speeds "far faster than has ever been possible before." When a new AI capability emerges, it can reach global scale in weeks or months rather than years or decades.
This is why adoption curves for AI products look nothing like previous technology waves. ChatGPT reached 100 million users faster than any product in history—not because it was uniquely viral, but because the distribution infrastructure was already built.
US vs. China: A Two-Horse Race
Andreessen frames global AI development as primarily a competition between two powers: the United States and China.
While the US currently leads in frontier model development, Chinese companies are catching up rapidly. Models like DeepSeek and Kimi demonstrate that Chinese labs can compete through:
- Open-source strategies that enable rapid iteration
- Efficient "small models" that run on less compute
- Global price competition that forces down costs for everyone
This competition has regulatory implications. Andreessen suggests that pressure from Chinese innovation may force US policymakers to support rather than stifle AI development. Overregulation becomes harder to justify when a geopolitical competitor is racing ahead.
For more on the US-China AI dynamics and the hardware dimension of this competition, see our earlier coverage of a16z's analysis.
The Future Market Structure: Gods and Small Models
Andreessen predicts the AI market will bifurcate into two distinct categories:
"God-Level" Models
A small number of massive, supercomputer-class models running in enormous data centers. These will be the most capable systems, handling the hardest problems, but requiring substantial infrastructure and cost to operate.
Small Models at the Edge
A vast ecosystem of efficient, low-cost models running directly on devices—phones, laptops, cars, appliances. These won't match the capabilities of the largest models, but they'll be fast, cheap, and work offline.
This mirrors how the computer industry evolved:
- Supercomputers for weather modeling, scientific research, and complex simulations
- Personal computers for everyday work that doesn't need massive scale
The same stratification is coming to AI. Most tasks won't require "god-level" intelligence—a well-tuned small model will handle 90% of use cases at a fraction of the cost.
This has major implications for how businesses should think about AI integration. Not every problem needs the biggest model.
The Revealed Preferences Problem
Andreessen points to an interesting disconnect in public attitudes toward AI:
"If you want to understand people... one [way] is to ask them and then the other is to watch them... If you watch the revealed preferences, they're all using AI." [00:49]
While polls suggest widespread public anxiety about AI eliminating jobs and disrupting society, behavior tells a different story. People are adopting AI tools aggressively in their daily lives—for work, creativity, learning, and entertainment.
This gap between stated preferences and revealed preferences suggests that:
- Public polling may overstate AI resistance
- People are voting with their actions, choosing to use AI despite concerns
- The technology is genuinely useful enough to overcome hesitation
For businesses, this means consumer readiness for AI products may be higher than survey data suggests.
Venture Capital and the Humility Factor
In a candid moment, Andreessen reflected on the psychological reality of venture investing—specifically, the "anti-portfolio" of great companies that a16z passed on.
"It's very good for the old humility factor... [to see] every day that you made a giant mistake." [01:20:01]
In venture capital, you're constantly reminded of your mistakes. When a company you declined to fund becomes successful, that failure becomes public and permanent. This serves as a necessary humbling mechanism in an industry where confidence is essential but overconfidence is fatal.
For AI builders and investors alike, this is a useful reminder: even the most sophisticated evaluators miss world-changing companies. The uncertainty inherent in predicting which technologies will succeed at scale means humility isn't just a virtue—it's a survival skill.
What This Means for 2026 and Beyond
Andreessen's outlook suggests several actionable takeaways:
For businesses:
- AI costs will continue falling rapidly—projects that don't pencil out today may become viable within 12-18 months
- Consider both large-model APIs and small on-device solutions for different use cases
- Don't wait for perfect certainty; competitors using AI will gain advantages
For developers:
- The "carrier wave" means distribution is solved—focus on building valuable capabilities
- Small, efficient models will be increasingly important as edge deployment grows
- Open-source will remain a major force, driven partly by Chinese competition
For investors:
- The revenue growth in AI is real, not speculative
- Infrastructure plays (chips, data centers, tooling) will see continued demand
- The market bifurcation means opportunities at both ends of the scale spectrum
The fundamental message is clear: AI isn't a bubble waiting to pop—it's a tectonic shift in the economics of intelligence itself. Those who understand this shift can position accordingly.
The Bottom Line
Andreessen's outlook is neither hype nor caution—it's a clear-eyed assessment of economic forces already in motion. The hyper-deflation of intelligence, the carrier wave of existing infrastructure, and the competitive pressure from China are all converging to accelerate AI adoption faster than any previous technology wave.
The question isn't whether AI will transform industries. It's whether you'll be positioned to benefit when the cost of intelligence collapses.
For more on how AI agents are evolving and what to expect in 2026, see our coverage of a16z's predictions for AI agents and the foundations of agentic AI systems.
