AI Is Growing Up Faster Than The Internet Ever Did

It’s official: AI is being adopted twice as fast as the web and mobile.

In under five years, Gen AI has gone from novelty to necessity. It’s now a mature software category with enterprise-scale momentum and we’re watching the ecosystem shift in real time.

a16z just published a deep-dive based on surveys with 100 CIOs and dozens of enterprise leaders. The takeaway? This isn’t experimentation anymore. AI budgets are being locked in. Procurement is getting serious. Custom builds are giving way to polished, off-the-shelf solutions.

If you’ve been waiting for “real” adoption, this is it.

A few things stood out:

• AI isn’t on the innovation fringe anymore. It’s in the budget. In production. In workflows.

• Enterprises aren’t loyal to one model—they’re building fleets, optimizing for speed, accuracy, and cost.

• Fine-tuning is out. Prompting and context length are in.

• And while many still parrot the “AI bubble” narrative, the numbers tell a different story.

Here are the 10 insights that matter most from the a16z report:

---

1. AI budgets are exploding
Enterprise LLM spend is up ~75% YoY. One CIO put it bluntly:
“What I spent in 2023, I now spend in a week.”

2. GenAI has moved into core budgets
Only 7% of today’s LLM spend comes from innovation budgets. That’s down from 25% last year. Translation: AI isn’t experimental anymore.

3. Multi-model strategies are here to stay
37% of enterprises run 5+ models in production. It’s less about vendor loyalty, more about task performance.

4. Clear leaders are emerging
OpenAI leads in overall adoption. Anthropic rules coding. Google’s Gemini 2.5 wins on cost-performance.

5. Closed-source is gaining ground
Models like Gemini Flash and Grok 3 Mini outperform open-source alternatives for the mid-market, nudging preferences toward closed ecosystems.

6. Fine-tuning is fading
Better prompts and longer context windows reduce the need for heavy customization, unless you’re in a niche domain.

7. Reasoning models are the next frontier
OpenAI’s o3 is already in production at 23% of surveyed companies. DeepSeek is gaining traction, but still early.

8. Model procurement is now traditional software buying
Think RFPs, cost-benefit spreadsheets, workload alignment. AI buying looks a lot like SaaS now.

9. Direct hosting is gaining trust
More companies are bypassing cloud intermediaries to go straight to OpenAI or Anthropic.

10. Switching costs are rising
Agentic workflows require custom prompts and tuning, making it harder (and costlier) to swap models.

---

This is what software maturity looks like: rapid iteration, hardened preferences, and procurement teams asking real questions about ROI.

The web took a decade to get here. Mobile nearly as long. AI? It’s done it in five.

And if the past year is any indication, the next five will move even faster.

---

Want more signal (and less noise) on where AI’s headed?
📬 Subscribe for weekly insights.