The AI bubble will pop. Intelligence won’t

Yes, it will.

But what won’t pop is intelligence itself. While Wall Street bids up mega-models with billion-dollar burn rates, AI is producing measurable returns elsewhere in transformational, if less glamorous, ways. 

Take Austin, Texas, where an on-premise AI-system helped local government process building permits in days instead of months. No spectacle. No headlines. Just efficiency gains that will outlast the market cycle.

That’s the point too often missed in the frenzy. Mega-models attract headlines, consume billions in capital, and struggle to demonstrate sustainable economics. Meanwhile, smaller, domain-specific systems are already delivering efficiency gains, cost savings and productivity improvements. The smart play isn’t to abandon AI, but to pivot toward models and deployments that will endure.

We’ve seen this movie before. Netscape once symbolized the internet revolution. Its spectacular IPO made headlines; its decline made history. But the collapse of early web darlings didn’t kill the internet, it revealed that the real value lay not in browsers but in the infrastructure beneath them.

AI stands at the same crossroads today. The platforms consumers know best — ChatGPT, Gemini, Claude — are extraordinary feats of engineering, but they don’t represent sustainable AI. They are breathtakingly expensive to run, but free or cheap to use. They deliver entertainment and convenience more than enterprise value. It’s fun to have ChatGPT turn out a poem, nice when it helps you smooth out an email — but it’s not mission-critical.

Economically, the hyperscale model doesn’t hold. Training and maintaining ever-larger systems yields diminishing returns while costs escalate into the billions. That’s why GPT-5 landed with a shrug. Scale alone is no longer impressive.

So what is?

The answer lies in focused deployments. Austin’s permitting office achieved in weeks what bureaucracy had delayed for years. Healthcare systems are running diagnostic models fine-tuned for their specialties that outperform general-purpose LLMs. Financial firms are already relying on BloombergGPT, trained on market data, which delivers better results in its domain than larger consumer platforms. These applications generate tangible ROI and do so sustainably.

The principle is simple: a massive general-purpose model can do many things at a passable level, but it rarely excels. A leaner system, built for a specific function and deployed thoughtfully, can deliver speed and accuracy where it matters most and at a fraction of the cost. This is the strategic, cost-effective way forward: integrate AI in tactical ways that directly serve the business, rather than chasing the illusion of a one-stop shiny new AI technology.

Think of it as staffing a project: 100 average consultants won’t outperform five experts.

Where the data lives matters just as much. Lighter models can be optimized to run locally on edge devices or inside secure enterprise facilities instead of depending on costly, centralized infrastructure. At webAI, for example, we’ve been able to shrink models by nearly a third while preserving accuracy. That changes the economics completely. Instead of routing every query through an expensive cloud data center, intelligence sits closer to the data it serves, making it cheaper, faster, more resilient and more secure. Just as important, enterprises retain ownership of their data and the insights built on it, which is not possible when relying solely on hyperscale providers.

Companies tied exclusively to mega-models are exposed to spiraling costs, energy scrutiny, and security vulnerabilities. Decentralized, specialized AI avoids those traps. IT also offers resilience and puts businesses on firmer ground for the regulatory scrutiny that is sure to come.

With this in mind, smart techno-optimists and AI investors need not panic when headlines warn of an “AI winter.” Yes, some companies will collapse under the weight of unsustainable economics, just as many did after the dot-com crash. But AI itself isn’t going away. It’s evolving toward networks of specialized systems that work together more like a city grid rather than a skyscraper.

For executives, the takeaway is clear: avoid chasing scale for its own sake. Instead, invest in AI systems that are efficient, close to your data, and tailored to specific business needs. Build for sustainability, not spectacle.

When the next AI earnings cycle sends markets into another frenzy, remember Austin’s building permits. The businesses building lean, domain-specific intelligence won’t be watching their valuations with the same anxiety. AI isn’t going back in the box. But the future won’t be bigger at all costs — it will be smarter, leaner and built for staying power.

The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.