The AI (r)Evolution is not dead
It's just not about who sells the best shovels any more
Back in May, a friend asked me on WeChat if I thought NVIDIA’s share price could keep climbing.
They were curious—not about the usual quarterly earnings chatter, but about the broader picture. “Can they stay on top of this AI boom?” they asked.
I told them I thought they would, at least until one of two things happened:
someone else figured out how to make silicon as good as theirs for training AI models
Or someone cracked the code on training those models more efficiently.
Eight months later, it feels like we’ve hit that second milestone.
This is not bad news for AI, in fact it’s amazing news. The AI race has never really been about who can make the best models. It’s about who can put those models to work in smart, meaningful ways.
The winners in this space won’t be the ones building from scratch—they’re the ones finding creative ways to integrate AI into existing systems or developing tools that let others do the same. It reminds me of the dot-com boom, where companies like Interwoven and Mediasurface were the darlings of the early web. Their valuations soared on the promise of proprietary content management systems that were supposed to transform how businesses managed their online presence. These platforms were complex, expensive, and designed to lock customers into a single ecosystem. For a time, it worked. But as the dust settled, it became clear that the value wasn’t in owning the entire stack—it was in providing flexible, accessible tools.
Eventually, open-source alternatives emerged, offering modular, customizable solutions for free—or at a fraction of the cost. What these open-source tools lacked in initial polish, they made up for with adaptability. They became the foundation for an ecosystem of plugins and integrations that let businesses tailor solutions to their needs. Interwoven and Mediasurface? They were replaced, left behind as the industry shifted toward tools that prioritized accessibility over exclusivity. The companies that thrived were the ones that saw the value in enabling users, not controlling them.
In the same vein, as we get better at developing frontier models, the best open-source ones are always going to learn from the leaders. I don’t agree with Elon Musk much these days, but he’s right that OpenAI was better as a not-for-profit.
So who are the AI companies I think are doing the right thing?
Zapier: The ultimate API factory. Not an AI company as such, but they have the plumbing to connect apps and automate workflows, putting AI in the center of this to find information, trigger actions, and keep things running without constant manual oversight is truly fabulous. I use it to handle most of my incoming newsletters.
MX8 Labs: Full disclosure, I’m a shareholder, but they’ve created a survey research platform with APIs built specifically to work with large language models. This means the AI can do 90% of the survey programming and analysis, reducing effort by an order of magnitude compared to historical platforms.
BOLT.ai: Instead of just offering coding tools, BOLT has gone the extra step of wrapping coding into an infrastructure that supports full application hosting. You can build with AI and seamlessly go from idea to execution, making it faster to iterate. If you’ve got a non-technical consumer proposition you want to build, this is where I’d go.
GitHub Copilot: A tool that doesn’t try to replace developers but works alongside them, adding just enough intelligence to make coding faster and smoother. It’s not the smartest AI, but it’s tightly integrated with a toolset that pretty much everyone uses. It will be interesting to see whether the integrations of Gemini into Google Workspace and Copilot into MS Office have the same impact.
Synthesia: Focussed purely on how to put human-like avatars for e-learning and marketing to work, they’ve turned AI into a tool for better communication—not just flashy tech for its own sake.
Each of these companies focuses on how AI can help in day-to-day tasks. Much of it is pretty unexciting but has a significant business impact.
Meanwhile, the press focuses on what can be done with AI instead of asking what should be done. Automatic screen control by AI? It’s futuristic and flashy, but what problem is it really solving? Platforms already run on APIs, and AI doesn’t need to “see” a screen to do its job. It’s a redundant layer that adds complexity without clear value.
In your AI journey, focus on how you can leverage AI to either do things more quickly or entirely eliminate processes that you previously needed to do. The winners won’t be the people with the highest-ranked large-language model. That will be some Chinese or Indian university, closely followed by a raft of other open-source models.
The winners will be about who can put AI to work. That’s much harder to measure, but that’s where the money will be found.
