The AI Camps — Calling the Roll

The AI revolution driven by large language models is going to change humanity as much as the smartphone did — maybe more. After the carpet-bombing of AI news from March through early April, some very preliminary observations are starting to emerge about what comes next, and where the investment opportunities might lie.

OpenAI's hundred-million-user lead, with Google in hot pursuit

Right now the camp that matters most is OpenAI-Microsoft. This camp is out in front by the widest margin, and it may well end up laughing last. The main reason is that OpenAI already has upwards of a hundred million users, all diligently helping train OpenAI's models without even realizing it. Large models touch on society far too broadly — regulation, job displacement and retraining — so this was never a purely technical problem. If you watch interviews with OpenAI's boss Sam Altman, you'll notice he has a genuinely deep grasp of society and philosophy. At least so far, he's managed to steer the product while having clearly thought through the various social problems that come with pushing technology this far ahead — he's not just some garden-variety tech nerd. That may well be the deeper foundation of OpenAI's long-term lead, because commercial applications at this planetary scale require grappling with genuinely massive social problems.

OpenAI's biggest rival is the Google camp. Google is generally regarded as the strongest single force in AI — if OpenAI hadn't come out of nowhere, everyone would have bet that Google would eventually be the one to build artificial general intelligence. Even the foundational concept behind this AI revolution, the Transformer, came out of Google.

But in practice, there's a gap between Google's performance and OpenAI's. The main reason is that Google's existing businesses are simply too successful — this AI revolution has waded into deep, murky social territory, and Google has a lot of non-technical considerations weighing on it. That said, Google isn't that far behind either. It still holds more data than anyone else, so it's very much still in the game.

I'd also flag two other camps worth watching: Twitter, and Meta's open-source Llama camp. Twitter's owner Elon Musk, in between building rockets and Teslas, has also announced he's building AI. Meta, meanwhile, has pushed out Llama, whose results are already clearly strong — and open-sourced it on top of that.

Unlike the popular impulse to dismiss Twitter as a source of nothing but bizarre tech headlines, I actually think Twitter's AI opportunity is a good one. The nature of Twitter's data is excellent — extremely well suited to training large models — and right now, at this very moment, countless people are uploading content to it, which is exactly the raw fuel large models need. Meta's open-source model, meanwhile, has been proven in practice to be remarkably effective, genuinely filling gaps that OpenAI can't address. Plenty of developers are already using it and have shipped real breakthroughs with it. On top of that, Meta's business instincts are the most aggressive of the bunch — it's willing to ship more radical AI products, and that kind of approach is exactly what could let it cut the corner and overtake the leaders.

Twitter is worth watching; the open-source camp led by Meta has real promise

Now look at Amazon, Apple, and the rest. Amazon makes its living selling cloud services, so it's mainly partnering with open-source models. Apple hasn't made any major announcements yet. Other national-level programs (Taiwan included) are essentially building backup plans — the goal isn't to displace the Silicon Valley giants, it's mostly about hedging against the risks of tech-giant monopoly. The other, smaller tech giants (a contradiction in terms, I know, but relative to the true leaders, everyone else falls into this bucket) are running the same backup-plan logic.

That raises a question: how much benefit is there really in using open-source models? Right now, the biggest advantage of open-source models seems to be that you deploy once and don't have to keep paying the big vendors afterward. But outside of enthusiasts and a handful of niche use cases, there are still plenty of problems to solve commercially. Because of gaps in data volume and compute, open-source models still trail the true large-vendor models by a wide margin. Still, open-source has one advantage: once it hits a certain threshold of quality, the gap versus a big vendor's model becomes merely "good" versus "very good" — and at that point, the cost and deployment advantages of "good enough" can let it catch up from behind. And some open-source large models have a big vendor backing them from behind the scenes anyway — Meta, for instance — so this fight is far from over.

Weighing all the big players together: the OpenAI-Microsoft camp should, on paper, have been trailing by a wide margin. By data volume, Google and Apple have the most. By customer count, Meta leads. By cloud compute, Amazon is right up at the top. Amazon also has voice data, and Google and Apple between them hold essentially all the data there is — OpenAI shouldn't even be in the conversation. Even with ultra-wealthy Microsoft in tow, the combined data on hand still doesn't match the other giants.

It could have gone even worse. OpenAI's main early backer, Elon Musk, split with OpenAI's boss Sam Altman. By Altman's account, Musk felt OpenAI was falling too far behind to keep going. That call wasn't necessarily wrong — in the early days, the budget gap between OpenAI and Google DeepMind was as much as fifty-fold. That's fifty times the machines and fifty times the people — how could you possibly compete against that?

But at a point when things weren't looking great for OpenAI, Altman made two calls that mattered. One was allying with Microsoft to solve the money problem. The other was picking the right technical path — pushing the large language model direction to its breaking point. A lot of AI luminaries looked down on that path at the time, but that one correct choice ended up shaping everything that followed. Microsoft, for its part, arguably had nothing going for it except money — but understanding its own weaknesses, refusing to shut itself in a room and tinker on its own, and correctly finding a small outfit like OpenAI to partner with in order to cut the corner and overtake everyone else — that, too, is a genuinely impressive call by Microsoft's boss Satya Nadella.

I think of this as a classic David-and-Goliath story (except this particular David is whale-sized — it's just that the other whales are even bigger). At least in large language models, for now, the leading pack looks set to keep pulling away from everyone else in stages — and that's something you can plan your investments around.