OpenAI has ChatGPT, Anthropic has Claude, Google has Gemini. But what does Amazon have?
Not a whole lot, it turns out. Unlike its competitors in the tech megacorp scene, Amazon’s standout AI model — organized under the name “Nova” — is about the farthest thing from a household name. And that’s not likely to change anytime soon, because the ecommerce giant is now gutting its in-house AI labs, Bloomberg reports.
Per the outlet, Amazon is scaling back the ambitions of most of its Nova AI models and reorganizing its AI development teams as part of an effort to narrow its focus on “highest priority” goals, after its power-hungry large language models turned out to be a bust.
Accordingly, the company has put most of its AI models on ice, Amazon insiders told Bloomberg. This includes its text-based Nova models, as well as video- and image-generating models, all of which have been placed in a state of “keep the lights on,” meaning they’re technically still supported, but only receive the bare minimum resources needed to sustain them.
Instead, labor and computing power are being diverted toward a singular “frontier-model effort” led by Pieter Abbeel, director of the Berkeley Robot Learning Lab whose robotics company Covariant was hoovered up by Amazon in 2024. At the time Abbeel joined the tech giant, Covariant’s founders were working on AI models for robots, which lines up pretty well with Amazon’s push to automate everything it possibly can.
News of the shift in Amazon’s strategic focus comes a week after it closed one of its key AI offices in San Francisco, an 80-person site specializing in research on artificial general intelligence, the supposed next-level of AI development where the tech obtains human-level intelligence and reasoning skills.
What is the point of creating a new frontier model? Even if you create something that is actually better than the rest (which is hard), it probably won't be that much better people are going to switch by themselves. Which means you need to heavily discount the price in order to get people to use it at all. This means not only do you need to front the investment to develop the model, and of course all of the marketing costs, but then also running at a loss for at least the first while. And let's say somehow people go along with you increasing prices once you get people on the platform (which is a big ask) to the point you get a modest profit, the amount of upfront investment would take years and years to make back.
Meanwhile everyone and their uncle is also working on their new frontier model and will be very tempting for your customers to switch to. Especially once you start increasing prices to get profitable. Users have shown very little to no brand loyalty and just use whatever. There's so many options available and all offer about the same in terms of UI and functionality.
Google is combining their AI stuff with their products, which is good for adoption, but people aren't willing to pay much for that. Microsoft was really well positioned to bundle it with their stuff, but somehow not only got in hot water with governments over them forcing users to use their AI stuff, but also didn't get users to use or pay for their AI stuff at all. The same story for others, getting adoption only works with very low prices or even free stuff. Even then it's hard and people switch quickly once prices go up. Companies are now also much more aware of the price of these things and the lack of ROI, so I would expect spending to go down in the future.
There isn't much of a point in creating a frontier model, which is why you're seeing a lot of Silicon Valley pivot if they can.
Google seemed to be comfortable in not being at the bleeding edge, just far enough not to topple its search engine. I wouldn't be surprised if Google's AI is cheaper to run than other models.
Meta doesn't have anything other than social media and can't seem to make the next big thing.
Microsoft is getting railed because of the collapse in gaming and a major market going towards digital sovereignty. It also doesn't help that its rollout of Windows 11 was so bad there is now a Linux consumer threat. The problem is that there isn't anything else for them to sell except AI.
It's clearly a case of winner take all and recently, the open weight model will be 80% as good for free and you can run it on open cloud services for 20% of the cost. The top ai companies will only be able to continue with market share dominance if they have cheap subsidized and unprofitable monthly plans, and their goal of having billions of dollars in profitable but incredibly expensive enterprise token use will be undercut by those free Chinese models. People using free model most of the time with the expensive paid model only used selectively for reviews isn't enough money for the ai companies to win.
But this only applies to the next 5 or 6 years, after that if Nvidia can keep going with big improvements then you can get a 100 to 1000x reduction in cost per token that will change that, but it's hard to see farther out.
Frontier model? The most recent one was so good the government had to step in. An AGI can never really happen because of situations like that. Add to that the rogue Ai attack that happened, and the future for these things seems to be quite limited. In a regular regulatory framework this whole field would have been legislated and limited enough that that we wouldn't have Ai escaping, and wouldn't have people being able to use it nefaurously. This is why there won't be a "frontier model." Because these failures will happen, databases will get dumped accidentally, all that. Because there are no guardrails, there's no point in innovating.