The Race for AI Independence
Every single day brings a new headline. Today's: Amazon steps up effort to build AI chips that can rival NVIDIA. This follows news a few days ago that Amazon was dangling billions of dollars more in investment into Anthropic, if they would commit to using these new chips. Anthropic is said to be reluctant given how wedded they are to NVIDIA's stack for their models.
Amazon is not alone here, of course. Google was early in building out their own chips. Microsoft is working on their own chips. Microsoft's partner OpenAI is also working on their own chips. Apple is already using their own chips and building more. Every AI company in China now basically must work on their own chips given the import restrictions. Meta seems perhaps most wedded to NVIDIA given the love-fest between Mark Zuckerberg and Jensen Huang on various stages around the world. But of course, they're also working on their own AI chips.
This is hardly surprising simply given the costs associated with NVIDIA's chips and the ramp that every single technology company is experiencing during our current AI revolution. This has quickly propelled NVIDIA from a key player in computer graphics to the most valuable company in the world in the span of mere months. Everyone (aside from perhaps Apple) is buying as many NVIDIA chips as they can get their hands on – even if it means not buying them directly from NVIDIA themselves, but in some cases from rivals.
And the spend. Oh, the spend. Amazon recently said it's going to spend $75B on capex this year, with much of it going towards the AI build up and out. The rest of Big Tech is in similar boats. With NVIDIA working on a yearly cadence for new chips, costs are now arguably scaling faster than AI itself. At some point, Wall Street is not going to like that too much. Probably sometime soon. And so, Operation: Independence is on.
But these chips are just one element of independence being sought in AI.
Microsoft now seems firmly committed to decoupling from OpenAI, the startup in which they've invested nearly $14B and have also ridden to all-time highs in terms of stock price. What started as hedging after the attempted coup a year ago now seems far more about ensuring their own AI technology is produced in-house.
Even GitHub, whose partnership with OpenAI, without question, has been a huge success from both a product and, importantly, business perspective for Microsoft is now diversifying.
On the flip side, OpenAI has been trying to break their own dependence on Azure (and NVIDIA chips) by securing their own deals outside of the Microsoft cloud.
OpenAI is also now building out their search product, in part with the help of Bing, but also on their own to be able to fill in that inevitable hole. Meta, meanwhile, is also apparently now building up their own search product (again) so as not to have to strike deals with Microsoft or Google on search for their AI products.
Apple announced their partnership with OpenAI at WWDC but before that tie-up has even formally launched, clearly there are some tensions between the two sides. And that may or may not lead to iOS granting access to other third-party AI functionality sooner rather than later. At the same time, that integration is likely to cause a huge spike in usage and, as such, money-making capabilities for OpenAI. Which Microsoft won't love as they'll be getting paid for it in part with money/credits they've given OpenAI...
But Apple, of course, famously aims to own their own stack in everything they do. To think they're going to rely on third-parties for AI indefinitely is to think that AI is not going to be a big deal. Apple is doing what they almost always do: partnering at first to get off the ground while they build their own capabilities behind the scenes. They're seemingly behind right now, but as with everything Apple, there's so much going on behind (beautiful and literal) curtains that it's hard to know for sure. Does Apple have their (excellent) in-house chip team focused on AI at the moment? I mean, they must, right?
The difference, this cycle, is that everyone is quickly joining Apple in such a crusade for independence. Again, I think because AI is evolving so fast and more so because the costs are ramping so fast with NVIDIA cornering the market so fast, all of Big Tech quickly realized they needed not just to diversify, but to try to gain "AI independence", as it were.
Google is probably best positioned here right now given all the ingredients they've long had, including, as mentioned, those TPUs. But they need to figure out how to turn a weakness – everything being so wedded to the Google ecosystem – into a strength. So far, that has seemingly not helped Gemini, but if Gemini gets good enough to help Google, that could change equations. Might the current morphing of AI we seem to be experiencing help them (or hurt them)?
Meta talks their big "open source" game, and while even though the powers-that-be disagree with their definition, there's no question that their approach has been beneficial in spreading Llama far and wide. They also have interesting AI conduits with their Ray-Ban smart glasses and, eventually, the 'Orion' true AR glasses. But right now it feels like they're simply shoving AI in the faces of their multi-billion user base. And that tends not to work without actual utility – just ask Google+.
Amazon seemingly is in need of help on a few fronts with regard to AI. These new Trainium chips sound good on paper, but again, there's a reason that partners may be reluctant to use them. CUDA is a hell of a drug. Alexa currently seems decidedly less "remarkable" and more of a mess as well, she's still nowhere to be seen despite early promises. Her baggage may indeed be holding Amazon back, but also the fact that there are so many different AI initiatives serving different masters and purposes can't help. They seemingly don't have the partnership problems with Anthropic that Microsoft does with OpenAI, but they're also not benefitting as much from that partnership, it seems. At least not yet.
Speaking of, while Microsoft is clearly trying to do their own things apart from OpenAI, there's no indication that any of it is actually working. It's early, but the sounds of complete and utter user silence are deafening. Whereas Apple and Meta hold the consumer keys, Microsoft has the ability to make AI happen in enterprise (with Google sort of in between). But they crave that consumer mineral. Always have. Always will. They got a taste with OpenAI, can they taste success without them? Also, should they work closer with Intel on AI chips?
Back to NVIDIA, they're in a great position now, to say the least. But might that change if we start to shift from training to inference? Or to different AI build-outs and capabilities that bespoke chips are better suited for? They're in a race of their own to become Intel, before becoming what Intel has become.