Comment by npunt
4 hours ago
Ok I see where you're going. I think we'd both agree that the PC market is mature and unlikely to grow further, their services revenues aren't looking sustainable long term, and that AI demand is high. Good food for thought!
Some facts to consider tho since I think you're underestimating the apple silicon change:
MacOS market share went from 7.5% (2020) -> 9% (2025)
Mac gross margins are harder to estimate but suggest a change from 24% (2016) [1] -> 41% (2025) [2] if you divide profits from revenues. These exclude services.
PC market profit share is >60%, as in Apple get most of the money made in the PC market.
So the idea that apple silicon didn't really move the needle isn't supported. Apple also cares about profit share far far more than market share. Market share can be a liability because of antitrust and supporting less desirable users and use cases. A few decades ago they chose not to get into the Netbook fad when everyone else was because of this philosophy, so its not new.
Apple is certainly ramping up their server hardware as it seems to be angling to run models on their own silicon, and Xcode cloud is a natural extension of that activity. I'm not sure that suggests selling servers or server capacity though. The end result is the same biz model: the people using this server capacity are still primarily individual customers.
The main question about the AI buildout right now is a ton of money is being invested and the returns on that capital are highly speculative. Everyone is investing, sure, but there's a lot of economic headwinds and its unclear where the value will ultimately accrue. If Apple jumps into something, they tend to do it with a decade-long perspective. What you're arguing for is that Apple should have jumped in early, which regardless of B2B v B2C, is not Apple's style.
If apple were to get into that and leverage their silicon advantage, they'd need to significantly increase their supply of silicon. They're already pretty much at capacity serving B2C, and they typically do big capital investments with places like TSMC to build out the capacity they need for nextgen process. Same with ram, probably even more of a limited resource right now. If they were to jump into enterprise in a big way, it wouldn't just be 'make more chips' it would be drastically increasing their capex to expand capacity, and that takes years of lead time to do. The AI buildout is very new, its really all just in the last 2-3 years. I just don't think they could have delivered on this idea right now, based on what they knew in ~2024 and the capacity they have available. If anything their strategy, whether intentional or not, is looking rather prudent. But another month or two passes and maybe that will change too. Things are changing very fast right now.
[1]: https://asymco.com/2016/11/02/wherefore-art-thou-macintosh/
[2]: https://asymco.com/2026/03/18/the-biggest-surprise-of-the-ma...
It just doesn't feel very "skate to where the puck is headed" to me. Apple is holding on for dear life to a business model that has very little momentum. Apple was ready for the AI buildout, which is what upsets me. The rackmount Mac Pro is the perfect product for Apple to sell datacenter customers, all they need to support are CUDA drivers (which are already ARM/BSD-ready) and some sort of PCIe interconnect. There was no "pivot" that was required, outside of their software policy. Maybe AI was a bubble, but missing out on that capital is a bigger problem than suffering from success a-la Nvidia. When AI hype dies down, there's going to be new markets that Apple has to be ready to capitalize on. AI and GPGPU compute have been good exercises for testing Apple's flexibility, but a lot of those advances are flushed down the toilet on hardware like the Mac Studio, which forces customers into non-scalable Thunderbolt RDMA clusters.
I'd argue the supply chain is a non-issue, too. Apple's chips are a high/low mix, and most of them go to TSMC when they really shouldn't. Apple's A-series SOCs could be manufactured by Intel or Samsung without setting them back too far on fab nodes. Mac chips could remain TSMC products, and Apple could pivot their high-memory SKUs to the Mac Pro form factor instead of forcing it into a non-scalable Studio chassis. That would give them a product that the industry could start to buy again, and they could even offer it as a consumer product if they wanted to thumb up their nose at Nvidia. The current Studio is a sad flagship to sell, reminiscent of the Trash Can form factor with even fewer hardware features.
It's Ternus' choice, and he seems to be open to new ideas. Apple isn't immune to bad decisionmaking though, and I bet that future generations will be confused why the ARM datacenter was dominated by non-entities like Ampere and Nvidia selling low-performance licensed cores.