During a Gold Rush, Sell Shovels

For a long time, the general narrative around Apple has been brutal: they missed the AI boat. Critics laughed at Siri’s underwhelming performance, pointed at the lack of its own frontier AI model, and the recent news that Apple is partnering with Google to power Siri’s next-gen capabilities. To many, Cupertino seemed hopelessly behind in a world where everything has to “be about AI”..

They seem to have missed the quiet plan that was slowly taking shape..

Apple isn’t losing the AI race – it is waiting it out,  by opting out of the direct fight for AI dominance, which consumes billions on AI infrastructure and compute,  and quietly building the complimentary hardware ecosystem instead. While AI labs and tech giants burn billions training ever-larger models in massive data centers, Apple has just executed a masterstroke with its latest hardware release: the new M5 Max and M5 Ultra Mac Studios.

Instead of competing to build the strongest AI software, Apple is selling high-margin pickaxes to today’s data miners – the growing crowd of AI tinkerers and hackers that don’t want to rely on API access to frontier models, but prefer to keep their inference on prem. 

The killer feature of the new M5 Ultra Mac Studio isn’t raw compute power, but its Unified Memory Architecture (UMA). The new flagship desktop can be configured with an astonishing 512GB of unified memory, enough to fit even very large LLMs, especially in quantised format. Having the possibility to run DeepSeek-V4-Pro on your local machine is simply mind-boggling.  Even though this comes at a hefty price, it is relatively low compared to the alternatives, such as stacks of GPUs in dedicated cases. And don’t even begin to compare the power requirements!

For AI researchers, engineers, and developers, this suddenly opens a whole new set of possibilities:

• No More VRAM Bottlenecks: Traditional GPU setups struggle with VRAM limits when loading massive Open Weights models locally.
• Massive Local Inference: Apple’s unified architecture treats system RAM as unified video memory, allowing developers to run enormous LLMs locally without needing a server farm.
• Zero Real Competition: In the desktop market, there is virtually no other single-box solution that offers this memory capacity and bandwidth at a fraction of server hardware costs.

So while everyone else spends billions digging for AI gold, Apple is catering directly to the builders, giving them shiny shovels at a nice premium. Reports indicate AI labs, research institutes, and independent engineers are buying these new Mac Studios by the dozen (did we mention they can be clustered?).

Outsourcing expensive and risky model development while supplying the entire developer ecosystem with the ultimate hardware sandbox turned out to be a case of brilliant positioning and foresight, rather than a failure. The strategy of supplying the tools necessary for the AI revolution is what took NVidia to being the most valuable company on the planet after all. Now Apple is poised to repeat this move but in a different market niche and context. 

Now where is the “Order now” button…