The infrastructure supporting Apple's artificial intelligence is undergoing a striking shift. What initially appeared to be a closed ecosystem is now opening up to integrate cutting-edge technology from Nvidia and Google with a very clear objective: to ensure that the processing of our data in the cloud is as secure as if it had never left the phone. This new phase of Private Cloud Compute seeks to resolve the dilemma of how to offer powerful AI features without compromising user privacy, especially when tasks exceed the computing capacity of an iPhone or Mac.
In this scenario, the Cupertino company has decided it can't do it all alone and has turned to Nvidia's powerful Blackwell GPUs . These processors not only ensure fast responses from Siri and smooth complex photo editing, but also implement what's known as confidential computing. Essentially, this means that information is isolated while being processed, preventing even system administrators from snooping on what we're accessing. It's a technical move intended to maintain Apple's reputation as the guardian of our digital privacy.
The power of Nvidia Blackwell at the service of privacy
The arrival of Blackwell B200 chips in Google Cloud's data centers, which Apple uses to bolster its infrastructure, marks a significant milestone. This technology enables the establishment of encrypted channels and remote verification , ensuring that the hardware has not been tampered with before sending any sensitive data to the server. For the average user, this means that the most demanding Apple Intelligence features—those that the phone cannot handle on its own—are sent to the cloud with a level of protection that until recently seemed like science fiction.
It's curious that Apple has opted for external infrastructure after years of championing its complete control over silicon. However, tests suggest that its own Private Cloud Compute systems needed a boost to run large language models with the fluidity demanded by today's market. By delegating part of this workload to Google and Nvidia, Apple gains the necessary power without compromising its privacy rhetoric, as the system's design prevents data from being stored or used to train other external models.
The 12GB RAM benchmark and hybrid processing
One detail that has caused quite a stir among users is the hardware barrier required to access these features. For the system to function correctly, a minimum of 12 gigabytes of RAM has been established , which excludes iPhone models that we still consider modern. The technical explanation is simple: for hybrid inference to be effective, the device must keep the local model in its working memory, and if there isn't enough space, the experience simply breaks down. This is where Private Cloud Compute takes over, handling what the local chip can't manage.
This division of tasks is what experts call hybrid AI inference. A smart router decides in milliseconds whether your request can be handled locally or if it needs to be sent to the servers. Simpler, everyday tasks remain on the local hardware , saving the company money and providing an instant response to the user. Conversely, if you request something that requires a much deeper understanding of the context, the heavy artillery of the cloud comes into play, always under the layer of security promised by new agreements with its technology partners.
The European Union's wall and the future of Siri
If we focus on what directly affects us in Spain and the rest of Europe, the situation is a bit more uncertain. Due to the requirements of the Digital Markets Act (DMA) , Apple has put the brakes on the rollout of these artificial intelligence features. The company fears that opening its system to third parties, as required by European regulations, could compromise the security structure they have built with Private Cloud Compute. For now, this leaves us in a holding pattern while other markets are already beginning to experience the benefits of this new, more capable and contextual Siri.
For those already making plans, it's important to note that access to these advanced cloud capabilities could end up being tied to certain subscription levels. While the basic features will be free, the intensive processing on high-performance servers has a real cost that Apple could try to pass on through its iCloud plans. It's a way to monetize a technology that, let's face it, costs a fortune to keep running 24/7.
The architecture being developed behind the scenes demonstrates that the future of our devices depends not only on their internal components, but also on an invisible network of ultra-secure servers working tirelessly. While in Europe we still have to wait for regulators and the company to reach an agreement, the move to combine Nvidia's talent and Google's infrastructure under Apple's umbrella makes it clear that the race for artificial intelligence is won through strategic alliances. Ultimately, what matters is that this revamped Siri understands our questions without us losing control over our personal information.