Encrypting data while it's being processed, not just before and after
Three of the most powerful names in technology are converging around a single question: what does it take to make an AI assistant truly capable, and at what cost to principle? Apple, long a guardian of user privacy and sovereign over its own silicon, is turning to Google's cloud and Nvidia's most advanced chips to rebuild Siri from the ground up — a quiet admission that even the most resourced companies must sometimes borrow the infrastructure of rivals to remain relevant. The arrangement, anchored by encryption technology that protects data even mid-computation, reflects the broader tension of this technological moment: the pursuit of intelligence at scale and the preservation of trust are not easily reconciled, but the attempt must be made.
- Apple's Siri has fallen behind competitors, and the urgency to close that gap is now driving the company into the arms of Google and Nvidia — an alliance that would have seemed unthinkable just years ago.
- Routing user voice queries through Google's cloud creates a fundamental tension with Apple's most cherished brand promise, and the company knows it — hence the insistence on Nvidia's confidential compute encryption as a non-negotiable condition.
- Nvidia's Blackwell B200 chips, the most powerful data center GPUs available, represent the computational muscle Apple could not quickly replicate on its own, making this partnership as much an admission of limitation as a strategic choice.
- Apple's own Private Cloud Compute platform now sits in an awkward position — its role unclear as some Siri workloads migrate to Google's infrastructure, leaving developers and users waiting for WWDC 2026 to understand how these two systems will coexist.
- The real verdict will arrive not in a press release but in the hands of users — when they discover whether this hybrid architecture delivers an AI assistant worth the philosophical compromises it required.
Apple is remaking Siri, and it is not doing it alone. According to reporting from The Information, the company is turning to Google's cloud infrastructure and Nvidia's flagship Blackwell B200 chips to power the next generation of its voice assistant — a meaningful departure from Apple's tradition of building on its own silicon and systems.
The arrangement is unusual in its ambition and its company. When users pose questions to Siri, some of those requests will travel through Google Cloud, processed using a licensed version of Google's Gemini AI model and the raw computational power of Nvidia's most advanced data center GPUs. Apple's reasoning appears pragmatic: delivering genuinely capable AI at scale demands resources that even Apple found more efficient to borrow than to build.
Privacy, however, remains a condition rather than a casualty. Apple has approved Nvidia's confidential compute technology, which encrypts user data not merely in transit or storage, but during active computation on the GPU itself — a technically demanding safeguard that Nvidia describes as preserving the confidentiality and integrity of AI workloads even in shared cloud environments. For a company that has made privacy a cornerstone of its identity, this protection appears to have been the price of admission.
The timing points toward Apple's Worldwide Developers Conference in 2026 as the likely stage for a public unveiling. Yet a significant question hangs over the announcement: how will this Google-routed approach coexist with Apple's own Private Cloud Compute platform, introduced as its in-house answer to privacy-preserving AI? Apple has not yet explained which requests will go where, or whether the two systems are complementary or competing.
What the partnership makes plain is that Apple has concluded it needs outside help to build the Siri that users now expect. The deeper test will come when that assistant reaches real hands — and when Apple must explain, convincingly, how a voice query routed through Google's data centers still belongs to the user who spoke it.
Apple is about to remake Siri, and it's not doing it alone. The company is turning to Google's cloud infrastructure and Nvidia's most powerful data center chips to power the next generation of its voice assistant, according to reporting from The Information. The move signals a significant shift in how Apple approaches artificial intelligence—one that leans on external partners rather than relying solely on its own silicon and systems.
The partnership brings together three technology giants in an unusual arrangement. Apple will use Google's fleet of Nvidia Blackwell B200 chips, Nvidia's flagship data center GPU built for training and running large language models. When users ask Siri questions, some of those requests will be routed through Google Cloud and processed using a licensed version of Google's Gemini AI model. It's a pragmatic solution to a hard problem: building genuinely capable AI features at scale requires enormous computational resources, and Apple apparently decided that tapping into Google's existing infrastructure made more sense than building it from scratch.
The arrangement comes with a privacy safeguard that Apple clearly considers essential. The company has approved Nvidia's confidential compute technology, which encrypts user data while it's actively being processed on the GPU. This is not encryption in transit or at rest—it's encryption during computation itself, a more demanding technical feat. Nvidia describes the technology as preserving "the confidentiality and integrity of AI models" and allowing sensitive workloads to run securely even in cloud environments. For Apple, which has long positioned privacy as a core brand promise, this layer of protection appears to be a non-negotiable condition of outsourcing Siri processing to Google's cloud.
The timing matters. Apple is expected to unveil the upgraded Siri at its Worldwide Developers Conference in 2026, giving the company a public stage to explain how it's balancing capability with privacy in its AI strategy. But questions linger about how this cloud-based approach fits with Apple's own Private Cloud Compute infrastructure, which the company introduced as part of its Apple Intelligence platform. Private Cloud Compute was supposed to be Apple's answer to privacy-preserving cloud AI—a way to do sophisticated processing without surrendering user data to third parties. Now Apple appears to be running some Siri workloads through Google's cloud instead. The company hasn't yet clarified how these two systems will coexist, or which types of requests will go where.
What's clear is that Apple has concluded it needs Nvidia's hardware and Google's scale to deliver the kind of Siri that users increasingly expect. The partnership is a tacit acknowledgment that even Apple's resources have limits when it comes to building world-class generative AI. The real test will come when users see what this hybrid infrastructure actually delivers—and whether Apple can maintain its privacy narrative while routing voice queries through Google's data centers.
Notable Quotes
Apple will enable Nvidia's confidential compute feature that encrypts data as it's being processed on the chips— The Information reporting
The technology preserves the confidentiality and integrity of AI models and allows sensitive workloads to run securely, even in cloud environments— Nvidia on confidential compute
The Hearth Conversation Another angle on the story
Why would Apple, a company famous for controlling its entire stack, suddenly outsource Siri to Google's cloud?
Because the math doesn't work otherwise. Building and running the kind of AI models that power modern voice assistants requires staggering amounts of compute. Apple decided it was smarter to pay Google for access to infrastructure that's already built and optimized than to duplicate that effort.
But doesn't that contradict Apple's privacy story? Sending voice data to Google Cloud seems like the opposite of what Apple usually does.
It does create tension. That's why the confidential compute piece matters so much. Apple is betting that encrypting data during processing—not just before and after—is enough to satisfy both users and regulators that the data isn't actually exposed to Google.
Is that technically true? Can Google really not see the data if it's encrypted while being processed?
In theory, yes. Confidential compute is designed so that even the cloud provider's administrators can't access the plaintext. But it's still a trust model. You're relying on Nvidia's implementation and Google's infrastructure to work as advertised.
What about Apple's own Private Cloud Compute? Wasn't that supposed to solve this problem?
It was. But Private Cloud Compute apparently isn't powerful enough or flexible enough to handle all the workloads Apple wants to run. So now Apple has two systems—its own and Google's. That's messy, but it's honest about the tradeoffs.
When will we actually see this in action?
WWDC 2026 is when Apple will show it off. That's when the company typically unveils major platform changes. But the real test is what users experience when they upgrade their phones and start using the new Siri.