Apple’s Siri overhaul leans on Google’s Gemini and Nvidia’s Blackwell chips, report says

Apple’s long-awaited effort to improve Siri, the voice assistant it began reworking in 2024, is increasingly taking shape as a partnership-driven fix rather than a fully self-contained rebuild, according to a report published Friday. TheStreet said Apple’s next Siri will rely on two outside technology providers, Google for the underlying large-language-model “brains” and Nvidia for the data-center chips that run some of the heavier cloud work. With WWDC starting June 8, the coming weeks will be the first chance for developers and early users to see whether the upgraded assistant can finally match Apple’s promises.

The reported approach splits Siri into two layers. Everyday, simpler requests would stay on the iPhone and be handled by smaller models Apple trained itself, while harder questions that require deeper reasoning would be routed to the cloud and answered using a custom version of Google’s Gemini model, built by Alphabet’s AI team. In other words, Siri would behave like it has enough local capability for quick tasks, but it would call out to cloud intelligence when the query is complex. Apple would still control the front-end experience, but the “thinking” for top-tier prompts would depend on external model performance.

TheStreet also reported that Apple agreed to pay Google about $1 billion a year for access to the Gemini model used in Siri’s cloud layer. It added that Apple could not get the full licensed model to run on its own on-premises “Private Cloud Compute” hardware environment, pushing at least some requests into Google Cloud instead. On the compute side, the report said the cloud queries run on Nvidia’s Blackwell B200 data-center chips, with user data protected using confidential-compute style encryption based on Nvidia hardware features. The same hardware tier that powers parts of the broader AI ecosystem would therefore be used behind Siri’s most demanding interactions.

Why does this matter beyond architecture? Because Siri’s reputation has become a business and brand problem. TheStreet said Apple showed a smarter, more personalized Siri at a developer conference in 2024 and promoted it in marketing tied to the iPhone 16 launch, but the feature release slipped and the ads were pulled back. It also cited a $250 million class-action settlement tied to false-advertising claims over delayed Siri capabilities. That combination of missed expectations and public embarrassment is part of the context behind Apple’s urgency to stabilize and modernize the assistant.

The reported move also touches a tension Apple has leaned on for years: privacy. In 2024, the report says Craig Federighi told developers that anything leaving the device should be processed on Apple’s own servers for privacy reasons. With Siri’s cloud layer now described as dependent on Google Cloud, some requests would travel to Google’s infrastructure, even if they are encrypted and handled within confidential-compute workflows. Tim Cook, according to the report, has tried to position the change as routine as Apple prepares to deliver a more personalized Siri to users later in 2026.

Apple’s own messaging ahead of WWDC provides the calendar pressure for the rollout. In a May 18 Newsroom update, Apple said WWDC26 takes place June 8-12 and “kicks off” June 8 with the Keynote and Platforms State of the Union, featuring “AI advancements.” In a March 23 Newsroom press release, Apple reiterated that the conference begins June 8 and is followed by online sessions throughout the week, with developers able to watch the announcements and join labs. TheStreet’s reporting suggests Siri’s early, developer-facing version could show up in beta soon after the June 8 start.

Still, there are key uncertainties. The most specific Siri details, including pricing, exact partner responsibilities, and the final privacy implementation, appear to come from third-party reporting rather than an Apple primary disclosure in the materials reviewed. Apple has not publicly confirmed the exact Gemini configuration, the specific Nvidia chip model, or the precise “what leaves the device” breakdown in a way that can be directly verified from an official source. What to watch next is whether the June release window produces an assistant that behaves consistently in everyday tasks, and whether Apple’s WWDC demos align with the cloud-and-chip dependency described in the report. If it does, the partnerships could quickly become invisible to users. If not, the risk is that the credibility gap Siri has already created could widen again before it is fully addressed.