Apple's Worldwide Developers Conference is scheduled to open Monday, June 8, at 1:00 p.m. ET, and the most consequential detail of its artificial intelligence overhaul concerns hardware Apple does not own. According to a report from The Information dated June 3, the rebuilt, Gemini-powered Siri will send its most demanding requests to Google Cloud, where they will be processed on Nvidia Blackwell B200 graphics processors rather than on Apple's own servers. For an installed base of roughly two billion Apple devices, that means a core piece of the iPhone's intelligence will soon run on infrastructure built by two of Apple's biggest competitors.

Apple Will Route Complex Siri Queries to Google's Nvidia B200 GPUs

Apple is expected to keep simple requests, such as setting timers or controlling smart-home devices, running locally on the iPhone. The shift applies to the harder cases. Queries that exceed on-device limits will fall back on one of Google's large Gemini models, and that processing will happen on Google's fleet of Nvidia Blackwell B200 chips, with user data encrypted on the silicon itself. The arrangement places Google and Nvidia at the center of the most important product Apple has shipped in years.

It also breaks with Apple's long-standing practice of controlling the full stack. The company has spent years arguing that secure software must rest on hardware it designs, and it built Private Cloud Compute on its own silicon for precisely that reason. Outsourcing inference for its flagship assistant to a rival's data center is a direct departure from that philosophy.

Private Cloud Compute Proved Too Slow for the 1.2-Trillion-Parameter Model

The reason for the change appears to be performance. Apple tried running a Gemini model inside Private Cloud Compute, but the system was too slow during testing to be usable at the scale Siri requires, according to the reporting. That outcome pushed Apple toward Google's existing infrastructure, which is already provisioned for trillion-parameter inference.

The detail matters because it revises the earlier understanding of how the new Siri would work. As recently as last November, the expectation was that the custom Gemini model would run exclusively on Apple's own servers. The June report indicates that plan changed, and it leaves open how Private Cloud Compute fits into the launch at all.

How Does Nvidia B200 Confidential Computing Protect Siri Data?

The privacy mechanism is the engineering core of the story. The Blackwell B200 is a multi-die data-center processor packing 208 billion transistors, built around a second-generation Transformer Engine that supports four-bit floating-point math for large-model inference and a fifth-generation NVLink interconnect that moves data between chips at 1.8 terabytes per second. It is designed specifically to serve trillion-parameter models, which is why a 1.2-trillion-parameter Gemini system maps onto it cleanly.

The feature Apple is relying on is Nvidia confidential computing, a hardware-based trusted execution environment that Nvidia extended from the CPU to the GPU. When the mode is active, the Blackwell chip encrypts all data held in GPU memory, including model weights, the user's input, and the inference result, while the computation is running. In multi-GPU configurations, the NVLink traffic between chips is encrypted as well, and the B200 is the first GPU to carry trusted-execution protection across its input and output paths.

For a reader, the practical meaning is specific. Apple's bet is that on-chip encryption keeps a Siri request confidential even while it is processed on a competitor's machine, so the data is never exposed in plaintext to Google's broader cloud. This is a fundamentally different design from Apple's original Private Cloud Compute, which promised hardened, stateless servers that retained nothing; the new approach instead keeps the prompt encrypted in hardware at every step, paired with a CPU trusted-execution environment and cryptographic attestation that verifies the chip before any data is sent.

The $1 Billion Gemini Deal Behind Siri's New Brain

The Nvidia hardware sits on top of a partnership Apple and Google announced January 12, 2026. Apple agreed to pay roughly $1 billion a year to license a custom 1.2-trillion-parameter Gemini model, a system about eight times larger than Apple's own 150-billion-parameter cloud model and far beyond the roughly three-billion-parameter model that runs on the device.

The economics invert a familiar relationship. Google already pays Apple an estimated $20 billion a year to remain the default search engine in Safari. Under the Gemini arrangement, the money moves the other direction, with Apple paying Google for the intelligence layer behind its own assistant.

DOJ Appeal Puts the Apple-Google Partnership Under Antitrust Scrutiny

The deepening relationship lands in a contested legal environment. The Department of Justice filed a DOJ antitrust appealin February 2026 challenging the September 2025 ruling that left the $20 billion search-default deal intact. Legal analysts have argued that the Gemini-Siri agreement raises the same structural concerns the government identified in the search case, because it routes the complex AI queries of two billion devices through a single dominant provider. Google, for its part, has said it will not receive Apple user data through the arrangement, and the contract reportedly bars Google from training on Siri queries.

That tension is the genuine conflict in the story. Apple is solving an immediate capability gap by leaning on the one rival regulators are actively trying to disentangle it from, which means the partnership could face conditions neither company has publicly addressed if the appeal succeeds.

New Siri Arrives in September, Not at Monday's Keynote

Anyone expecting the finished assistant on Monday will likely wait longer. WWDC is where Apple is expected to preview iOS 27 and its companion operating systems, with developer betas to follow. The fully conversational, Gemini-powered Siri is reported to launch in September alongside the next iPhone, after partial features shipped in earlier iOS 26 point releases. Monday is the formal unveiling and the developer on-ramp, not the consumer release date.

Frequently Asked Questions

Is Siri using Google Gemini?

Yes. Under a partnership announced in January 2026, the rebuilt Siri is based on a custom Gemini model that Apple licenses from Google. Apple's own smaller models still handle simple, on-device requests.

When is the new Siri coming out?

The Gemini-powered Siri is reported to launch in September 2026 alongside the next iPhone, not at the June 8 WWDC keynote. WWDC is expected to preview the software and open it to developers first.

Does Apple share Siri data with Google?

Google has said it will not receive Apple user data through the deal, and Apple is relying on Nvidia confidential computing to encrypt requests while they are processed. The contract reportedly prevents Google from training its models on Siri queries.

What is the Nvidia B200?

The Blackwell B200 is Nvidia's flagship data-center GPU, built for large-scale AI training and trillion-parameter inference. It includes hardware confidential computing that encrypts data while the chip is processing it.

For complex requests, the most personal interface on two billion Apple devices will soon run on Google's cloud and Nvidia's chips, defended by on-chip encryption rather than Apple's own servers, with the full assistant due in September and an antitrust appeal hanging over the partnership that makes it possible. That combination, not any single feature, is what a reader should weigh when the keynote begins Monday.