Perplexity Partners with Intel: Local AI Models and Hybrid Inference Come to Laptops

Perplexity Partners with Intel

Perplexity AI CEO Aravind Srinivas recently thanked Intel CEO Lip-Bu Tan and the Intel team on social media, announcing that the two companies are collaborating to bring local AI models and hybrid inference capabilities to Intel Core Ultra Series 3 laptops. This partnership marks Perplexity's official entry into the on-device AI space for personal computers.

Perplexity AI was founded in 2022 by Aravind Srinivas, a former Google and OpenAI researcher. The company positions itself as an "answer engine" rather than a traditional search engine. Unlike Google Search, which returns a list of links, Perplexity generates structured answers with cited sources. As of 2024, Perplexity's valuation has surpassed $9 billion with over 15 million monthly active users, making it one of the most credible challengers to Google Search. Its business model includes a free basic tier and a $20/month Pro subscription that provides access to multiple large models including Claude and GPT-4.

What Is Hybrid Inference? Perplexity's Edge-Cloud Approach

Based on Perplexity's messaging, the core concept behind this collaboration is "Personal Computer with local models and hybrid inference" — running local models on the PC while leveraging cloud capabilities for combined reasoning.

In practice, when users interact with Perplexity, some AI inference tasks will be handled directly on the laptop's NPU (Neural Processing Unit), while more complex tasks are offloaded to the cloud. An NPU is an accelerator chip specifically designed for machine learning inference. Compared to general-purpose CPUs, NPUs feature hardware-level optimizations for matrix operations and tensor computations, enabling AI inference at significantly lower power consumption. The key advantage is energy efficiency — NPUs consume only a fraction of the power a GPU would need for the same AI inference workload, which is critical for battery-powered laptops.

The core idea behind Hybrid Inference is dynamically allocating compute resources based on task complexity. In a typical architecture, a lightweight "router model" first assesses the complexity of a user's request: simple text completion, formatting, or basic Q&A is handled directly by a small model running on the local NPU, while queries involving multi-step reasoning, cross-document synthesis, or real-time web retrieval are forwarded to cloud-based large models. This architecture must address several key technical challenges: context synchronization between local and cloud models, maintaining a consistent user experience during seamless handoffs, and ensuring accuracy of the intelligent routing strategy.

This hybrid inference architecture delivers several notable advantages:

• Lower latency: Simple queries get instant responses without network round-trips
• Privacy protection: Sensitive data can be processed locally without uploading to the cloud
• Offline availability: Basic AI functionality remains accessible even without an internet connection
• Cost reduction: Less cloud compute consumption benefits both users and Perplexity

Intel Core Ultra Series 3: Powering On-Device AI

Intel Core Ultra Series 3 (codenamed Lunar Lake) is Intel's processor platform built for the AI PC era, featuring an integrated NPU capable of 48 TOPS (Trillion Operations Per Second). Microsoft has defined 40 TOPS as the minimum threshold for Copilot+ PCs, meaning Core Ultra Series 3 already meets and exceeds the industry's hardware requirements for AI PCs. This provides a solid hardware foundation for running small-to-medium language models on laptop devices.

Deploying large language models to edge devices requires a suite of model compression techniques. Mainstream methods include Quantization — reducing model weight precision from FP32 to INT8 or even INT4, which can shrink model size by 4-8x with limited performance degradation; Knowledge Distillation — using a large model to guide the training of a smaller one; and Pruning — removing network connections that have minimal impact on output. Currently, edge devices can typically run models in the 1B-7B parameter range, such as Microsoft Phi-3 and Google Gemma 2B, which are specifically optimized for on-device deployment and can achieve near-large-model performance on specific tasks.

Intel has been aggressively promoting the "AI PC" concept, but the market has seen few truly compelling on-device AI applications. Intel formally introduced the AI PC concept in late 2023, defining it as a personal computer integrating three compute engines: CPU, GPU, and NPU. This strategy emerged after Intel fell significantly behind NVIDIA in the data center AI chip market, pivoting to focus on the PC edge AI market for differentiation. Intel CEO Lip-Bu Tan officially took the helm in March 2025 — previously the long-time CEO of Cadence Design Systems, he's known for his pragmatic engineering management style. Since taking over, he has accelerated AI PC ecosystem development and actively pursued partnerships with leading AI application companies.

As one of the fastest-growing AI search products, Perplexity's involvement is undoubtedly a major boost for Intel's AI PC ecosystem. For Intel, this is not just a product-level collaboration but a critical case study proving the real-world value of its NPU.

Industry Trend: AI Moving from Cloud to Edge

This partnership reflects a significant industry trend — the shift from pure cloud inference to hybrid edge-cloud inference.

Over the past two years, virtually all mainstream AI applications have relied on cloud-based large models for inference tasks. But as edge chip capabilities continue to improve and model compression techniques advance, more AI capabilities are being pushed "down" to end-user devices. Apple Intelligence, Copilot+ features on Qualcomm Snapdragon X Elite, and now the Perplexity-Intel collaboration all validate the viability of on-device AI.

For Perplexity, this also represents a strategic platform expansion. Moving beyond a purely web and mobile AI search tool to deep PC integration means Perplexity is positioning itself to become an indispensable part of users' daily computing experience — not just another browser tab.

Questions Worth Watching

The specific implementation details of this partnership haven't been fully disclosed yet. Several questions are worth tracking:

1. Local model capability boundaries: Which functions can be handled entirely on-device? What are the model size and precision specifications?
2. User experience differences: Will response quality differ noticeably in hybrid inference mode compared to the pure cloud version?
3. Intel exclusivity: Will this feature eventually expand to other chip platforms like AMD and Qualcomm?
4. Pre-installed or user-installed: Will Perplexity come pre-installed on laptops with Intel Core Ultra Series 3, or will users need to download and install it themselves?

Regardless of these open questions, the Perplexity-Intel partnership provides a concrete and compelling use case for AI PCs, bringing the concept of "on-device AI" one step closer to everyday users.