Perplexity – Weekly Recap

Perplexity advanced its enterprise strategy this week, emphasizing security, infrastructure efficiency, and data-centric integrations for its Computer product. The company introduced a direct connection to Snowflake, enabling users to query live warehouse data with SQL, filters, and metrics for analytics, dashboards, and workflow automation.

The Snowflake integration targets use cases such as pipeline analysis, product usage tracking, and customer segmentation, while allowing administrators to retain control over access rights, business definitions, and shared logic. This positions Computer as a self-service analytics layer atop existing data warehouses, aiming to deepen relevance in data-heavy enterprise environments.

Perplexity also underscored a “secure by default” architecture for Computer, featuring hardware-isolated sandboxes for each task and virtual private cloud separation of storage and compute. Additional controls include short-lived proxy tokens instead of raw API keys for agent authentication and ML-based scanning of external content through classifiers and a BrowseSafe model.

File connector data is encrypted in transit and at rest, with uploaded files automatically deleted after seven days, addressing common governance and compliance concerns. These measures are designed to appeal to enterprises and regulated industries where risk management, data isolation, and content screening are prerequisites for large-scale AI deployment.

On the infrastructure front, Perplexity highlighted research on serving post-trained Qwen3 235B models using NVIDIA GB200 NVL72 Blackwell racks. Benchmarks indicate significantly lower NVLS all-reduce latency and reduced mixture-of-experts prefill combine times versus the prior Hopper generation, improving high-throughput inference performance.

The company reports that Blackwell’s rack-scale NVLink domain, enhanced memory bandwidth, and Tensor Cores boost both prefill and decode workloads, enabling faster responses at lower serving costs. If realized in production, these efficiency gains could strengthen Perplexity’s unit economics and competitiveness in AI search and assistant markets.

Collectively, the week’s developments suggest Perplexity is investing heavily in secure, scalable infrastructure and tight data-warehouse integration to win enterprise and institutional workloads. The combination of governance-focused architecture, performance optimization, and Snowflake-centric analytics capabilities points to a strategic push upmarket and a more defensible position in the AI tools landscape.

Overall, Perplexity’s week was marked by coherent progress on product robustness and enterprise readiness, which may support deeper adoption among data-centric and security-sensitive customers over time.