phoebe leu

Nvidia is increasing prices on Grace Blackwell 300 and Vera Rubin 200 server racks by ~17% for early-to-mid 2025 shipments, driven primarily by memory chip cost increases from Samsung, SK Hynix, and Micron, with storage and networking also contributing; most increases expected to be passed to end customers.

Perplexity's annualized revenue has reached ~$750M, a significant uptick from months ago, driven by AI search and agent products (Computer Use); the company doesn't train foundation models but routes tasks to various models including post-trained open models, and collaborates closely with Nvidia on hardware design and real-world usage data.

Nvidia completed a $6B deal for Poolside (initially reported as license-and-hire), marking a shift toward traditional equity investments in AI model companies alongside Groq and Infabrica deals, as part of its strategy to expand the open-model ecosystem that drives GPU demand.

Nvidia is in talks to invest multiple billions in Perplexity at a $30B valuation (50% above last round), shifting from a license-and-hire model to equity investment; this follows similar deals for Groq, Infabrica, and Poolside ($6B), as Nvidia builds an open-source model ecosystem (Nemotron family) to drive broader GPU demand beyond top-tier buyers.

Nvidia completed a $6B license-and-hire transaction for Poolside, an AI coding agent company, as part of its strategy to acquire model talent and technology; the deal structure preceded the potential Perplexity equity investment.

Nvidia is in talks to invest multiple billions in Perplexity at a $30B valuation (50% above last round), shifting from a license-and-hire model used for Groq, Infabrica, and Poolside to a traditional equity stake, signaling deeper alignment with AI application-layer companies that drive compute demand.

Perplexity's annualized revenue has reached ~$750M, up significantly in recent months, driven by AI search and agent products like Perplexity Computer Use; the company doesn't train base models but routes tasks to various models and collaborates closely with Nvidia on hardware design and real-world usage data.

Nvidia releases open-weight models (Nemotron family) with training data and methodology for free to enable more companies to access affordable, fine-tunable models, driving broader compute demand and GPU sales; the strategy is explicitly business-minded, not charity, with internal models quantifying the compute demand uplift.

Nvidia completed a $6B license-and-hire acquisition of Poolside (coding model startup), absorbing its team to accelerate Nvidia's own open model development; Poolside talent expected to enhance Nvidia's next 1-trillion-parameter Nemotron model.

Nvidia is in talks to invest multiple billions in Perplexity at a $30B valuation (50% above last round), shifting from a license-and-hire model to equity investment; Perplexity provides real-world AI usage data that informs Nvidia's hardware and model development, with companies meeting multiple times weekly.