Snowflake’s Cortex Training moves fine-tuning off the cluster, Perplexity’s pplx-embed-v2-context turns whole documents into searchable vectors, and NTT DATA adds 12MW of AI-ready capacity in Jakarta. Meanwhile, chips, optics, software, and infrastructure partnerships accelerate the race for practical AI deployment. Scroll down to catch the signals that matter.

🧠 Snowflake Moves Model Training Off-Cluster

Snowflake opened public preview of Cortex Training, a serverless platform for fine-tuning and deploying open-weight models. It supports LoRA, full-parameter training, supervised fine-tuning, reinforcement learning, DPO, and distillation while managing GPU operations, experiments, evaluation, and production deployment within Snowflake’s governed environment.

Serverless training could make customized models faster and cheaper to operationalize.

💼 Ondine Expands Its AI Investment Mandate

Ondine Capital announced its third fund, targeting the full AI stack—from foundation models and semiconductors to infrastructure, enterprise software, and physical AI. The broad mandate reflects how investment opportunities increasingly span compute, data, deployment systems, robotics, and the industrial technologies connecting them.

Capital is following AI across the entire technology supply chain.

🧩 Perplexity Upgrades Multimodal Retrieval

Perplexity introduced pplx-embed-v2-context, open-weight vision-language embedding models available in 9-billion and 0.6-billion parameter versions. They can generate int8 vectors for complete documents or context-aware chunks, helping retrieval systems preserve relationships among text, images, layout, and broader document meaning.

Context-aware embeddings could improve search quality without forcing systems to process every document repeatedly.

🏗️ VMware Builds Governed AI Data Foundations

VMware announced general availability of AI-ready data foundations in Tanzu Platform. The offering is designed to help enterprises create governed, context-rich data products for AI applications and agents, addressing the reliability, lineage, access-control, and operational challenges that often undermine otherwise capable models.

Enterprise AI increasingly depends on trustworthy data products, not models alone.

🛡️ Sundae Labs Acquires Aigensei

Sundae Labs acquired Aigensei, bringing its truth-telling engine and AI development team into the AGNi platform. The deal aims to strengthen systems designed to reduce hallucinations and improve factual reliability, combining specialized verification technology with a broader platform for building and deploying AI applications.

AI reliability is becoming a product layer—and an acquisition priority.

🔌 Tower and Innolight Advance Near-Package Optics

Tower Semiconductor and Innolight continued their near-package optics collaboration, targeting higher-performance connectivity for next-generation AI infrastructure. Bringing optical components closer to processors can help address bandwidth, power, and signal-integrity constraints as accelerated computing systems grow denser and data movement becomes a dominant bottleneck.

The next AI performance gains may come from networking architecture as much as silicon.

⚡ Onsemi Prioritizes AI Data Centers

Onsemi is prioritizing AI data-center customers after artificial-intelligence revenue surged 175% year over year. The company’s focus highlights demand for power-management and semiconductor components that help servers handle higher compute loads, tighter efficiency requirements, and the thermal challenges created by increasingly concentrated AI workloads.

AI demand is lifting the broader infrastructure component market—not just accelerator sales.

🚀 AMD Rides Accelerating Data-Center Demand

AMD forecast quarterly revenue above estimates, supported by accelerating data-center demand for its AI chips and graphics processors. The outlook reinforces AMD’s expanding role in the infrastructure race, where customers are seeking additional accelerator supply, competitive performance, and alternatives for scaling large-model training and inference.

Data-center momentum is turning AI hardware competition into a broader market contest.

🌏 NTT DATA Opens 12MW Jakarta Facility

NTT DATA opened its Jakarta 2 Annex Data Center, an AI-ready facility delivering 12MW of IT capacity. The site combines advanced cooling, resilient power, and carrier-neutral connectivity, while new low-latency optical links between Indonesian facilities support high-density workloads and faster movement of data across the company’s regional footprint.

Southeast Asia is becoming a more consequential location for AI-ready capacity.

🌐 North Yorkshire Targets Gigawatt-Scale AI Capacity

Drax, Brookfield, Radiant, and Qai announced a £20 billion partnership to develop up to gigawatt-scale AI data-center capacity in North Yorkshire. The proposal links major infrastructure investment with power generation, site development, and advanced computing, underscoring the enormous energy and capital requirements of future AI campuses.

AI infrastructure planning is expanding from individual facilities to regional power-and-compute ecosystems.

What trends are you tracking? Reply with your take, or send this to a colleague shaping AI strategy.

Share your unique referral link

Sponsor this newsletter in 2027: 8 placements on dates you choose, from EUR 2,400, until 30 November. Book: 7wdata.be/sponsor-2027