OpenAI Accelerates Custom Silicon Initiative with Broadcom Collaboration

OpenAI and Broadcom announce co-development of OpenAI-designed AI accelerators and scale-out Ethernet hardware.
Deployment planned to start in phases from the second half of 2026, targeting 10 GW of compute capacity for OpenAI’s data infrastructure.
Broadcom integrates OpenAI’s model R&D learnings into specialized chip architectures and rack systems.
Move enables OpenAI to diversify compute sourcing beyond established GPU vendors such as NVIDIA and AMD.
Broadcom’s share price rallies following publication of details on partnership scale and timeline.
Leadership from both companies position the deal as key to unlocking next-gen model capabilities and performance.

NVIDIA’s Spectrum-X Ethernet Switches Power Meta and Oracle’s Next-Gen AI Data Centers

Meta and Oracle select NVIDIA’s Spectrum-X platform to build out giga-scale AI networking infrastructure in 2025 and beyond.
Ethernet backbone engineered for ultra-low latency, scalable bandwidth, and advanced congestion management.
Spectrum-X targeted to serve hyperscale training and inference workloads in multi-tenant environments.
NVIDIA positions the platform for use in diversified cloud AI architectures, not just proprietary fabrics.
Adoption follows recent cloud trends toward open, flexible networking stacks for large-scale model deployment.
Technical briefings highlight predictable performance and compatibility across complex topologies.

SB 243 law passed to govern AI chatbot products, including disclosure, verification, and crisis protocol requirements.
New rules demand platforms implement age checks and break reminders for minors using AI chat features.
Crisis-response systems become mandatory for self-harm scenarios, with state oversight and reporting obligations.
Presenting chatbot products as healthcare professionals is now explicitly prohibited.
Penalties for non-compliance extend up to $250,000 for severe violations.
Law takes effect January 2026, marking the first comprehensive state-level regulatory standard for consumer AI safety.

Power Integrations and NVIDIA collaborate to enable higher-voltage power delivery for modern AI data centers.
Higher-voltage architectures reduce energy loss and simplify distribution across large compute clusters.
Industry experts point to direct impact on operational costs and long-term scalability.
Technical innovation focuses on minimizing conversion stages and maximizing efficiency end-to-end.
Rising AI workload densities make power system advances critical for campus expansion.
Strategic partnership seen as part of wider movement to optimize power, cooling, and grid integration for AI growth.

U.S. data center investment reported to have risen over one third in the first half of 2025, reflecting a structural shift driven by AI infrastructure needs.
Factory construction slows by comparison, showing change in national capital expenditure trends.
AI campuses increasingly anchor local economies, job markets, and utility upgrades at regional level.
Land, energy, and cooling considerations dominate decision-making for new hyperscale locations.
AI demand for compute and storage is reshaping investment priorities among major operators.
Market and policy experts expect infrastructure build-outs to further accelerate in coming fiscal quarters.