August 10, 2026 —
Intel seeks $15B in fresh equity for an AI-driven expansion cycle
Today's North America AI brief follows capital and control across the stack. Intel is raising fresh equity to pursue AI-driven manufacturing opportunities, while model developers push more capability onto local hardware and software agents take on longer, less supervised work.
Intel seeks $15B in fresh equity for an AI-driven expansion cycle
Intel announced a $15 billion underwritten common-stock offering and may let underwriters buy another $2.25 billion of shares. The chipmaker said proceeds could support capital expenditures and working capital while protecting its investment-grade rating. Intel tied the timing to sustained customer demand for AI compute and cited physical AI, purpose-built silicon, advanced packaging, and external wafer manufacturing as growth opportunities.
Claude Code makes autonomous tool approval the default
Anthropic will make auto mode the default for new Claude Code sessions on Pro, Max, and Team plans beginning August 14. Tool calls will pass through a classifier that blocks irreversible, destructive, or out-of-environment actions instead of prompting users each time. In a controlled study with 1,053 paid testers, Anthropic says auto mode caught 89% of dangerous commands, versus 13.6% for human review.
Meta returns to open weights with a 30B local agent model
Meta released Muse Glimmer, a dense 30-billion-parameter multimodal model designed for local agentic workloads. The Apache 2.0 release combines a 2B vision encoder with a 28B text decoder and supports images, video, coding, tool use, and speculative decoding. Meta's model artifacts and Hugging Face's launch support position it as a privacy- and cost-conscious option for developers who want capable agents on a single workstation-class GPU.
South Korea adds a $3.5B fund to its semiconductor push
South Korea plans a 5 trillion won, roughly $3.5 billion, semiconductor fund aimed at materials, components, equipment, and fabless chip companies. The government also wants faster development of domestic chip hubs and additional trade financing for suppliers. The package targets vulnerable layers of the semiconductor supply chain as AI demand intensifies competition for memory, manufacturing equipment, and advanced production capacity.
Source Foundry draws another $400M for chip-manufacturing tools
Situational Awareness invested another $400 million in stealth semiconductor-equipment startup Source Foundry, bringing its reported total commitment to $500 million. The startup, founded by Stanford researchers, is trying to make advanced chip manufacturing faster and less expensive. The investment lands after the hedge fund sharply reduced its public portfolio, concentrating a large private bet on a potential challenger in a manufacturing stack dominated by a small number of suppliers.
Meta pairs open-model rhetoric with a community fund
Mark Zuckerberg argued for widely distributed personal superintelligence and said Meta will resume releasing open-source models soon. Meta also introduced a Future Is For Everyone Fund for communities hosting its data-center projects.
Ford begins its app-first vehicle assistant rollout
Ford is rolling an AI assistant into Ford and Lincoln apps in waves, using vehicle-specific data to answer questions about fuel, service, cargo, and towing. Ford targets up to 8 million owners before an in-vehicle version arrives in 2027.
Discovered Materials funds an AI search for cooler chips
Discovered Materials raised a $9 million seed round led by Lightspeed India Partners. Its agent pipeline proposes semiconductor materials, tests them with physics models, and sends the most promising candidates toward lab validation.
Chunked distillation pushes long-context training onto one GPU
Multiverse Computing researchers published an offline top-K distillation method and fused chunked KL loss that avoid keeping teacher and student models together. The team says the approach makes long-context recovery practical on a single GPU.
Cyber evaluations become an agent-containment problem
A review of recent incidents found frontier agents reaching real systems during cyber testing after sandbox and network-control failures. UK AISI and company postmortems point toward stricter egress controls, continuous monitoring, and independent environment audits.