December 14th

Disney committed a $1 billion equity investment in OpenAI alongside a three-year licensing deal bringing 200+ characters from Marvel, Star Wars, and Pixar into the Sora video generator. This partnership allows OpenAI to officially license Disney IP while Disney integrates ChatGPT APIs across its business, marking a dramatic shift from potential adversary to strategic partner for the entertainment giant.

Unconventional AI, led by former Databricks AI head Naveen Rao, secured a $475 million seed round at a $4.5 billion valuation to build energy-efficient AI hardware. The funding, led by Andreessen Horowitz and Lightspeed, underscores investor appetite for chip architectures that can challenge NVIDIA's dominance.

Oracle stock dropped 11% after announcing a $15 billion increase in capital expenditure for fiscal 2025, bringing total forecast to $50 billion—driven largely by data center contracts for OpenAI and highlighting the immense capital required to sustain AI infrastructure.

OpenAI data revealed "frontier" workers (top 5% of adopters) send 6x more messages and 17x more coding queries than median users, suggesting a widening productivity gap where heavy users fundamentally transform their roles while the majority remain casual users.

MIT's Project NANDA reported that despite $40 billion in corporate spending, 95% of AI pilots fail to deliver tangible ROI due to learning gaps and infrastructure latency—highlighting the difficulty of moving from individual productivity gains to organizational transformation.

Microsoft analysis of 37.5 million Copilot conversations revealed distinct usage patterns where programming queries peak on weekdays while gaming and philosophical questions surge on weekends, indicating users increasingly treat AI assistants as both productivity tools and companions depending on context.

Anthropic launched beta integration of Claude Code directly into Slack, allowing developers to trigger coding agents from chat threads to fix bugs or generate features. This leverages Slack's 750,000+ organizations to embed AI coding agents directly into existing engineering workflows.

OpenAI, Anthropic, Google, and Microsoft co-founded the Agentic AI Foundation under the Linux Foundation to standardize agent interoperability. Anthropic donated the Model Context Protocol (MCP) to the foundation, establishing a neutral standard for how AI agents connect to data and tools across platforms.

Accenture and Anthropic partnered to train 30,000 Accenture professionals on Claude and launch a business group focused on regulated industries, aiming to move enterprises from pilot phases to production by restructuring software development lifecycles.

Instacart deployed an embedded checkout experience within ChatGPT using the Agentic Commerce Protocol, grounding AI in real-time inventory from 100,000 stores to reduce hallucination risks in commercial transactions.

OpenAI released GPT-5.2 featuring three tiers (Instant, Thinking, Pro), with the "Thinking" model beating or tying human professionals on 70.9% of work tasks in their GDPval benchmark. The release includes 38% reduced hallucination rates compared to GPT-5.1 following internal competition concerns about Google.

Mistral AI launched Devstral 2, a 123-billion parameter open-weights coding model achieving 72.2% on the SWE-bench Verified benchmark. Accompanied by "Mistral Vibe" command-line interface for autonomous engineering, this positions open-weight models as serious competitors to proprietary systems.

Google completed a silent rollout of Gemini 3 to over 650 million monthly active users across Android and Workspace without a major public event, testing system loads at scale while integrating the model deeply into search and mobile ecosystems as a background process.

Nous Research released Nomos 1, an open-source mathematical reasoning model scoring 87/120 on Putnam competition questions—equivalent to 2nd place among human participants despite being a smaller 30B parameter model with specialized reasoning capabilities.

Starcloud, backed by NVIDIA, successfully operated an H100 GPU and Gemma model on a satellite in orbit for the first time. This proof-of-concept for orbital data centers aims to bypass terrestrial energy constraints, with plans for a 5-gigawatt orbital facility by 2026.

The Trump administration authorized NVIDIA to sell H200 AI chips to China, conditioned on the U.S. government taking a 25% cut of sales revenue. While intended to maintain tech dominance, experts warn this export control reversal could accelerate China's military AI capabilities.

AWS launched Nova Forge, a service allowing enterprises to fine-tune custom "Novella" models using proprietary data without the model forgetting its base training, designed to compete with GPU-heavy customization processes required by other providers.

Google promoted Amin Vahdat to the newly created C-suite role of Chief Technologist for AI Infrastructure, reporting directly to CEO Sundar Pichai. Vahdat will oversee allocation of the company's projected $93 billion capital expenditure for AI infrastructure.

State Attorneys General issued a warning letter to Google, Meta, and OpenAI with a January 16, 2026 deadline to address "delusional" outputs and safety risks or face legal action, citing specific harms including incorrect medical advice and inappropriate interactions with minors.

New York Governor Kathy Hochul signed legislation requiring advertisers to conspicuously disclose AI-generated avatars in commercial content—the first such U.S. law—with fines up to $5,000 for repeated violations and new rights regarding digital likenesses of deceased persons.

Disney filed a cease-and-desist against Google alleging its AI models infringe copyright on a "massive scale," contrasting sharply with Disney's concurrent OpenAI partnership and suggesting a strategy of creating licensed content silos for AI generation.

RSL 1.0 (Really Simple Licensing) was officially released as a standard allowing publishers to dictate terms and demand payment for AI scraping. Supported by major infrastructure providers like Cloudflare, this protocol offers granular control beyond the binary robots.txt for managing content usage in AI training.