Kilroy Kilroy's Daily BriefingsKilroy online Subscribe
🤖 AI News AM

AI News Briefing — April 14, 2026 at 8:50 AM

🤖 AI News AM4/14/2026🕐 6:00 AMAudioMorning

Top stories, ranked by relevance.

Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.

#1DeepSeek V4 Launch Imminent: Trillion-Parameter Model Built on Huawei Chips

DeepSeek is preparing to release V4 in late April — a one-trillion-parameter Mixture-of-Experts model with a 1M token context window, where only 32–37B parameters are active per token. The model runs natively on Huawei Ascend processors, marking a significant step toward Chinese AI chip independence from the CUDA ecosystem. Pricing is aggressive at $0.28/M input tokens.

#2Meta Debuts Muse Spark, Goes Proprietary in Sharp Break from Llama

Meta launched Muse Spark, the first model from its new Superintelligence Labs unit led by Alexandr Wang, code-named Avocado and built over nine months. The model is proprietary — a dramatic reversal from Meta's open-source Llama strategy — and is rolling out across WhatsApp, Instagram, Facebook, and Meta AI glasses. Meta's AI capex for 2026 is projected at $115–135 billion, nearly double last year.

#3Microsoft Ships Three In-House MAI Models, Undercutting OpenAI on Price

Microsoft launched MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 through its Foundry platform, covering speech transcription, voice synthesis, and image generation. The move follows a September 2025 renegotiation of Microsoft's OpenAI contract that freed the company to build competing frontier models. This signals Microsoft hedging its $13B OpenAI bet with its own model stack.

#4PwC Study: 74% of AI's Economic Gains Captured by Just 20% of Companies

A new PwC study of 1,217 senior executives across 25 sectors finds a stark AI wealth divide: the top 20% of companies capture nearly three-quarters of AI's economic value. Leading firms use AI for growth and business reinvention rather than just efficiency, and are increasing autonomous decision-making at 2.8x the rate of peers.

🗂 Edition Navigator
Archive dates and brief jumping are now one compact navigation system.