AI Briefing

Iran just publicly threatened to annihilate OpenAI's $30 billion Stargate data center in Abu Dhabi, which means AI infrastructure is now officially a military target. Meanwhile, Google dropped Gemma 4 under Apache 2.0 and it's already the #3 open model in the world, and a Tufts research team figured out how to cut AI energy use by 100x by teaching neural networks to actually think logically. Oh, and OpenAI is quietly reshuffling its entire C-suite weeks before a potential IPO.

01IndustryTom's Hardware ↗

Iran's IRGC Threatens 'Complete Annihilation' of OpenAI's $30B Stargate Data Center in Abu Dhabi

Iran's Islamic Revolutionary Guard Corps released satellite imagery pinpointing OpenAI's 1-gigawatt Stargate data center in Abu Dhabi and threatened its 'complete and utter annihilation' if the US strikes Iranian power plants. The threat follows actual Iranian drone strikes on AWS data centers in the UAE and Bahrain in March that knocked two of three availability zones offline for over 24 hours. This marks a chilling new phase where AI compute infrastructure has become an explicit military target, forcing the industry to reckon with the fact that concentrating billions of dollars of strategic compute in geopolitically sensitive regions carries real wartime risk.

02Open SourceGoogle Blog ↗

Google Releases Gemma 4, Its Most Capable Open Model Yet, Under Apache 2.0

Google DeepMind released Gemma 4, a family of open-weight models in four sizes (2B, 4B, 26B MoE, and 31B Dense) built on the Gemini 3 architecture. The 31B Dense model currently ranks #3 globally on the Arena AI leaderboard, outcompeting models 20x its size, while the 26B MoE activates only 3.8 billion parameters during inference for exceptional speed. The entire family ships under an Apache 2.0 license with support for 140+ languages, native function calling, and video/audio inputs. This is Google's strongest play yet in the open-weights race, and the permissive license is a direct challenge to Meta's Llama and the wave of Chinese open models.

03ResearchScienceDaily ↗

Neuro-Symbolic AI Breakthrough Cuts Energy Use by 100x While Boosting Accuracy

Researchers at Tufts University led by Matthias Scheutz demonstrated a neuro-symbolic AI approach that slashes energy consumption by up to 100x compared to standard neural networks while dramatically improving accuracy. Their hybrid system, which combines neural networks with human-like symbolic reasoning, achieved a 95% success rate on the Tower of Hanoi puzzle versus just 34% for conventional models, trained in 34 minutes instead of over a day, and used only 1% of the energy for training. The work will be presented at the International Conference on Robotics and Automation in Vienna this May, and represents a potential paradigm shift as the AI industry grapples with data centers consuming 415 terawatt hours of power annually.

04IndustryCNBC ↗

OpenAI Reshuffles C-Suite as COO Moves to 'Special Projects' and AGI CEO Takes Medical Leave

OpenAI announced a sweeping leadership reshuffle: COO Brad Lightcap moved to 'special projects' overseeing complex deals and a private equity joint venture, AGI CEO Fidji Simo took medical leave for her neuroimmune condition POTS, and CMO Kate Rouch stepped down to focus on cancer recovery. Former Slack CEO Denise Dresser will absorb Lightcap's commercial duties, while President Greg Brockman covers for Simo and former Meta CMO Gary Briggs takes over marketing on an interim basis. The shakeup comes at a precarious moment, with OpenAI approaching $25 billion in annualized revenue and reportedly taking early steps toward an IPO, raising questions about organizational stability at the most consequential AI company in the world.