AI Daily Brief — 14 January 2025
Tuesday’s headline was Biden’s second AI executive order in two days — EO 14141, directing the Department of Defense and Department of Energy to lease federal land for gigawatt-scale frontier AI data centers, paired with new clean-power generation to match the load. MiniMax dropped a technical-report bombshell: a 456-billion-parameter MoE with a novel Lightning Attention architecture and a 4-million-token inference context. Robotics-foundation-model startup Skild AI raised $1.4 billion.
Top stories
- Biden signs EO 14141 — “Advancing United States Leadership in AI Infrastructure.” The order directs DoD and DoE to each identify at least three lease-ready frontier-AI-datacenter sites on federal land by Feb 28; competitive 30-day solicitations open Mar 31; winning bidders begin construction by Jan 1 2026 and must reach full capacity by Dec 31 2027. The Interior Department must designate at least five Priority Geothermal Zones by Mar 15. Developers must procure matching new clean power (geothermal, nuclear, solar, wind, hydro, marine). via White House · via Data Center Dynamics
- MiniMax posts MiniMax-01 paper on arXiv — 456B parameters, 4M-token context, Lightning Attention. The technical report (2501.08313) describes a hybrid Mixture-of-Experts architecture interleaving seven lightning-attention blocks with one softmax block across 80 transformer layers, 45.9B active parameters, trained on 1M-token context and extrapolating to 4M tokens at inference. Claimed parity with GPT-4o and Claude 3.5 Sonnet at 20-32x longer context. Public weights follow the next day. via arXiv
- Skild AI raises $1.4B for an “omni-bodied” robotics foundation model. One of the largest single-day funding events in robotics history; Pittsburgh-based Skild positions itself as the company building one general model that ships across diverse robot bodies. via TechStartups
- Aikido Security hits unicorn status with $60M round. The dev-time defense platform — security tooling embedded into developer workflows — closes a $60M raise that vaults it past $1B valuation. via TechStartups
Who shipped
Western frontier labs were quiet on products. OpenAI, Anthropic, Google DeepMind, and Meta made no dated launches; the dominant noise around their corner of the industry was post-Diffusion-Rule lobbying.
Open-source pulse
MiniMax‘s paper was the day’s open-source signal. Lightning Attention positions linear-style attention as a credible alternative to vanilla Transformer attention at frontier scale — a thesis many smaller labs had argued and few had demonstrated at 456B.
Research & papers
“Addressing the Sustainable AI Trilemma: A Case Study on LLM Agents and RAG” (2501.08262) flagged the energy-equity-productivity tradeoff embedded in agent and RAG designs — a continuing theme as the inference-time scaling era began consuming compute at production rates.
By the numbers
- 3 federal sites per agency required by Feb 28 under EO 14141
- 456B total / 45.9B active parameters in MiniMax-01
- 4M tokens — MiniMax-01 extrapolated inference context
- $1.4B — Skild AI’s robotics round
- Most-mentioned actor: Biden administration
Compiled by AI Feed’s editor from verified web sources for 14 January 2025.