AI This Week

NOAA unveiled AI-driven forecasting systems that promise quicker, more accurate predictions at a fraction of computing cost. Officials switched them on Wednesday for use. The new models complement, not replace, physics-based staples like the Global Forecast System and GEFS. Trained on decades of analysis data, they cut compute needs by 91% to 99% and can extend forecast skill by up to a day. AIGFS produces a 16‑day outlook using just 0.3% of GFS resources and finishes in about 40 minutes. AIGEFS adds probabilistic guidance, while a Hybrid‑GEFS blends AI with traditional ensembles to capture uncertainty. NOAA still works on hurricane guidance and outcome diversity. Leaders cite lower costs but note the heavy energy footprint of model training.

Applications to the UK Civil Service AI & Data Challenge jumped 160% year over year, with 252 ideas submitted versus 97 last time. DSIT, the Cabinet Office, and NTT DATA UK&I run the program, which invites staff to propose AI and data uses to improve services, then builds cross‑department teams to pitch to departmental CIOs. The winner receives £50,000 of development support. DWP led submissions, with HMRC and Defra tied for second. A record 339 civil servants volunteered to join project teams. Judges will pick eight ideas to advance in the new year. Past winners include AI4Peat, which mapped UK peatland drainage at national scale, and Project Constellation, which created a real‑time view of prison accommodation to save officers’ time.

Global data center investment hit US$61 billion in 2025, driven by surging AI workloads that demand dense compute, advanced chips, and reliable power. The total spans mergers, acquisitions, and spending on new builds and upgrades across major markets, marking the sector’s strongest year yet. Hyperscalers including Microsoft, Amazon, and Google push expansion while tapping bond markets and private equity, shifting from cash-only funding. More than 100 deals show broad participation. McKinsey projects AI-related data center spending could reach US$7 trillion by 2030. Virginia and Texas lead in the United States, with Europe and parts of Asia drawing capital for low‑latency services. Power constraints loom, prompting grid strategies, long‑term contracts, and on‑site generation. Developers pursue renewables, nuclear, and advanced cooling amid ROI and community concerns.

Amid growing concern over AI in film and TV, a group of entertainment figures has launched the Creators Coalition on AI to defend creators’ rights and set clear standards. The 18 founders include Daniel Kwan, Joseph Gordon-Levitt, Natasha Lyonne, Janet Yang, David Goyer, Paul Trillo, and others. They position CCAI as a cross-industry hub that will pursue four pillars: transparency; consent and compensation for content and data; job protection with transition plans; guardrails against misuse and deepfakes; and safeguarding humanity in the creative process. More than 500 artists back the effort, including Cate Blanchett, Rian Johnson, Phil Lord, Kristen Stewart, and Taika Waititi. The coalition formed after a wave of tech agreements that alarmed creators and sparked demands for shared principles.

2025 marked a turning point as travel brands shifted from conversational bots to operational AI that drives bookings, revenue, and faster service across channels. Maya’s COO Benjamin Manzi outlines five shifts: production over pilots; conversion over conversation; trust built through data governance, hallucination prevention, brand tone discipline, and risk management; augmented agents that amplify human teams; and deep integration with live inventory and workflows. The outlook for 2026 sharpens the focus: reliability, scale, and intent‑driven discovery. Expect quicker responses, sharper lead qualification, and more tailored guidance, with clear guardrails and human oversight. Systems that handle real volumes and edge cases will win, and only a few will scale across markets and languages.

Experts at Stanford’s Institute for Human-Centered AI forecast a dramatic shift for artificial intelligence in 2026, predicting the year will mark a turn from creative hype to sober measurement. This new era will subject systems to exacting tests for accuracy, risk, and value. Computer scientists anticipate a surge in AI sovereignty, with nations building their own models and data centers. They also project new interfaces will move beyond today's chatbots. Meanwhile, legal scholars expect domain-specific benchmarks will hold AI accountable, and healthcare leaders predict hospitals will demand strict return-on-investment frameworks for new tools.

The U.S. Department of Energy has signed agreements with 24 organizations, including Microsoft, Google, Nvidia, Amazon Web Services, IBM, Intel, Oracle, and OpenAI, to advance its Genesis Mission. The initiative seeks to apply artificial intelligence to speed scientific discovery and bolster U.S. energy and security capabilities. It aims to lift scientific productivity and curb dependence on foreign technology. Partners will build AI models for nuclear energy, quantum computing, robotics, and supply chain optimization. The effort follows a White House executive order directing AI deployment in energy innovation, advanced manufacturing, and national security. It extends prior DOE work with industry on high-performance computing at Argonne and Los Alamos labs. The department plans wider ties with universities and non-profits.

Texas startup Last Energy has raised $100 million to build and deploy mini nuclear reactors for AI data centers. The company plans modular units that deliver steady, carbon-free power near compute clusters. The funding backs factory production, site development, and customer agreements aimed at round-the-clock reliability as data center demand surges. Power-hungry AI training strains grids, and companies seek cleaner supply. Last Energy pitches shorter build times, standardized components, and predictable costs compared with large nuclear plants. The company targets private power deals that bypass grid bottlenecks and match data center load. Contracts would supply dedicated, on-site power to new global server farms. Investors are betting that next-generation nuclear can anchor the next wave of AI growth across markets

Nvidia is moving beyond selling AI chips by releasing Nemotron 3, a family of open models plus the training data and tools to customize them. The lineup includes Nano (30B parameters), Super (100B), and Ultra (500B). Nvidia says benchmark scores place the models among the best downloadable options. The company also shipped a hybrid latent mixture‑of‑experts design for agent building and libraries for reinforcement learning. The push arrives as OpenAI, Google, and Anthropic develop their own chips and as US firms share less about research, while Chinese companies release powerful open models more frequently. OpenRouter data shows open models handled about a third of tokens in 2025.

Google is turning any pair of headphones into a live translation device. Using its Gemini AI, the company’s Translate app now delivers real-time, speech-to-speech translation directly to your ears. This is not a recording; it's a conversation. The new beta function supports more than 70 languages and works to preserve the tone and cadence of the original speaker, resulting in more natural dialogue. Beyond live audio, the update also improves text translation by better interpreting context from idioms and slang. Google is also adding new tools for language learners. The initial beta is available for Android users in the US, Mexico, and India, with a wider release expected in 2026.