Musk commits xAI to more gas turbines; NVIDIA ships Nemotron Diffusion

TechCrunch reports xAI is now buying $2.8 billion more unregulated natural-gas turbines to power its Memphis-area data centers, abandoning the "solar-electric economy" commitment Elon Musk wrote into Tesla's 2023 Master Plan 3. The pitch for the eventual fix: SpaceX-launched orbital solar arrays that the company claims can deliver more than five-times the energy of terrestrial panels with constant illumination. A SpaceX filing argues "terawatt-scale annual AI compute growth" will eventually force data centers off-planet — a long-horizon framing that does not yet address the methane the Memphis turbines burn today.

NVIDIA Labs has quietly posted Nemotron Diffusion to Hugging Face — a 14-billion-parameter language model that switches between autoregressive decoding, parallel diffusion decoding, and a "self-speculation" mode that drafts with diffusion and verifies with autoregression, all without changing model weights. The accompanying technical report claims a 2.2-times throughput lift over the comparable Qwen 3 8-billion-parameter baseline at matched accuracy, scaling to 850 tokens per second on a GB200 (a 3.3-times lift). Base, instruct, and vision-language variants are all open-weight; the architecture is positioned as a path from memory-bound to compute-bound inference as GPUs keep outrunning memory bandwidth.

Axios reports that the Genesis Mission — the federal scientific-research push launched late last year to fuse quantum computing with AI — has selected Reflection AI as the foundational intelligence layer for the Department of Energy's 17 national laboratories. Reflection's customizable open-source models will run on DOE compute and be deployed across active research projects. CEO Misha Laskin framed the choice as a policy bet, telling Axios "you can't do scientific discovery on a closed model" — a deliberate counterpoint to the NSA's Anthropic contract that the White House cleared yesterday.

Leading the Future, the OpenAI- and Andreessen Horowitz-backed super PAC that surfaced earlier this spring, is doubling down in Kentucky's Senate race with a $750,000 buy supporting Rep. Andy Barr through the primary and general election. The group plans roughly $140 million in total 2026 cycle spending and has explicitly framed its mission as countering "AI-doomer sentiment" and securing congressional allies for a single federal regulatory framework — a deliberate alternative to the state-by-state laws now active or pending in California, Colorado, New York, and Texas. The Kentucky move is the third announced spend after earlier plays in primaries that already broke for industry-friendly candidates.

Axios reports the National Sheriffs' Association is acknowledging a steady decline in traditional neighborhood-watch enrollment as Ring doorbells, Nextdoor alerts, and license-plate readers automate community monitoring across thousands of US municipalities. Ann Arbor, Michigan removed more than 600 watch signs after concluding the program was encouraging racial profiling, and criminal-justice researcher Mary Dodge points to platform reports such as "I saw a person of color walking through my neighborhood, and I don't like it" as evidence that automation is amplifying bias rather than filtering it out. The story is the clearest local-government data point so far in the AI surveillance debate.

IEEE Spectrum covered new work from the Technical University of Denmark and Trinity College Dublin: a portable millimeter-wave radar paired with a machine-learning classifier reads micro-Doppler signatures from individual flying insects and identifies them to species with 85 percent accuracy, and to family (bees versus wasps) with 96 percent accuracy. The team trained on more than 70 radar-reflection features per insect; the peer-reviewed result was published in PNAS Nexus on April 28. The proposed next step is a field-deployable trap that lets insects fly through, classifies them in real time, and releases them unharmed — a foundation for a global pollinator-population database as wild bee numbers continue to crash.

A new University of Oxford paper introduces CUSP — Cutoff-conditioned Unseen Scientific Progress — a 4,760-event benchmark that asks frontier AI systems to forecast which research directions are feasible, explain the underlying mechanism, design a candidate solution, and predict timing. Across biology, chemistry, and physics, the models reliably pick plausible directions but "fail to reliably predict whether scientific advances will be realized" and show systematic timing errors with overconfident uncertainty estimates. Curiously, AI progress itself is more predictable than progress in the natural sciences — a useful tempering of the "AI scientist" narrative as Google's Co-Scientist and SandboxAQ-on-Claude integrations roll out to actual research labs.