Good morning. It's Tuesday, July 15th, 2026 — and we've got a packed briefing today.

**Story one.** Australia just unveiled what it's calling the world's first comprehensive national AI framework. Prime Minister Albanese announced the plan this morning in Sydney — and it's the kind of thing other countries will probably copy. The new rules make data center operators legally responsible for underwriting their own electricity generation, paying full grid connection costs, cutting power use during peak demand, and meeting strict water efficiency standards. Alongside that, the government is establishing a new Office of AI with regulatory backing. Critics are already calling hyperscale data centers "energy vampires," but the Albanese government is walking a tightrope — trying to capture AI investment while protecting grid stability. If this model works, expect it to show up in EU and US state-level policy proposals within the year.

**Story two.** Politico has a deep dive this morning on Anthropic's lobbying strategy — and it's a fascinating contrast with OpenAI. While OpenAI lobbies for uniform, streamlined federal AI rules, Anthropic is doing the opposite: pushing states to adopt progressively tougher AI safety laws. Their head of state and local lobbying, Cesar Fernandez, described a deliberate march through California, New York, and Illinois — each one setting a higher bar than the last. It's classic regulatory one-upmanship. The strategy lets Anthropic position itself as the safety-first company while raising the compliance floor beneath competitors. Clever. Whether lawmakers recognize the chess move is another question.

**Story three.** NVIDIA is going all-in on the open model strategy. Yesterday they published a piece on their Nemotron open model initiative, positioning it as giving enterprises and nations "AI they can trust, control, and customize." And today in Japan, NVIDIA and partners are showcasing a full-stack AI and robotics push — covering everything from manufacturing to infrastructure. Combined with the Nemotron LangChain agent benchmarks from last week, the pattern is clear: NVIDIA is building an open model ecosystem that competes directly with the big proprietary labs. If your business AI needs a vendor-specific lock-in pitch, the Nemotron line is getting very compelling.

**Story four.** A new open-source model out of China is turning heads. Unisound just released the U2 — a Mixture of Experts with 266 billion total parameters but only 10 billion active at inference time. Benchmarks are competitive: Unisound claims scores in the mid-to-high 70s on GPQA Diamond and SWE-bench, with strong MATH results — though independent verification is still catching up. The price is notable too — at just 15 cents per million input tokens, it's orders of magnitude cheaper than frontier models. Another data point in the story that cheap, good-enough models are eating the inference market for batch workloads. If you don't need real-time interaction, the math on these Chinese MoE models is getting hard to ignore.

**Story five.** And in rocket news — Starship Flight 13 is go for tomorrow. SpaceX has rolled Booster 20 back to Pad 1 for final checkouts. The launch window opens Wednesday evening at 6:45 PM Eastern — that's 3:45 PM Pacific if you're planning to watch. This is the flight that's supposed to finally nail the booster landing after Flight 12's cascade failure. New hardware and software modifications across both stages, plus the first deployment of Starlink V3 satellites. Flight 13 has a lot riding on it.

And that's all for today. Have a great Tuesday.