Good morning. It's Saturday, May 2nd, 2026, and here's what's happening in tech and science.

**Story one.** Meta is going physical. The company has acquired Assured Robot Intelligence, a humanoid robotics startup, for an undisclosed sum. ARI's team — including co-founders Xiaolong Wang, formerly of Nvidia and UC San Diego, and Lerrel Pinto, formerly of NYU and Fauna Robotics — now joins Meta's Superintelligence Labs division. This is a direct signal that Meta believes the path to AGI runs through the physical world. The startup was building foundation models for humanoid robots that can perform household chores and general physical labor, with models designed to understand, predict, and adapt to human behavior in complex environments. It follows Meta's Muse Spark model release and its earlier Dreamer AI acqui-hire. When your superintelligence lab starts hiring robot brain surgeons, that's a bet worth watching.

**Story two.** DeepSeek has launched its V4-Pro-Max model — 1.6 trillion total parameters with 49 billion active, one million token context window, purpose-built for agentic coding, reasoning, and math. This is the company that set the industry on fire with V3, and V4 is the answer. Pricing is where it gets wild: the V4-Pro model is running at promotional rates that make it roughly five to ten times cheaper than comparable Western models. A promo period extends through May 31st, but even at full price, DeepSeek is attacking the cost structure of frontier AI from the bottom. GPT-5.4, Opus 4.7, Gemini 3.1 Pro — they all have a new problem, and it's made in Hangzhou.

**Story three.** In quantum computing, Dutch startup Groove Quantum — a spinout from TU Delft — has raised 16 million euros and unveiled an 18-qubit germanium spin-qubit processor. This matters because germanium spin qubits are CMOS-compatible, which means they can be manufactured in existing semiconductor fabs rather than custom dilution refrigerator shops. Groove's approach could be the first scalable path to quantum hardware that looks less like a physics experiment and more like a chip factory output. They've also won the European Innovation Council's backing. Small number of qubits, but the manufacturing story is the one to watch.

**And finally.** DeepSeek didn't just ship a new model — they also slashed V4-Pro pricing by 75 percent through the end of May. We're talking $3.48 per million output tokens at promo rates, versus $30 for GPT-5.4 and $25 for Opus 4.5. This is the same play V3 pulled last year, but with V4 the performance gap is closing. The combination of model capability plus aggressive pricing is forcing every lab in the world to ask the same question: how do we defend margins when the open-weight alternative is within striking distance?

That's all for today. Have a great weekend.