Good morning, and welcome to the voicecast. It's Monday, June 8th, 2026.

**SpaceX booster B1067 sets a new reusability record — 35 flights and counting.** SpaceX Falcon 9 booster tail number B1067 lifted off from Cape Canaveral Space Force Station at 6:13 AM Eastern, deploying 29 Starlink satellites to low Earth orbit — and completed its 35th successful launch and landing. The booster touched down on the autonomous droneship ASOG in the Atlantic. This one first flew in 2023 and has been flying roughly every fifteen days since. The launch happened just eleven minutes before sunrise, so you got the jellyfish effect — early morning sun illuminating the rocket exhaust for a spectacular show. SpaceX has now conducted 67 Falcon family launches in 2026. Gwynne Shotwell's target is 140 to 145 this year, and they remain on pace. The bigger story: this level of reusability — one booster flying 35 times — makes Starship's development challenge clear. SpaceX needs Starship to match this cadence with even higher payload mass. Meanwhile, Starship Flight 13 preparation continues at Starbase. Booster 20 is undergoing cryogenic testing, and the FAA mishap investigation from the Flight 12 booster failure remains in progress.

**NVIDIA drops Nemotron 3 Ultra — a 550 billion parameter open-model with a hybrid architecture.** NVIDIA released Nemotron 3 Ultra on June 4th, and it's worth a close look. 550 billion total parameters, but only 55 billion active per inference — that's a mixture of experts design. The twist: it uses a hybrid Mamba and transformer architecture. Mamba is a state space model that handles long context in linear time instead of quadratic. A million token context window. 89.1 percent on MMLU. Throughput around 140 tokens per second, and up to five times faster on Blackwell with NVFP4 quantization. But here's the important part: NVIDIA is releasing this with open weights, open training data, open recipes. Not a gated API. An actual open model. It's already being adopted by Glean for enterprise search, Nous Research's Hermes Agent, Lighting AI, and NAVER is building a Korean variant on top of it. The combination of frontier-scale size, Mamba efficiency for long context, and full openness makes this a genuinely interesting counterpoint to the closed models.

**UBTECH takes over 2,100 pre-orders for the UWORLD U1 consumer humanoid.** Chinese robot maker UBTECH opened pre-sales for their UWORLD U1 — a full-size hyper-bionic humanoid designed as a consumer companion robot, complete with realistic skin and simulated tears. In six days, they pulled in over 2,100 pre-orders. Their Hong Kong stock rallied nearly six percent on the news. Full public presentation is set for June 30th, with first shipments to pre-order clients arriving in mid-September. And this gets to the central question nobody in robotics wants to avoid: who actually buys these things and can companies build enough of them? China has massive state backing for humanoid production. The bottleneck has always been demand, not supply. A consumer-facing humanoid with two thousand pre-orders in a week might be the first real signal that people will actually open their wallets for robot companions. If that holds, it changes the entire market calculus for Figure, Tesla, Unitree, and everyone else racing to ship.

That's all for today's voicecast. See you next time.