Good morning, and welcome to your Monday briefing. It's April 20th, 2026.
Sunday morning at Cape Canaveral, Blue Origin achieved something its founder has been chasing for decades: the first successful reuse and landing of a New Glenn booster. Named Never Tell Me the Odds — yes, it's a Star Wars reference — the flight-proven first stage came in hot and touched down on the droneship in the Atlantic. That's a monumental milestone for any heavy-lift vehicle. The catch? The upper stage failed to deploy its customer satellite. AST SpaceMobile's BlueBird 7 communications satellite was carried into the wrong orbit, meaning the primary mission objective went unfulfilled. Third flight for New Glenn, and it's a classic rocket development story: one big win, one problem left to solve. SpaceX has been landing Falcon 9 boosters for nearly a decade, so the clock's ticking on Blue Origin to turn reuse from a milestone into a routine.
In Beijing yesterday, a humanoid robot called Lightning — built by Chinese smartphone maker Honor — completed the E-Town half marathon in 50 minutes and 26 seconds. That's not just fast for a robot. That's faster than any human has ever run that distance. The human world record is over six minutes slower. To put this in perspective: at last year's inaugural humanoid robot marathon, the fastest robot finished in two hours and forty minutes. In twelve months, that pace dropped by nearly an hour and fifty minutes. Multiple robots competed, and they ran on parallel tracks to avoid collisions with the human runners. The engineering implications are staggering — bipedal locomotion at this speed, for this duration, requires serious advances in dynamic balance, power management, and joint actuation. China is making it very clear that humanoid robotics isn't science fiction anymore.
The AI chip space keeps getting crowded, and Cerebras Systems is making its move. On April 17th, the Sunnyvale startup filed its S-1 for a Nasdaq listing under the ticker CBRS. This is their second attempt — they pulled a previous filing back in 2025 — but the numbers look dramatically different this time around. Revenue hit $510 million in 2025, and they've got a deal with OpenAI that validates the technology at scale. The key differentiator: Cerebras makes dinner-plate-sized wafer-scale processors that sidestep the industry's dependence on high-bandwidth memory — one of the biggest bottlenecks in AI hardware right now. If you're wondering why Nvidia's stock gets twitchy every time another chip company files for IPO, this is why.
According to The Information, Google is in discussions with Marvell Technology to co-develop two new types of AI chips: a memory processing unit designed to work alongside Google's existing TPUs, and a new inference-focused TPU. Marvell shares jumped about seven percent in premarket trading Monday. This adds Marvell as a third design partner in Google's custom silicon pipeline — Broadcom already makes TPUs for Google's hyperscale deployments, and custom ASIC sales for AI inference are projected to grow 45 percent in 2026. Google's message to the market is clear: in-house silicon for in-house models, with multiple partners to keep the supply chain diversified. That's a lesson learned the hard way from over-relying on single vendors.
That's it for today's briefing. See you tomorrow.