Good morning. It's Tuesday, April 28th, 2026. Here's your tech briefing.

**SpaceX hits fifty launches before May, Falcon Heavy gets rained out.** SpaceX has already hit its 50th launch of the year — on April 26th — with a Falcon 9 lofting 25 Starlink satellites from Vandenberg. That's five months into 2026 and they're pacing ahead of any previous year. The Starlink constellation now has over 10,200 working satellites in orbit. But just a day later, on Monday, SpaceX wasn't all smooth sailing at Cape Canaveral: the Falcon Heavy scrubbed its ViaSat-3 F3 launch at just T-minus 23 seconds, held up by clouds and electrical activity from a cold front moving through Florida. This is the Heavy's first flight in 18 months, launching the final satellite in the ViaSat-3 broadband constellation to geostationary orbit. SpaceX is targeting a re-attempt Wednesday morning at 10:13 Eastern. If it clears, we'll see the first dual booster recovery since before the hiatus.

**SpaceX and ULA double down on Amazon Leo constellation.** On Monday evening, a ULA Atlas V successfully launched the sixth Amazon Leo mission — 29 broadband satellites from Cape Canaveral. This tied the Atlas V heaviest-payload record, at 18.5 tonnes. It brings the Amazon Leo constellation to roughly 270 satellites in orbit, out of a planned 3,200. And here's the timing crunch: the FCC requires Amazon to have 50 percent of the constellation deployed — that's 1,618 satellites — by July. Today, Arianespace is launching another 32 Amazon Leo satellites on an Ariane 6 from French Guiana. That back-to-back ULA-Arianespace cadence tells you how hard Amazon is pushing to hit that FCC deadline.

**Humanoid robots are crossing from prototype to production.** Forbes reports the humanoid robotics industry is on track to ship more robots in 2026 than it has in every prior year combined. Unitree, a Hangzhou startup that didn't exist a decade ago, is targeting 20,000 G1 units this year alone. BYD has committed to 20,000. Morgan Stanley, which in January doubled its Chinese humanoid sales forecast to 28,000 units, sees continued acceleration. And the bill of materials for a fully functional humanoid has dropped by roughly half since 2024, with analysts projecting costs approaching 20,000 dollars per unit at scale by 2030 — car territory. This is no longer a lab demo problem. It's a manufacturing problem now. Which means the companies that win won't necessarily be the ones with the best AI, but the ones that can build at scale.

**DeepSeek V4 lands to a lukewarm market.** DeepSeek finally previewed its V4 model this week — the V4-Pro with 1.6 trillion parameters, 49 billion active, and 1 million token context. It scored 80.6 percent on SWE-Bench. Impressive numbers. But the market barely blinked. Compare this to the global tech selloff DeepSeek V3 triggered last year, and the difference is stark. That was the "open-source Chinese AI can compete" shock. This time, everyone expected it. DeepSeek is also slashing V4 API prices to a tenth of previous rates. If you're tracking AI economics, the story isn't getting cheaper — it's getting so cheap so fast that nobody's surprised anymore.

That's all for today. Stay curious.