What’s at stake for next week’s Starship flight, where the real AI safety focus should be and more in this week’s Prototype. To get it in your inbox, sign up here .

Next week , SpaceX plans the first orbital flight of its Starship rocket, a mere five years or so later than originally planned .

The Elon Musk-founded rocket company has big plans for this first flight. Beyond getting to orbit for the first time–and conducting the many systems tests that entails–Starship will also have 26 of its next generation Starlink satellites on board, and each one will have 10x the broadband capacity of the current generation.

Starlink is increasingly important to the nearly $2 trillion (market cap) company. The internet service pulled in $11.4 billion in revenue in 2025 (out of a total $15.5 billion) and Morgan Stanley projects Starlink’s revenue alone will grow to $18.5 billion by the end of the year.

The company’s internal focus reflects this. In 2020, around half of SpaceX’s rocket launches were geared towards deploying Starlink satellites. Today, that’s closer to 80%–even as its number of launches per year has steadily increased. That’s caused a squeeze for the space industry, leaving some satellite operators waiting years to get their spacecraft to orbit.

Meanwhile, other launch providers who could meet that demand like Blue Origin have seen multiple delays, leaving customers desperate to get their moneymaking satellites in the air.

In theory, Starship should ease this bottleneck. It has a much larger payload capacity than SpaceX’s workhorse Falcon 9 rockets, and Elon Musk has repeatedly promised there will be “hundreds” of launches a year once Starship’s operations are fully up and running.

But in the meantime, the company is already beginning to sunset some planned Falcon 9 deployments, analyst Kim Burke at Quilty Space told me. And even if Starship does get up and running, that doesn’t necessarily mean relief for customers. “Not sure industry is clocking that an operational Starship might be an oxygen mask that falls out of the ceiling of a plane–but SpaceX puts it on first before helping the rest of the cabin,” she said.

Starship also has a long way to go. This orbital flight is important, but it’s not testing key systems, Burke added. “No booster catch. The V3 booster has yet to complete a landing burn in two attempts, and this one is headed for the Gulf. No ship catch, either. And no V3 stage has reflown,” she noted. “That leaves an uncomfy gap between what Flight 14 can prove and what SpaceX has already wagered.”

All that said, Burke noted that SpaceX’s latest moves suggest confidence that it has ironed out Starship’s issues, and may hit its target of beginning commercial operations in the fourth quarter. “You don’t push all your chips in unless you’re sure you’ll hit,” she said, before adding. “Well, it’s SpaceX, so I guess they might.”

Focusing On AI’s Real Risks

A lot of digital ink has been spilled over an “existential” threat from AI over the past few weeks, with leaders of some frontier AI companies like Anthropic, OpenAI and SpaceX suggesting that the pace of AI development should slow down in order to prevent potential risks.

Bill Gates is the latest tech leader to chime in, recently stating that AI tools could “cause a billion deaths”–which is relatively conservative compared to the claim endorsed by Sam Altman and Dario Amodei that “mitigating the risk of extinction from AI should be a global priority.”

The case for AI as an existential threat seems rather attenuated to me, with a lot of unstated assumptions in the chain of logic. That’s not to say there aren’t risks involved in current AI tools. There are quite a few, but most of them fall into two categories.

First: AI tools can amplify malicious behavior. Take deepfakes: They didn’t come into existence with AI–they just required much more skill to produce. But AI tools have caused an explosion of AI-powered misinformation across the internet. They can also speed up hacking–my colleague Tom Brewster showed this week how a Chinese hacker was able to attack more than 100 companies using agents from Anthropic and DeepSeek—and boost common scams , which might cost the economy billions.

Second, it’s become clear that AI agents have capabilities that far outpace their governance. At their core, AI agents run on probabilities rather than certainties, their performance can be difficult to predict upon deployment. My colleague Brewster also reported earlier this month that a company deployed models for testing, only to discover that the agents had inadvertently hacked several organizations. One of OpenAI’s models inadvertently breached Australia’s healthcare databases earlier this year, and the company reported on other “misalignment” issues earlier this month along with a promise to disclose such events in a more transparent and timely manner.

Every major industry on Earth conducts rigorous safety tests of its products. AI companies shouldn’t be different. Which is why moves like OpenAI’s transparency pledge and Anthropic partnering with Accenture to independently test its products seem more prudent to me than some of the headline debates, because they’re paying attention to real issues rather than sci-fi ones.

In my mind, focusing on AI risks should mean being less worried about Skynet hijacking the nuclear arsenal and more about an AI agent virtually robbing a bank when asked to help a user save money on plane tickets.

Space Stuff: Hubble Network, which connects satellites to devices on Earth via Bluetooth, just raised a $200 million round to support building a 60-satellite constellation in orbit. Meanwhile, Asteroid mining startup Astroforge is developing an AI model , called Solo, which it says will be able to autonomously fly in deep space, no human mission control needed. It’s planning a demonstration mission with a spacecraft powered by the model.

AI-powered Biotech: Enveda, which uses AI to look for drug candidates that naturally occur in plants and microbes, raised a $311 million series E round to support its ongoing clinical trials. Basecamp Research, which has built proprietary datasets to support AI in finding potential new disease treatments, raised a $140 million series C round. And frontier AI giant Anthropic is setting up its own labs to support drug discovery research.

Humanoid Robots: Both Tesla and China-based Xpeng have started to scale up production of humanoid robots, though a lot of questions remain about how practical they’ll be to mass produce and how useful they’ll be once deployed to customers. Meanwhile, Japan is reportedly exploring the military potential of humanoid robots–what could go wrong there?

What’s Entertaining Me This Week

I’m about three-quarters of the way through Wisdom Corner , the latest from David Heska Wanbli Weiden. It’s a sequel to his 2021 novel, Winter Counts , again centered on the character Virgil Wounded Horse, who works as a vigilante on a Lakota reservation in South Dakota. On reservations, only the Department of Justice can legally handle felony cases–and it often declines to do so. Virgil fills the gap by meting out street level justice to bad guys–but the books also interrogate that choice and what it costs him. These are great classic noir stories in a setting often given short shrift in pop culture, and Weiden’s prose is sharp and engaging. I’m already looking forward to the next book if Weiden chooses to write one.