A new design for orbital data centers. Greener cement. The sweet spot for getting enough sleep. All that and more in this week’s edition of The Prototype. To get it in your inbox, sign up here.
As America’s tech companies try to build infrastructure to keep up with the demand for AI, so follows the backlash from communities over concerns about water use and energy costs, especially as the price of electricity continues to climb. That’s why, as we’ve discussed a lot here, companies are increasingly looking to space as a solution. Google and SpaceX made headlines this week on word of a potential collaboration to build orbital data centers.
Another company entered the fray this week. That’s Cowboy Space Corporation (formerly known as Aetherflux), which this week raised $275 million at a $2 billion valuation with its own plan to build space-based data centers–20,000 of them, according to an FCC filing. Cowboy was founded by CEO Baiju Bhatt, who previously cofounded fintech company Robinhood. He told me what makes building these facilities in orbit attractive is long lead times and resource demands of installing them on Earth. “Space offers an abundance to solve this problem,” he said.
The biggest source of abundance? Energy. Bhatt said Cowboy will be launching its satellites into specific orbits that will keep the Sun shining on them 24/7, enabling them to get maximal power, as though “they were in the Sahara at lunchtime.” Another key differentiation for Cowboy is that they don’t design satellites within the standard confines of typical launch providers, its spacecraft will be purpose-built for data centers–which will be housed in the upper-stage of the rocket itself rather than being a separate satellite. Bhatt says this eliminates redundancies and will simplify design by letting the company start from scratch rather than try to fit into existing architectures. (This isn’t a new idea–the world’s first space station, Skylab, was built into the upper stage of a rocket.)
Using Earth’s orbit rather than Earth as a data center destination does pose significant challenges from both an economics and physics perspective, as I’ve covered before. But Bhatt believes many of those can be overcome with Cowboy’s plan to use the upper stage of its rocket to house 1 MW worth of chips, rather than distributing compute among many satellites as other companies plan to do.
“There are several advantages to reusing the upper stage and as much of the onboard hardware as possible for a data center, including mass savings–which can reduce launch costs–and larger operational volumes for computing equipment,” Nikoloas Gatsonis, an aerospace engineering professor at Worcester Polytechnic Institute told me via email. But with the scale comes challenges, he added, namely in dissipating the massive amounts of heat generated by the GPUs so they don’t damage the circuits.
George Lordos, a research scientist at MIT, echoed this concern to me, noting that a similarly scaled system to what’s on the International Space Station would be very heavy and might not result in any real economic advantage. Bhatt told me that Cowboy plans to use the faring of the spacecraft as a giant radiator to handle that heat, though design is still in progress.
Cowboy is unapologetically ambitious in its schedule, saying that it plans for a data center launch by the end of 2028. Lordos noted that this would be one of the fastest design-to-launch deployments in the history of commercial space, noting it took SpaceX and Blue Origin a decade to get to a reusable rocket like Cowboy plans to build. Bhatt acknowledges the challenge of this to me, but told me the company has hired veteran aerospace engineers like the former head of Blue Origin’s propulsion team, who have already managed this feat before.
“I’ve disrupted a big industry before,” Bhatt added. “I know what it takes. It requires moving with urgency and setting a bold and aggressive timeline and vision.”
Discovery of the Week: A Greener Way To Make Cement
Cement is a ubiquitous building material, found in streets and buildings around the world. It’s also a major contributor to climate change–just under 5% of total carbon emissions. Which is why laboratories and startups have been working on ways to make greener kinds of cement. But one challenge there is that even if someone does it, the industry may be slow to adopt a new kind of material.
That’s what’s interesting about a new paper published last week, which describes not a new formula for cement, but rather a less carbon-intensive process for making Portland cement, the most common type used.
Conventionally, cement is made by grinding up limestone and heating up to around 3,000 degrees. That uses an intense amount of energy, and releases CO2 as part of the process. The new method described in the paper, however, uses silicate rocks like basalt as the main ingredient, which are turned into cement electrochemically rather than through heat, which uses less energy.
This doesn’t mean companies are going to start changing their ways tomorrow. For one, the process right now is significantly more expensive, so more work will have to be done to bring that down. On the bright side, new industrial chemistries are often expensive at the start but come down in price as more efficient ways to accomplish the task are discovered. Also, this method has the potential to extract aluminum and iron from the rocks simultaneously, which adds more revenue opportunities for businesses that pursue it, even if the total cost isn’t as competitive as cooking limestone.
The Hot Take: The Brain Is A Data Frontier
Each week, I ask investors for their take on tech trends within their industries. Today the answers come from Luca Giani, a senior principal at Alumni Ventures who focuses on “mission-driven ventures” in healthcare. Giani himself has cofounded multiple biotech companies.
What tech is being overhyped right now?
The race to put the internet in your brain. There is genuine excitement around consumer brain-computer interface (BCI) applications, including messaging by thought, seamless AI integration, cognitive augmentation. And I understand the appeal. But we need to tread carefully. Brain interfaces that work reliably and safely for medical and health reasons are a very different proposition from consumer applications. The unintended consequences around what that does to the brain, its plasticity, and cognitive overloads are not trivial. The first important thing BCI can do in the next decade is help patients who have no options, and be a companion for everyone’s daily healthcare tracking and prevention. That is where the science should lead, and where the field should concentrate its credibility.
What should more people be talking about today?
Neural data as an AI training asset. The AI world is obsessed with compute: more chips, bigger clusters, faster inference. What we’re not talking about enough is the data frontier. High-quality recordings from the human brain, particularly from deep structures that current devices cannot safely access, are extraordinarily rare and extraordinarily valuable. As AI models for brain health, neural decoding, and biomarker discovery mature, the companies that have been quietly accumulating proprietary neural datasets will hold an asset that cannot be replicated easily. We are at the very beginning of understanding what that data is worth. Companies like Axoft [an Alumni Ventures portfolio company], able to generate high-density subcortical recordings from deep brain regions, are building the device businesses today and the data infrastructure businesses of the future. The AI community has not fully noticed this yet, but I bet it will.
What are we all going to be talking about in five years?
The democratization of BCI. I mean that in two directions simultaneously. On the clinical side, brain-computer interfaces will become a routine diagnostic tool in neurology ICUs and operating rooms, the way an EEG or an MRI is today. Real-time, high-resolution neural monitoring at the bedside will change how we assess concussions, stroke, TBI, disorders of consciousness, and seizure risk in ways that are simply not possible with current tools. On the consumer side, non-invasive neural wearables will follow the trajectory of continuous glucose monitors and smartwatch ECGs, moving from exotic and expensive medical devices to affordable, widely used health tools that track cognitive state, flag early neurological changes, and adapt in real time to improve outcomes. The combination of better hardware, AI-driven signal processing, and falling manufacturing costs will make this transition faster than most people expect. Five years from now, monitoring your brain health passively and continuously will seem as normal as tracking your heart rate.
On My Radar
Big Tech keeps moving into drug discovery: Isomorphic Labs, a subsidiary of Google’s parent company Alphabet, raised a $2.1 billion investment to further its goal of using AI to design new medicines, which it plans to bring to clinical trials soon. But it’s not alone: earlier this month, TikTok parent Bytedance started presenting its therapies at global scientific conferences. Nvidia, Microsoft and other Silicon Valley giants are also moving into the healthcare space. Meanwhile, the healthcare companies are moving into tech as the likes of Lilly and Roche are building supercomputers and software to accelerate the use of AI to find new treatments for disease.
Anthropic v. The Pentagon – Week 11: It’s been more than two months since the Department of Defense labeled Anthropic a “supply chain risk,” which theoretically bars it from being used by federal contractors for military purposes. Nevertheless, Reuters reports that the military is using the company’s Mythos model to identify and patch key software vulnerabilities.
Pro Science Tip: Get Enough Sleep–But Not Too Much
Getting enough sleep–the sweet spot is 6 to 8 hours a night–correlates to lower risks of disease and early death, according to a new study published this week. The research looked at the UK Biobank, which has around 500,000 participants who share their health, lifestyle and biological data. The researchers used this to look at the relationship between sleep duration and the individuals’ biological ages, which use different biomarkers from different organ systems to track aging. What they found was that a U-shaped pattern emerged: sleeping less than 6 hours a night and more than 8 hours were both associated with faster aging, with the least amount coming from those getting between 6.4 and 7.8 hours of sleep a day. Of course, this is just a generalized pattern and individuals vary, but it does support the idea that there may be an optimal amount of sleep for each person to maximize their health. The study isn’t definitive–it’s not clear whether the sleep duration led to the health problems or if the health problems disrupted sleep. More study remains, but in the meantime, if you’re getting around 6 to 8 hours, you’re probably in a good spot.
What’s Entertaining Me This Week
I recently watched the movie Deep Cover, distributed on Amazon, and enjoyed it immensely. The premise is pretty simple: a police officer (played by Sean Bean) hires an improv troupe (played by Bryce Dallas Howard, Orlando Bloom and Nick Mohammed) to go undercover for a minor sting operation. Suffice to say, the comedians are soon way over their head and find themselves in the middle of a war between rival gangs. Bloom was the standout for me, giving a positively unhinged performance as a “deep method” actor who does improv in between gigs twirling signs and doing pizza commercials.
