October 2, 2026 · ITK Podcast · Hosts: Alex and Jordan

Alex: Welcome to the I T K Podcast, Information Technology. I'm Alex.

Jordan: And I'm Jordan. It is Friday, October second, twenty twenty-six, and this is our Friday bonus roundup of the stories that broke since Thursday's episode.

Alex: We lead with the weirdest idea in tech right now. Yesterday afternoon, at about two thirty eastern, a SpaceX Falcon Nine lifted off from Vandenberg in California carrying a Google satellite that wants to put AI in orbit.

Jordan: Google's Project Suncatcher. A refrigerator sized satellite carrying four of Google's own AI chips, the Trillium generation Tensor Processing Units. Planet Labs built it together with Google, and the team has already made contact. It is up there, and commissioning has begun. This is a year long experiment to answer one question. Can AI run in space?

Alex: And the ambitions are cosmic. Google says solar panels in orbit can be up to eight times more productive than on Earth, with near constant sunlight. The long term vision is constellations of satellites, linked by lasers, doing AI compute in orbit, powered by the sun. The project lead says this could bypass the energy and political constraints strangling data center expansion on the ground.

Jordan: Right, and this is where we do the ITK thing, because the folks at NPR and Scientific American looked at this and said, hold on, the economics are deeply, stubbornly uncertain. This prototype will run Google's Gemma model in fifteen minute bursts before heat forces a shutdown. Fifteen minutes of AI, then cool down. The solar panels generate about one kilowatt, enough electricity to run a hair dryer.

Alex: The first AI computer in space has the power budget of a hair dryer. Google's own analysis says space based computing only approaches the energy cost of Earth data centers if launch prices fall below two hundred dollars per kilogram by the mid twenty thirties. That is a very big if, riding on a whole lot of physics.

Jordan: And heat. You cannot air condition a satellite. There is no air. Dumping heat into space is brutally hard, and AI chips are basically heat factories. That is the real engineering wall here, and I don't think we can say it loudly enough.

Alex: True, but credit where it is due. Some things can only be tested in space. They are collecting real data on launch vibration, radiation, and the brutal temperature swings of orbital flight. A real experiment beats a white paper every single time. And note the timing. This comes as terrestrial data centers face real backlash on the ground. SpaceX and Blue Origin are planning similar projects. The sky might literally become the data center.

Jordan: Same week, different moonshot from Google. DeepMind introduced SynthID Bio on Wednesday. It is a watermark for AI designed proteins.

Alex: So SynthID is the thing that watermarks AI generated text and images, and now they have adapted it to biology. The idea is to weave a faint statistical tell into the amino acid sequence, or into the three dimensional shape of a protein, so later you can detect that it was designed by AI, even after the physical protein is synthesized in a lab.

Jordan: This matters because of a real problem. DNA synthesis companies screen every order against databases of known threats. A novel sequence used to be safely treated as an undiscovered natural organism. Now AI can invent entirely new proteins that have never existed, so that assumption is broken. A watermark gives screeners an automated signal that an order came from a trusted model with safeguards built in, instead of slow manual reviews.

Alex: And the lab validation is real. DeepMind tested watermarked designs on three targets, including the spike protein of the virus that causes COVID, and the watermarked designs matched the unwatermarked ones on binding strength. They still worked. They also fine tuned part of AlphaFold three so the signature is baked directly into the model's weights.

Jordan: Here is the catch, straight from Nature. The watermark can be scrubbed. Someone who wants to erase the made by AI tag can run a watermarked protein through another design tool and generate a new sequence that keeps the function but loses the signal. So this is one layer, not a solution.

Alex: To DeepMind's credit, they say exactly that. It is a proof of concept, and they are publishing the methods paper, open sourcing the code and the lab data, and releasing the model weights. The honest framing is refreshing, and honest framing is what we are here for.

Jordan: Now the money beat. SoftBank completed its final thirty billion dollar investment tranche to OpenAI on Thursday, partly funded through an eleven billion dollar bond sale. Meanwhile OpenAI is in talks for yet another thirty billion dollar round, at a one point four trillion dollar valuation, with an IPO pushed to twenty twenty seven at the earliest.

Alex: And Anthropic is going the opposite direction. Public first. Their IPO is expected in November, and the prospectus suggests they want to spend five hundred eighteen billion dollars on cloud computing and infrastructure. Meanwhile K K R warned this week that roughly six hundred billion dollars in AI related debt could pose serious risks if power grids, data centers, or chips fail to meet expectations. One point four trillion dollars against roughly a forty billion dollar revenue run rate. Is this the start of a trend, or one very good month dressed up as a trend?

Jordan: Also on the hardware front, CoreWeave is now offering NVIDIA Vera Rubin systems on its cloud, with Cognition, the lab behind the Devin software engineer, as the first customer. Early tests show up to four point eight times the token throughput of the previous generation. The memory bottleneck story we covered yesterday just keeps writing itself.

Alex: Which brings us to the part of the show where we pay the bills. A word from our sponsor.

Jordan: Tired of tech that does not work?

Alex: Call the people who have been fixing it for thirty years. at Your Service Techs.

Jordan: That's at with a lowercase at, folks, because we are classy.

Alex: at Your Service Techs. Thirty years of experience with computers, servers, networks, websites, and surveillance. From Grand Rapids, Michigan to Phoenix, Arizona, they keep businesses running and the bad guys out of the camera blind spots.

Jordan: In Grand Rapids, call six one six, five three five, two zero two two. In Phoenix, call six zero two, two three two, six zero six zero.

Alex: That is at Your Service Techs, on the web at phxfix dot com. They fix it so you can forget it.

Jordan: Beautiful. Now the health beat.

Alex: Two stories this week, and they pair beautifully. First, a small study from the University of Texas Health San Antonio. Twenty older adults with obesity, average age about seventy, did an eight week exercise program. Half got a daily drink with branched chain amino acids, the building blocks found in meat, dairy, and beans. The other half got a placebo.

Jordan: After eight weeks, the amino acid group reported a forty five percent drop in fatigue and a twenty nine percent drop in depression symptoms. The placebo group barely improved, and their fatigue actually rose about five percent.

Alex: Here is the claim, the evidence level, and the unknowns. Claim: amino acids plus exercise help older adults feel less tired and happier. Evidence level: tiny. Twenty people, eight weeks, one study. The mechanism is only a hypothesis, that inflammation changes how amino acids are processed in the brain. Unknowns: everything long term. Does it last? Who does it work for? The researchers are already planning larger trials, including one with military veterans.

Jordan: And here is the counterpoint, the one Dave Asprey keeps warning about. Researchers at Queen Mary University of London found a molecular switch, a compound called nine nine one, that directly activates A M P K, the cell's energy sensor, the same pathway that exercise and metformin touch. Longer healthspan, supposedly.

Alex: Until you read the fine print. In their experiments, higher doses of the compound shortened lifespan in some cases. Too much of the switch jammed on was worse than leaving it alone. The lead researcher said, flatly, that he would not want anyone reaching for a supplement on the back of this.

Jordan: That is the dose makes the poison principle. Gary Brecka talks about this constantly, and Asprey's whole philosophy is foundations first, minerals, vitamin D, B vitamins, before you touch anything exotic. One tiny encouraging study, one promising compound with a sharp edge. Both say the same thing. The evidence is early, and anyone selling you certainty is selling you something.

Alex: That is our Friday bonus. Google put AI chips in orbit, DeepMind is watermarking proteins, the AI money keeps flowing, and the dose still makes the poison.

Jordan: Question everything, including us, especially us.

Alex: We will see you Monday with the full weekly episode. Until then, stay curious.
Question everything — including us… especially us.