The cry of “atoms, not bits!” — a phrase capturing Silicon Valley’s growing obsession with physical manufacturing over digital products — reached a fever pitch last week with word that Jeff Bezos is putting together a $100 billion fund to roll up and automate factories.
But automating factories isn’t purely a hardware problem. It increasingly depends on sophisticated software and AI tools, and that shift is reshaping the companies building the infrastructure of the physical manufacturing world.
Karthik Gollapudi, the CEO of Sift Stack, an El Segundo, California, company whose tools support the design and manufacturing of complex machines like spacecraft and cars, is feeling the ground shift underfoot. He says these changes have reshaped his company’s focus in the last six months.
Gollapudi and his co-founder, CTO Austin Spiegel, started the company in 2022 after working on software tools at SpaceX that managed the huge amount of telemetry data — real-time performance information streamed from sensors on physical components — during testing, manufacturing, and launch.
Most companies building advanced machines use off-the-shelf database tools or cook up their own Python scripts, but Sift saw the opportunity to provide companies with a best-in-class tool. Customers range from United Launch Alliance, a major US rocket builder, and other defense contractors, to robotics and power grid management startups.
However, Gollapudi says that the arrival of AI tools for data analysis forced a change at his business. The kinds of customized workflows that once stood out as the company’s signature offering have become table stakes in a world of AI and deep learning models. But the company’s ability to manage data infrastructure had suddenly become more valuable.
“Our long-term vision of how we saw this playing out over five years is actually being played out this year,” Gollapudi told TechCrunch.
Techcrunch event
San Francisco, CA
|
October 13-15, 2026
That means managing the intense data flow from today’s software-intensive machines. Some vehicles the company works with have more than 1.5 million sensors streaming data concurrently, across multiple formats and time scales.
Organizing and storing that data for AI applications is the company’s goal—”a lot of the value is in exposing that to be machine readable,” Gollapudi said. If AI agents are going to make decisions about manufacturing or analyze test data to flag potential problems, Sift’s goal is to make that data available to them.
Jeff Dexter, the VP of software at Astranis, a satellite company that uses Sift to manage test, manufacturing, and operations, said that good data infrastructure matters for companies like his that might do 10 million automated software tests in a day.
“Inevitably, it gets to a point where it’s costing us millions of dollars per month just to store data,” Dexter said. “It’s really like, is this a million dollars well spent? With technology like Sift, I don’t worry about how much data is there.”
Gollapudi told TechCrunch that Sift raised a $42 million Series B in 2025 at $274 million post-money valuation, led by StepStone with participation from GV (Google’s venture arm), Riot Ventures, Fika Ventures, and CIV.
