Modern rockets, satellites, defense systems, and autonomous vehicles generate more sensor data than their operators can parse. That gap is a problem — and Sift thinks it is a $42 million problem.
Sift, an El Segundo, California startup building the data infrastructure layer between raw hardware telemetry and AI control systems, announced a $42 million Series B round on March 25. StepStone Group led the round, with participation from GV, Riot Ventures, Fika Ventures, and CIV, according to SpaceNews. The company has now raised roughly $67 million total, including a $17.5 million Series A led by GV in June 2024 and a seed round in 2023. Gollapudi told TechCrunch that the Series B closed in 2025 at a $274 million post-money valuation.
The company was founded in 2022 by Karthik Gollapudi and Austin Spiegel, both ex-SpaceX engineers. Gollapudi led flight software for Dragon; Spiegel worked on the Starlink Constellation Tools team. A third co-founder, Neel Kujur, serves as chief technology officer.
Modern rockets, satellites, defense systems, and autonomous vehicles are computers wrapped in steel, generating millions of data points per second from hundreds of sensors, but AI cannot interpret any of it without structure, Sift said in its announcement. The pitch: AI-controlled hardware needs structured sensor data the same way software engineering needed observability tools in the 2000s. Without a data plane, AI systems that are supposed to flag anomalies or guide manufacturing decisions are working blind.
Some vehicles Sift works with have more than 1.5 million sensors streaming data concurrently, across multiple formats and time scales, according to TechCrunch. The company ingests that stream, time-aligns it, runs automated anomaly detection, and outputs structured, queryable data that both engineers and AI systems can work with. Sift also integrates with Palantir Foundry and its AIP platform — relevant because Palantir has been positioning hard in the aerospace and defense data stack.
Jeff Dexter, vice president of software at Astranis, a Sift customer, said his company might do 10 million automated software tests in a day. The storage costs become real fast. Inevitably it gets to a point where it is costing us millions of dollars per month just to store data, Dexter told TechCrunch. With technology like Sift, I do not worry about how much data is there.
Sift is not the only company that has noticed the problem. Boeing's CST-100 Starliner flew the wrong orbit in 2019 because of a clock sync error — $600 million in write-offs and a crewed flight program set back years. The ispace lunar lander crashed in 2023 due to a software miscalculation during descent. Both failures involved telemetry interpretation problems at scale. Gollapudi and Spiegel told TechCrunch they surveyed 17 companies before building, and the breadth of the problem only increased their conviction.
Sift's customer list spans launch providers, satellite operators, and defense hardware shops: Astranis, K2 Space, Astrolab, United Launch Alliance, True Anomaly, Mach Industries, and Parallel Systems. The common thread is that every one of these programs is trying to run AI-assisted decision-making on sensor data streams that were never designed to be machine-readable.
Software observability matured over two decades, Gollapudi said. Hardware companies still rely on spreadsheets and tribal knowledge. The implied sequel: hardware observability is overdue, and the arrival of AI control systems makes it urgent rather than merely annoying.
StepStone is not a typical space investor — the firm focuses on private markets and infrastructure debt, not orbit and payload. That a private equity-adjacent firm led a Series B for a 62-person telemetry startup is a signal that the problem has crossed a threshold: hardware companies are now willing to pay for professional tooling rather than patching together Python scripts and hoping the anomaly that blows up the vehicle also blows up the log file.
The staffing trajectory tells a similar story. Sift had 12 employees in 2023 when its seed round was announced. The Series B announcement did not specify current headcount, but PitchBook lists 62 total employees. That is not a large team for the problem they are trying to solve, which means the next hires will be telling. Whether Sift can build a sales and customer success motion at scale — not just engineering depth — is the open question the Series B was designed to fund an answer to.
The round closed in 2025 and was announced in March 2026. That is a long lag for a company that presumably had term sheets and press releases ready to go. Possible explanations range from regulatory review to a customer or partner wanting the announcement timed to a specific event. Giskard should push on this before publish.
What to watch next: whether Sift can retain and expand within its existing customer base, since no retention or net revenue retention figures were disclosed. The observability market for software is crowded — Datadog alone is a public company worth tens of billions. The hardware observability market does not have that name recognition yet. That is either the opportunity or the warning.
