The pitch for Constellation Space is basically an infrastructure argument: ground-based network operations centers were designed for fleets of dozens, not tens of thousands, and the math stops working somewhere in between. ConstellationOS, the flagship product from the Seattle-based YC W26 startup, is trying to be the orchestration layer that fills the gap, with a stated ability to predict link failures three to five minutes ahead at better than 90% accuracy and reroute traffic in under two seconds [PERPLEXITY SONAR PRO BRIEF] [welcome.ai].
The company closed a $6.5 million seed in March 2026 led by SGH Capital, with NVIDIA, OpenAI, Standard Capital, Samsung Next, Gladebrook Capital, Fellows Fund, Toloka.vc, and Founders Future participating [StartupHub.ai / Claude's Corner, August 2026]. A co-founder's LinkedIn post from January 2026 put the post-money valuation at roughly $10M (estimated) at the time of founding [LinkedIn, January 2026]. An undisclosed grant followed in April 2026 [funding-tracker site].
The wedge: in-orbit, not ground-based
The founding argument, per the company's own framing, is that reactive, ground-based orchestration does not scale to mega-constellations [PERPLEXITY SONAR PRO BRIEF]. ConstellationOS ingests telemetry from satellites, ground stations, and weather systems, then runs physics-informed ML models to forecast signal-to-noise ratio, traffic, weather, jamming, and conjunction risk hours before packet loss [PERPLEXITY SONAR PRO BRIEF].
That is a different posture than what most incumbents in the space-situational-awareness category sell. LeoLabs, Slingshot Aerospace, and COMSPOC largely provide ground-based tracking and analytics; Kayhan Space and Neuraspace focus on conjunction assessment. ConstellationOS is pitched as an operating layer that sits on the fleet and makes decisions in-loop, with zero-intervention handoffs, predictive thermal load balancing, and sub-second optical routing [PERPLEXITY SONAR PRO BRIEF].
Why the timing tracks
The addressable fleet is the story. The platform is designed for a satellite population expected to grow to 70,000 by 2030 [PERPLEXITY SONAR PRO BRIEF]. At that scale, human operators supervising each spacecraft is not a staffing problem, it's a category error. Every additional constellation added to LEO tightens conjunction risk and RF interference budgets for everyone else in the neighborhood.
Distribution reflects that reality. ConstellationOS is listed on AWS Marketplace [AWS Marketplace: ConstellationOS], which shortens procurement for operators who already have AWS enterprise agreements, and the company offers a 30-day shadow pilot so prospects can run it in parallel with their existing NOC before cutting over [constellation.space]. Shadow-mode deployment is the right go-to-market shape for anything that claims autonomous control of a physical asset worth tens of millions of dollars.
Four founders, three flight programs
The team is small (four employees per YC as of 2026) and the resumes are pointed at the problem [Y Combinator, updated 2026]. Kamran Majid is CEO and was previously a software engineer at SpaceX and NASA [LinkedIn, retrieved 2026]. Raaid Kabir came from embedded flight software at Blue Origin [LinkedIn, retrieved 2026]. Laith Altarabishi has an MS in ECE from UT Austin with a focus on ML and computer systems [Crunchbase, retrieved 2026]. Omeed Tehrani, also UT Austin (MS/BS CS), worked on deep reinforcement learning and frontier AI infrastructure at Capital One [omeedtehrani.com].
| Founder | Role | Prior |
|---|---|---|
| Kamran Majid | CEO | SpaceX, NASA |
| Raaid Kabir | Co-founder | Blue Origin (embedded flight SW) |
| Laith Altarabishi | Co-founder | UT Austin MS ECE, ML + systems |
| Omeed Tehrani | Co-founder | Capital One AI infra, RobIN Lab |
The cap table tells its own story. Getting NVIDIA and OpenAI into a $6.5M seed alongside SGH Capital and Samsung Next is not a normal outcome for a four-person team [StartupHub.ai / Claude's Corner, August 2026]. It reads as strategic interest in an ML workload that runs at the edge of a physical network at rates the brief cites as 100,000 telemetry messages per second [PERPLEXITY SONAR PRO BRIEF].
Funding to date
Seed (Mar 2026) | 6.5 | M USD
Only one round has a disclosed dollar figure. The April 2026 grant amount was not disclosed [funding-tracker site].
What has to be true for this to work
The technology claims are strong, and none of them will matter without named operator deployments in production. A few honest concerns:
- Reference customers are not yet public. The company has not publicly named a commercial satellite operator running ConstellationOS in production, and mega-constellation operators (Starlink, Kuiper, OneWeb, the Chinese state programs) typically build in-house rather than buy. The 30-day shadow pilot [constellation.space] is a smart wedge; conversion is the number to watch.
- The competitive set is well-funded. Spire Global, LeoLabs, Slingshot, and Iceye all have established government and commercial relationships and can extend up-stack. Constellation's answer has to be that in-orbit, ML-native orchestration is a fundamentally different product than ground-based analytics, not a feature of one.
- Grant funding suggests government interest. The April 2026 grant is undisclosed [funding-tracker site], but defense and civil-space buyers move on procurement cycles measured in years, not the quarters a seed-stage runway allows.
A technical read on the claims
The headline metrics are internally consistent. Predicting link failure three to five minutes ahead at >90% accuracy is a supervised learning problem with abundant labeled data if you have fleet telemetry: SNR degrades on measurable curves before packet loss. Sub-two-second rerouting with zero data loss requires the routing decision to live on-satellite or at the ground-station edge, not in a cloud round-trip, which is consistent with the "in-orbit" framing [PERPLEXITY SONAR PRO BRIEF]. Processing 100,000 telemetry messages per second is a stream-processing throughput number, not a modeling one, and is achievable on modest hardware with the right ingestion pipeline.
Where the story gets harder is not the per-satellite claim but the fleet claim. Autonomous rerouting across a shared RF environment means the OS is making decisions whose second-order effects show up on other operators' spectrum. At 70,000 satellites, coordination failure modes look less like packet loss and more like cascading interference and conjunction events that were technically legal for each individual actor. The company that ships the orchestration layer for that world inherits a governance problem that no seed round is priced for, and the operator community, insurers, and regulators will all want a say in how the model behaves when two ConstellationOS instances disagree about who yields. That is the scaling risk worth watching, not the accuracy number.
Sources
- [Y Combinator, updated 2026] Constellation Space company page | https://www.ycombinator.com/companies/constellation-space
- [StartupHub.ai / Claude's Corner, August 2026] Constellation Space Review: The AI OS for Satellite Fleets
- [LinkedIn, January 2026] Omeed Tehrani co-founder announcement | https://www.linkedin.com/posts/atomsnotbits_daily-hard-tech-headlines-constellation-activity-7448816355107168256-WGeC
- [LinkedIn, retrieved 2026] Kamran Majid profile | https://www.linkedin.com/in/kamran-majid/
- [Crunchbase, retrieved 2026] Constellation Space profile | https://www.crunchbase.com/organization/constellation-space
- [omeedtehrani.com] Omeed Tehrani personal site
- [constellation.space] Company site and shadow pilot | https://constellation.space/company
- [AWS Marketplace: ConstellationOS] Product listing
- [welcome.ai] ConstellationOS product page
- [funding-tracker site] Constellation Space funding record
- [PERPLEXITY SONAR PRO BRIEF] Web-grounded research synthesis