Tera AI
San Francisco, US Β· 20 known investors
Tera AI develops foundation models for autonomous systems that enable real-time navigation and mapping on resource-constrained hardware using monocular cameras, applicable across diverse platforms.
Also known as Tera Β· TeraAI
Founders & leadership
Investors Β· 20
Also in the syndicate Β· 16
Funding
SEC filings, press & company announcements$7.8M disclosed across 1 of 3 rounds Β· 2025
- $7.8MSeedMar 2025 Β· 2 sources
Felicis (lead), Inovia Capital (lead), Caltech + Wilson Hill
Source β
Source: company announcements and press reports β follow each round's link for the claim.
Company profile
researched Aug 2026Tera AI is a San Francisco-based robotics software company developing a platform-agnostic, software-only spatial reasoning system that lets robots and vehicles localize, map and navigate using cameras already installed on the platform, without LiDAR, high-precision GPS receivers or added sensors. The company describes its offering as a unified spatial AI stack covering four domains: air (quadcopter, helicopter, fixed-wing), onroad (rideshare, delivery), offroad (small and large ground vehicles) and sea (small and large surface vessels), with capabilities including GPS-denied camera-based navigation, collaborative autonomy, 3D mapping and semantic understanding, and camera-based range estimation.
Technically, Tera AI positions its models as foundation models for spatial reasoning that support zero-shot navigation in previously unseen environments, applied at inference time in a manner the founder compares to a large language model. The system builds persistent internal map representations that a robot carries across missions rather than one-off scans, and is described as able to read maps from camera imagery from arbitrary viewpoints. The company states it has demonstrated GPS-denied navigation on long-wave infrared imagery and low-altitude flight over feature-sparse desert terrain in real time on constrained compute with a monocular, narrow field-of-view camera.
Its published research includes STRIDE, a spatio-temporal road image dataset of 131,000 panoramic StreetView images permuted into 3.6 million observation-state-action sequences covering Silicon Valley, and TARDIS, a 1-billion-parameter autoregressive multimodal world model built on STRIDE that jointly predicts position, motion and next observations. Reported TARDIS results include a 41% FID improvement over Chameleon-7B on controllable photorealistic synthesis, georeferencing within 10 meters from a single street-view image the majority of the time, and 77% of self-generated actions remaining within road lanes. The dataset, code and checkpoints were released on Hugging Face.
Founding story
Tera AI was founded in 2023 by Tony Zhang, who serves as founder and CEO. Zhang previously led machine learning efforts at Google X on geospatial AI, working on sensing, prediction and optimal control models and collaborating with academic labs, and holds a Caltech PhD in neural computation under computer-vision researcher Pietro Perona, focused on sequential decision-making in brains and machines. The company began with geospatial foundation model research, including early experimentation in areas such as insurance, before concentrating on general-purpose robotic navigation delivered in software.
Business model
Tera AI sells software rather than hardware, delivered as an over-the-air software update that runs on a robot's existing camera and GPU. The founder characterizes the product as operating-system-like software licensed to mobile robotic platforms, with an eventual vision of downloadable capability modules analogous to an app store. Integration is offered as a supported process, with the company assisting partners through deployment.
Sources describe a software product sold to robotics manufacturers and platform makers via over-the-air updates; specific pricing or contract structures are not disclosed in the available sources.
Traction
As of March 2025 the company was testing its product with multiple U.S. robotics companies and planned to deploy its initial solution on embedded devices that year. Later reported validation includes real-time navigation during sub-100-ft-AGL desert flights with a narrow-FOV camera, GPS-denied navigation on LWIR imagery, and reduction of new-platform integration to under 40 engineer-hours. The company cites selection to the CB Insights AI 100 for 2026. One aggregator lists an employee range of 201-500, which is not corroborated by other sources.
Latest developments
The most recent dated material is a 30 July 2026 company post on GPS-denied visual navigation using long-wave infrared imagery, preceded by a 16 June 2026 post on reducing platform integration to under 40 engineer-hours and an 2 April 2026 post on ultra-low-altitude flights in the Ocotillo desert. The website (copyright 2026) advertises open ML systems, 3D computer vision and forward-deployed engineering roles and cites selection to the CB Insights AI 100 for 2026.
βΈFull profile β market position, technology, go-to-market, geography, history, risks & controversies
Market position
Tera AI positions itself as offering the first unified spatial AI stack spanning air, road, offroad and sea platforms, contrasting its approach with hardware-centric autonomy stacks that depend on expensive sensors (the company cites examples such as $50,000 localization sensors, $100,000 LiDAR systems, $10,000 high-precision GPS receivers and $30,000 IMUs) and with per-platform navigation software that does not generalize. Third-party recognition cited includes selection to the CB Insights AI 100 for 2026.
Hardware agnosticism and a pure-software delivery model are the stated core differentiators: the system is designed to run on cameras and compute already present on a platform, generalize zero-shot to new environments without per-site retuning, and operate without GPS. The company also claims cross-domain generality across vehicle types and sensing modalities (including thermal), and a materially compressed integration timeline (under 40 engineer-hours versus an industry-typical 3-6 months).
Technology
Camera-based (often monocular, narrow field-of-view) spatial foundation models for mapping and navigation that run on-device on limited compute. The stack supports GPS-denied operation, works on visible and long-wave infrared imagery, builds persistent map representations, and performs cross-view map matching without LiDAR or custom-collected maps. Research artifacts include the STRIDE dataset (131k panoramic images, 3.6M sequences) and TARDIS, a 1B-parameter early-fusion multimodal autoregressive transformer world model that unifies perception, mapping and planning into a single sequential prediction loop. Deployment uses containerized delivery, platform-based calibration and automated bench validation.
Go-to-market
The company markets partnership engagements directly from its website ('Partner with us'), emphasizing demo software integrable in a few days with company-supported onboarding. It works with camera and vehicle OEMs to ensure compatibility across manufacturers, and as of March 2025 was testing its product with several U.S. robotics industry players. It also uses published research, technical blog posts and validation results as credibility signals.
Primarily robotics manufacturers that already have customers but face difficulty extending their systems to new autonomy platforms, situations and environments. Application areas cited include robotic manipulation, mobile robotics, automated driving, uncrewed aerial vehicles, ground vehicles and surface vessels, with adjacent mentions of digital twins and urban planning, and environmental monitoring.
Geography
Headquartered in San Francisco, California, with open roles listed in San Francisco (on-site) and San Diego, California (hybrid). Its STRIDE dataset covers Silicon Valley, and field testing referenced includes the Ocotillo desert. Stated clients and test partners are primarily U.S.-based.
History
Founded in 2023, the company worked in stealth on spatial reasoning AI for roughly a year before emerging publicly on 19 March 2025 with a $7.8 million seed round. In July 2025 it published its first research paper and released the STRIDE dataset and TARDIS model checkpoints on Hugging Face. Company blog posts through 2026 describe an expanded cross-domain product framing (air, onroad, offroad, sea), a compressed integration workflow, ultra-low-altitude desert flight trials, and GPS-denied navigation on thermal imagery. Its site references two years of R&D behind models runnable on limited compute.
Risks & controversies
Sources contain conflicting figures for the seed round ($7.8M per TechCrunch and VCNewsDaily versus $8.0M with a much longer investor list and a 1 March 2025 date per the Startup Intros aggregator), and the aggregator's 201-500 employee range is uncorroborated and appears inconsistent with an early-stage seed company. Many performance claims (integration time, desert flights, LWIR navigation, CB Insights selection) originate from the company's own website and blog rather than independent verification. No revenue, customer names or commercial contracts are disclosed in the available sources.
Compiled by commissioned research from 8 cited public sources β announcements, filings, and press listed under research sources below.
Key figures
latest reportedCompany-reported or press-reported figures, each dated to when it was claimed β not independently audited.
Competitors Β· 5
by search overlap
Companies competing with Tera AI for the same Google search keywords, organic and paid, via search-intersection analysis.
Timeline Β· 7
launches, deals, and filingsTera AI published results showing its GPS-denied visual navigation stack running on LWIR (thermal) imagery without a separate, modality-specific navigation stack.
Tera AI reported cutting the time to stand up its autonomous navigation software on a new platform from an industry-typical 3-6 months to under 40 engineer-hours using containerized delivery, platform-based calibration and automated bench validation.
Tera AI reported software-only visual navigation during flights below 100 ft AGL over feature-sparse desert terrain, running in real time on a limited compute node with a narrow field-of-view camera.
Company blog post arguing that existing navigation stacks are tailored per platform and rely on domain-specific assumptions that do not generalize.
Tera AI states it was named one of 100 private AI companies in CB Insights' AI 100 list for 2026.
Tera AI released its first research paper, 'TARDIS STRIDE: A Spatio-Temporal Road Image Dataset and World Model for Autonomy', along with the dataset, code and pretrained TARDIS checkpoints on Hugging Face.
Tera AI came out of stealth announcing a $7.8 million seed round co-led by Felicis and Inovia Capital, with participation from Caltech, Wilson Hill and Naval Ravikant. Funds were earmarked for deploying its first solution on embedded devices and expanding the technical team.
$7.8M source β
Dated company events from announcements, filings, and press; legal rows summarize public dockets and regulator releases.
In the news
βΈResearch sources Β· 8
primary sources listed
- Tera AItera-ai.com Β· web
8 public sources were cited for this profile; the first-party ones are listed here.
Frequently asked questions
- What does Tera AI do?
- Tera AI builds software-only, camera-based spatial AI for GPS-denied mapping and navigation across air, ground and sea platforms.
- Who are Tera AI's investors?
- Tera AI's investors include Felicis Ventures, Inovia Capital, Zypsy Capital, Wilson Hill.
- How much funding has Tera AI raised?
- Tera AI has disclosed $7.8M raised across 1 of its 3 known rounds.
- Where is Tera AI headquartered?
- Tera AI is headquartered in San Francisco, US.