Ailytics, a Singapore vision-AI company, announced a $7.4 million Series A on 26 October, led by Tin Men Capital. The company calls the round oversubscribed. The technical claim underneath it is the part worth reading twice, because it is an argument against the way most of this market has been funded.
What Was Announced
Tin Men Capital, a Singapore enterprise and frontier-tech firm, led the round — the same firm that led what Ailytics says was a $2.7 million pre-Series A in April 2024. Joining it: SEEDS, an arm of SG Growth Capital, the investment platform of Singapore’s Economic Development Board and Enterprise Singapore; Safie, which the release describes as Japan’s largest cloud-camera company; one of Japan’s largest construction companies, which the announcement does not name; family offices; and returning investors.
No valuation accompanies the round. That is not an inference from the release’s silence — the funding trackers that list the company record the post-money valuation as undisclosed, and the individual investors’ contributions are not broken out either. Total raised to date is a figure this publication is going to decline to print. The company’s own announcement states no cumulative total, and the trackers that carry one disagree with each other — DealStreetAsia put the pre-round equity total at $6.36 million — in ways that suggest they are measuring different things at different dates rather than converging on a number.
“We’ve already proven that advanced computer vision can run on a decade-old camera,” said Prateek Manocha, co-founder and chief technology officer.
Ailytics traces its origins to a 2019 joint research project between the National University of Singapore and the Housing & Development Board, and started out on Singapore public-housing projects.
The Bet: Software on Hardware Somebody Else Already Paid For
Manocha’s sentence is a positioning statement disguised as an engineering note. Nearly every jobsite-vision company this publication has covered sells, ships or specifies hardware. Sensera Systems raised $27 million on the strength of self-contained, solar-powered, wirelessly connected cameras already installed on thousands of sites — the software layer being the bet, but the cameras being the moat. EarthCam’s October announcement ran seven new AI models across its own camera estate. OpenSpace built a business on a worker walking a site with a 360 camera on a hard hat.
Ailytics’ claim is that it needs none of that, because a construction site already monitored by CCTV — installed for theft and access control long before anyone thought about AI — is already instrumented. How widely that holds is the load-bearing assumption, and this publication found no public measure of CCTV prevalence on job sites to test it against. The capital cost has been paid, the network is in, the mounts are on the poles. What is mostly missing is anyone watching the footage.
The commercial consequence is a different sales motion, and that is where the model will be tested. Selling software against cameras a client already owns means no hardware margin, no installation revenue and no physical lock-in — but also no procurement cycle for capital equipment, no site-mobilisation schedule and no deployment crew. It is a thinner, faster business if the computer vision really does degrade gracefully on old, badly placed, low-resolution cameras pointed where a security contractor decided to point them in 2017. That is a materially harder technical problem than running fresh 4K from a position you chose, and the claim that it is solved is Ailytics’ own.
Tin Men’s stated read is about the economics rather than the vision stack. “Ailytics has also reached 15 markets at very robust unit economics and client churn below 1 per cent,” said Jeremy Tan, co-founder and managing partner at the firm.
The Traction Numbers, As the Company Reports Them
The release carries a dense set of figures, all of them the company’s own: more than 400 deployments across 15 countries, up from 70 projects in four countries at the time of the pre-Series A; more than 5.7 million hazards detected; more than 500 million square metres of active worksites monitored. Customers, it says, report a 73 per cent reduction in total recordable incident rate and 84 per cent fewer safety non-compliances. Those last two are customer-reported outcomes relayed by the vendor, which is a weaker evidentiary chain than it reads as, and the release does not say over what period, across how many customers, or against what baseline.
The named deployments are specific enough to check against, which is more than many announcements offer. On the construction side: the contractors Woh Hup, Obayashi and Soilbuild, HDB projects, and the Changi Airport Terminal 5 build, where the release says more than 1,000 cameras are involved. Beyond construction: NatSteel, the PaxOcean shipyard in Indonesia, DHL, and Rome’s Leonardo da Vinci International Airport. Countries named include the United States, Australia, the Netherlands, Hong Kong, Japan, the UAE and Saudi Arabia. The Safie relationship predates the round — the two signed a memorandum of understanding in April 2026 — which makes an investor on the cap table also a distribution channel, and is the most structurally interesting thing in the investor list.
Why Safety Vision Keeps Attracting Money
The regulatory and research backdrop has been moving toward this for a year. NIOSH’s October bulletin on construction robotics made a narrow, awkward argument that factory safety engineering assumes a workspace that holds still and a job site does not — a problem which, for autonomous machines, is unsolved. Passive observation sidesteps the whole difficulty. A camera watching for a worker inside an exclusion zone does not need to navigate, grip or hold still; it needs to classify. FORT Robotics raised $20 million for the interlock layer that stops a machine; Ailytics is selling the layer that notices a person.
The use of funds is pointed at expansion rather than new capability: local teams in the United States, Japan, Australia and the Gulf, mid-sized American industrial customers, the company’s own domain-specific models, and keeping inference costs low as volume grows. That last item is an admission worth noting. A business built on running continuous inference over somebody else’s camera estate has its gross margin written in GPU time, and the cost curve of watching 500 million square metres is the number that decides whether the thin, fast version of this business works.
What to Watch
Whether the decade-old-camera claim survives contact with the American mid-market. Singapore public housing and Changi T5 are large, well-run, heavily regulated projects where the CCTV estate is professionally specified. A mid-sized US contractor’s site is a different population of cameras entirely, and generalisation across it is the whole thesis.
And whether the unnamed Japanese contractor stays unnamed. A large construction company on the cap table alongside Safie’s camera infrastructure looks less like a financial position than the outline of a Japanese go-to-market — the kind of arrangement Shimizu’s investment in Noble Machines made explicit in October. The release does not characterise it that way, and the company has not said it intends to.
