Kos raised $12M to automate the paperwork behind America's AI data centers
The April financing funds a virtual finance employee for invoices, purchase orders and lender draws; Kos does not construct data centers.
By Ryan Merket · Published
Primary source: X
Why it matters
The AI buildout creates accounting work alongside compute demand. Kos is betting finance operations become a six-figure software category before payment errors and paperwork slow projects.
Kos raised $12 million in inception capital on April 20th to put an AI worker inside the finance departments of data-center developers and other industrial companies, giving Tanuj Thapliyal (@tanujt) a second startup after co-founding the computer-vision company Spot AI.
https://x.com/tanujt/status/2102087429549167058
The financing was co-led by 8VC and XYZ Ventures, with MVP Ventures and Launch Search Partners participating, according to Kos's original funding announcement. The round also brought industry operators onto Kos's cap table, including former Equinix CFO Keith Taylor, STACK Infrastructure CFO Heather Paduck, former Crusoe CFO Matthew DeNezza and Sachin Katti, a Stanford professor and OpenAI infrastructure executive.
Thapliyal returned to the five-month-old round on September 21st with a 49-post X thread declaring that China was winning the AI race and that Kos had raised the money "to build thousands of new AI data centers in America." Kos is not a developer, construction contractor or compute provider. Its product handles the financial paperwork generated by those projects.
That distinction defines the business. Kos reviews invoices against contracts, generates and routes purchase orders, processes lender draw requests and supports accounting work such as bank reconciliations and month-end close. Kos operates through email, Slack, Microsoft Teams, phone calls and video meetings, using its own account, workstation and software licenses inside a customer's organization.
Rather than asking finance employees to learn another dashboard, Kos asks customers to treat the software like a remote colleague. A worker can forward an invoice by email, question Kos over chat or teach it a new process through screen sharing. Kos says its system then repeats that process within the rules established by the customer, while human managers supervise its work.
Accounting for the AI buildout
Kos is targeting a cost center created by the scale and speed of infrastructure construction. Data-center finance teams must reconcile lengthy contractor invoices, equipment orders and change requests against contracts that can run to hundreds of pages. Errors can delay payments or move more cash than a project actually owes.
The Information reported around Kos's April launch that the company had signed multiple customer contracts. Thapliyal said contracts generally carried six-figure prices, roughly comparable to hiring a specialized finance employee. Kos has described its customers only by sector, saying the software is in use at infrastructure and defense companies.
The customer concentration also explains the investor list. 8VC and XYZ have placed repeated bets on defense, industrial and infrastructure startups, while Kos's individual shareholders have managed the finance functions of companies building and operating data centers. Kos is using those investors as both distribution and product expertise: the people financing the software have held the jobs it is supposed to automate.
Kos's broader claim is that a shortage of specialized accountants could become a constraint on deploying infrastructure capital. That thesis does not depend on Thapliyal's assertion that China is winning an AI race. It depends on Kos processing financial documents accurately enough for CFOs to delegate consequential work to software.
Thapliyal's second infrastructure startup
Thapliyal previously worked at Cisco's Meraki before studying at Stanford, where he conducted research under Katti. He later co-founded Spot AI in 2018 with Rish Gupta and Sud Bhatija, building software that connects AI analysis to security-camera systems. Spot AI currently says it serves 1,200 US customers and has raised $110 million.
At Kos, Thapliyal reunited with Manikanta Kotaru, another former Stanford researcher advised by Katti. Kotaru earned an electrical engineering doctorate at Stanford after studying at IIT Bombay and later worked at Microsoft Research and on Azure products. He serves as Kos's co-founder and chief technology officer.
The founders are applying that systems background to an accounting product whose hardest requirement is reliability. A model can summarize a contract convincingly and still miss the line item that causes an overpayment. Kos says it addresses that risk by grounding work in each customer's defined procedures, maintaining audit logs and assigning a human manager to the AI worker.
The $12 million gives Kos time to prove that architecture across customers with different contracts, enterprise software and approval chains. Thapliyal's latest thread packages the company as part of America's strategic AI buildout. Kos's nearer-term job is narrower and measurable: read the invoices, follow the process and keep the money moving.