DoiT buys Attribute, and AI cost attribution moves from billing tags to the kernel

DoiT has acquired Attribute, the AI cost-attribution company founded by Izhak Zimmermann and Liad Tropp, and folded its technology into a product meant to show which customer, feature or agent produced each piece of an AI bill ^[1]. What is worth noting is not the price, which nobody involved has confirmed, but the measurement point: Attribute watches consumption as it happens in the kernel instead of reconciling billing tags afterwards ^[5][6].

The sequence is unusual. DoiT launched the integrated Attribute product on July 7 ^[2]. CTech publicly reported the acquisition on August 17 and estimated the price at roughly $65 million ^[3]. The exact closing date is not established, DoiT and Attribute have not confirmed the figure, and CRN reported that financial terms were undisclosed ^[4]. On the reported dates, the product was in market about six weeks before the deal was public ^[1].

Mechanically, Attribute runs a lightweight eBPF sensor that monitors runtime activity at the Linux kernel level ^[5]. DoiT says the sensor maps GPU, CPU, memory, network, I/O and model API consumption to the process, container, pod or request that caused it ^[6], then joins those observations to provider cost data covering OpenAI, Anthropic, Google Gemini and AWS Bedrock, automatically splitting cached, reasoning, input and output tokens ^[7]. The output assigns cost to a customer, feature, workload or AI agent rather than leaving it pooled in a shared cloud account ^[8]. DoiT also says installation takes about 15 minutes with no SDK, tagging policy or application code changes ^[9]. Those deployment and coverage claims come from the vendor and have not been independently benchmarked ^[10].

The thesis is stated plainly by Zimmermann in DoiT's July 7 announcement: "You can't tag your way to the truth inside a shared GPU or a single Bedrock account" ^[11]. The founders built the company on the view that tagging systems used to divide conventional cloud bills cannot keep pace with shared GPUs, LLM gateways and agentic workloads ^[12]. The practical version of that problem: an operator may know its total OpenAI or Bedrock bill and still lack the data to say whether an AI feature carries a healthy gross margin ^[13], and shared credentials, gateways and clusters can obscure whether an expensive model request came from a paying customer, an internal experiment or an autonomous agent stuck in a loop ^[14]. DoiT CEO Vadim Solovey argued in a July 7 post that AI infrastructure was built for speed and scale, leaving financial attribution to be reconstructed after usage ^[15].

On the money, CTech reported Attribute raised about $13.5 million from Mensch Capital, SCapital, HarelTech and IBI ^[16], though Startup Nation Central lists the funding as undisclosed ^[17]. If both the raise and the $65 million estimate hold, the exit is roughly 4.8 times capital raised ^[2]. Neither the consideration structure nor Attribute's revenue was disclosed, so any return calculation is guesswork ^[18]. Startup Nation Central dates the founding to January 2023 ^[19], which does not reconcile with the description of a business founded roughly three and a half years before the deal ^[20].

Competitors are approaching the same allocation problem from the billing side: Finout markets Virtual Tags for allocating AI spend to teams, product lines, features and customer segments, and for cost per inference, customer or feature ^[21], and Vantage introduced an LLM token-allocation preview in February ^[22]. Zimmermann is now general manager of Attribute at DoiT, with Tropp listed as co-founder and CTO ^[23].