Andreessen Horowitz puts $38 million behind doxx.net's server-free networks for AI agents

Doxx.net's closed beta began in December 2025 ^[17], roughly ten months before the October 1 announcement ^[3][13]. More than 38 million potential threats over that stretch comes to about 3.8 million a month, or something like 125,000 a day ^[14]. The count is the company's own, and it covers potential threats, not confirmed attacks ^[18]. A blocked-threat tally rises with the number of users and with how broadly the filter defines a threat. A large figure can come from a small, busy test group as easily as from a large population.

Who generated it matters for the agent pitch. The platform is built for individuals, businesses and AI agents alike ^[8], so the tally covers all three. Doxx.net argues that intermediaries in a network become new weak points once software acts without a person approving each step ^[16]. Agentic Defined Networking, its product ^[5], is meant to cut that exposure by taking centralized servers out of the path ^[16].

Taking the servers out also changes where the money goes. With no servers in the design ^[15], the round is not buying a hosting fleet. I'd expect most of the $38 million ^[1] to go to engineers and to pulling users into the open beta. The company did not disclose how it plans to allocate the capital ^[12].

There are three ways the round could be priced. One is on the founder. Barrett Lyon has founded at least three companies before this one ^[7], and a lead check from Andreessen Horowitz ^[1] into a product only now opening to the public ^[6] leans on that record. Another is a privacy thesis. Joel De La Garza, the a16z partner, said: "The internet was never designed for privacy." ^[10] Nico Berardi of Animo Ventures said the platform has the potential to reshape how users interact online ^[11]. The third possibility is that the paying customer turns out to be people and businesses who want private networks, with agent traffic arriving later ^[8].

I think the founder reading fits the evidence best. The only traction on record is a self-reported filter count from a closed test ^[18]. These reports also cover a single deal, too small a sample to call a funding trend in agent security ^[1]. The counter-case is that the lead investor saw closed-beta data beyond the headline count, and that ten months of blocking at that volume is a fair early signal for a security product ^[13][14]. The view is wrong if open-beta figures show agents generating a large share of traffic, or businesses paying for their own DNS roots and address space ^[15] within a year.