Rafael Quintanilla raises $38 million for Arrakis' industrial AI agents
The former Accel investor assembled a Palantir-heavy team and reached a reported $140 million valuation within seven months.
By Ryan Merket · Published
Why it matters
Arrakis is testing whether forward-deployed engineers, outcome-based pricing and model choice can turn industrial AI pilots into a repeatable software business.
Rafael Quintanilla, a former Accel investor, brought Arrakis out of stealth on July 22 with $38 million in funding and a plan to install AI agents inside the operational systems of industrial businesses. Tech.eu reported that the total includes a $30 million Series A involving Blossom Capital and Accel, after Accel backed Arrakis' seed round in March.
Quintanilla founded Arrakis with Haroun Beltaifa, Romain Fouilland and Mikhail Galkov. Beltaifa and Fouilland previously worked at Palantir, while Galkov was an engineer at Delivery Hero. Quintanilla crossed the table from venture capital after developing Accel's investment thesis around defense and industrial resilience, work that took him through the US, Europe and the Middle East.
That research convinced Quintanilla that AI capital and product development were concentrating on office work while industrial operators still needed AI tied into core operations. Arrakis is built around that physical-economy thesis: applying agents to industrial systems where data, workflows and human approvals are already embedded.
Arrakis is Quintanilla's attempt to turn that venture thesis into an operating business. Arrakis targets aerospace, energy, logistics, manufacturing, construction and telecommunications, where an AI system has to work across procurement records, operational data and established workflows rather than live in a standalone chat window.
A fast financing cycle
Arrakis' financing moved unusually quickly. A US securities filing records $7.33 million sold to 28 investors in an offering whose first sale occurred on March 20. A second Form D reproduced by StreetInsider records another $30.41 million sold to 11 investors after a first sale on June 18. Together, the filings support the rounded $38 million total.
Fortune reported that the Series A valued Arrakis at $140 million post-money, citing Quintanilla. The round included GFC, MainObject and Rerail alongside Blossom Capital and Accel. Individual investors included Datadog co-founder and CEO Olivier Pomel, OpenAI head of business products Olivier Godement and Cambridge Aerospace founder Junaid Hussain.
The capital arrived before Arrakis had built a large organization or disclosed a broad customer base. Fortune put Arrakis at roughly 15 employees and five customers. Arrakis has identified customers only by description, including New York-listed enterprises in shipping, energy and other industrial sectors. Revenue, contract values and customer names remain undisclosed.
The speed of the financing reflects the wager investors are making on Quintanilla and his founding group. Accel knew Quintanilla as an investor and researcher; he co-authored Accel's 2024 Euroscape report on AI software. About half of Arrakis' team came from Palantir, Quintanilla told Fortune, giving Arrakis a pool of engineers familiar with deploying software in complex enterprises rather than selling self-serve tools.
Arrakis also has corporate structures ready for expansion. The UK entity was incorporated on December 19, 2025 as Silima Limited and renamed Arrakis Technologies on January 8. The March securities filing lists a New York address. Fortune reported that Arrakis plans to triple its roughly 15-person headcount and open outposts in New York and the Middle East.
Software with engineers attached
Arrakis combines a software platform with forward-deployed engineers who work directly with customers. The approach borrows from the Palantir playbook: begin with a specific operational problem, connect to the systems employees already use, and expand after producing a measurable result.
Fortune reported that one procurement customer cut cycle times by 90%. Arrakis has not named that customer or disclosed the implementation details behind the reduction.
Arrakis has positioned its commercial model around outcome-linked fees. Outcome-linked pricing gives customers a concrete test for projects that might otherwise become open-ended consulting engagements. It also puts pressure on Arrakis to show that its agents can survive contact with real operational data and produce repeatable results.
Arrakis says its platform is model-agnostic and can sit across model providers. Arrakis has not published independent evaluations showing how much its orchestration improves output quality or reduces token costs.
Model flexibility has practical value for industrial customers. A procurement or logistics system may remain in production for years, while the leading model provider can change within months. Arrakis is betting that large customers will pay an intermediary to choose and route models, protect customer data and maintain the surrounding operational logic.
The consulting question
The forward-deployed model gives Arrakis access to the messy details that decide whether enterprise AI reaches production. It also creates the central business question facing Arrakis: how much of each deployment can become reusable software.
Engineers working alongside customers can shorten implementation cycles and uncover repeatable product requirements. Heavy customization can also turn growth into a headcount problem, with each new contract demanding another deployment group. Arrakis' plan to triple its staff so soon after launch shows that people remain a substantial part of the delivery model.
Quintanilla is positioning Arrakis against two established groups. Consulting firms sell large transformation programs staffed by people, while Palantir sells a mature data and operations platform with its own forward-deployed engineering organization. Arrakis is offering narrower projects, outcome-linked fees and the freedom to switch between model providers.
The Palantir comparison carries particular weight because Arrakis has recruited heavily from the company and is pursuing many of the same operational environments. Palantir already has the customer relationships, deployment experience and product breadth that Arrakis still needs to earn.
Arrakis' early customers will determine whether Quintanilla found a repeatable opening or assembled a well-funded implementation practice. The first five contracts are too small a sample to settle that question. They are enough to show why a former investor left Accel: industrial AI spending is moving from model experiments toward the difficult work of connecting software to the systems that run physical businesses.