Covers Wednesday, Oct 28 through early Thursday, Oct 29 (Eastern). Wednesday's money went to the layers underneath AI: a Korean inference-chip maker whose largest backer is a state fund, a cloud-security company that more than doubled its valuation in under nine months, and a startup that owns the factory instead of selling robots to one. Eight rounds in the window cleared our checks, plus one Oct 27 round the last Daily Brief missed (marked "catch-up"). Every round below is confirmed by the company, its lead investor or both. Figures companies report about themselves are labeled company-reported.

FuriosaAI, the Seoul-based designer of AI inference chips, closed a $620M (KRW 820 billion) Series D at a post-money valuation of about $2.9B (FuriosaAI release · FuriosaAI blog · Bloomberg Law / Bloomberg News). The company does not name a lead. South Korea's National Growth Fund put in $280M (KRW 370 billion), the largest amount from any investor. International investors added $110M in total, with U.S.-based Infinitum the largest of them, alongside Jericho Capital, QUDA Capital Partners, Cisco Investments, Abu Dhabi's NeoFort Holding and Worldline Capital. Existing shareholders contributed $125M, including $22M from the Korea Development Bank; NAVER (through NAVER Ventures), DSC Investment and KDB Capital, its three original seed investors, all returned. New Korean institutions include the Export-Import Bank of Korea, which FuriosaAI says chose it for the bank's first direct investment in a startup, UAMCO ($30M) and Hanwha Asset Management. FuriosaAI's RNGD card, a 180-watt, air-cooled accelerator, entered mass production in January 2026. The company plans 20,000 units this year and more than 40,000 in 2027, and names LG Group, Upstage and Samsung SDS as deployments (company-reported). Part of the money goes to a third-generation chip it is developing with Broadcom and other partners. Why it matters for investors: Yesterday's Daily Brief covered a U.S. cloud raising money to rent out specialist inference chips from Nvidia's rivals. Today one of the companies that makes such chips raised six times as much. FuriosaAI's pitch rests on power, not raw speed: it says RNGD matches the throughput of Nvidia's RTX PRO 6000 card for card while serving twice as many users per kilowatt (company-reported). That matters because data centers are now short of power as well as chips. The cap table is the part to study. A state fund supplied roughly 45% of the round, and Korean state-linked lenders, including the Korea Development Bank and the Export-Import Bank of Korea, also invested. Capital like that can be patient, but it can also price a round on industrial policy rather than on returns. Diligence questions: How much of the 2027 volume is already under signed orders? How exposed is the roadmap to scarce memory and leading-edge foundry capacity? Do any state-linked investors hold terms that shape where the company manufactures or sells?

Historical lens. A state fund on a chip company's register has a famous precedent. TSMC's annual report for 2024 on Form 20-F says the company was founded in 1987 as a joint venture among the R.O.C. government and other private investors, and incorporated on February 21, 1987 (TSMC Form 20-F). A technology agreement with Philips, filed with the SEC, records that TSMC was set up by the Executive Yuan's Development Fund, N.V. Philips and investors from the Republic of China to make chips to its customers' own designs (Philips–TSMC agreement). The same 20-F says the National Development Fund, one of TSMC's largest shareholders, has held a board seat since the founding and owned 6.38% of the company as of February 28, 2025. It adds that the fund has at times sold shares in the form of ADSs. Our interpretation, not established fact: Korea's National Growth Fund, the largest investor in FuriosaAI's Series D alongside corporate backers such as Cisco Investments and NAVER, resembles that founding mix of state money and industry partners. TSMC's record shows a state holder can stay for decades and also sell. Which role Korea's fund will play at FuriosaAI is not yet known.

Upwind Security, a cloud and AI security company founded in 2022, closed a $400M Series C at a $3.8B valuation, led by Bessemer Venture Partners (Upwind release via GlobeNewswire · Upwind blog · TechCrunch · SiliconANGLE). Canapi Ventures is new. Every existing investor returned: Cyberstarts, Greylock Partners, Craft Ventures, Leaders Fund, Salesforce Ventures, TCV, Alta Park Capital, Penny Jar Capital, Cerca Partners, Swish Ventures and Picture Capital. Bessemer also led the $250M Series B that valued Upwind at $1.5B in January 2026, less than nine months ago. The company says its revenue has quadrupled in 2026 but gives no dollar figures (company-reported). Its platform builds a live picture of what is actually running across cloud workloads, APIs and AI agents. It then uses that runtime data to rank the exposures attackers can actually reach, and runs AI agents that investigate threats, test the attack surface and fix problems. Named customers include Siemens, Roku, Carvana, Nubank, Supabase and ClickUp. The money goes to product development and sales. Why it matters for investors: This is a markup of about 2.5 times in under nine months, from the same lead, on growth the company describes only as a multiple. Insiders who know the numbers paid up, which is a signal. But outside investors are being asked to accept a 4x growth claim with no revenue base attached. The bet is that AI agents create a new kind of insider: software with credentials that changes production systems faster than people can review it. That pushes buyers toward platforms that watch live behavior rather than periodic scans. Diligence questions: What is the revenue base behind the 4x? How much of it comes from AI-specific products versus classic cloud posture tools? How does the company hold its ground against the big security platforms that bundle the same features?

Foundry Robotics, based in Emeryville, Calif., raised an oversubscribed $100M Series A led by Marlinspike Partners (Foundry Robotics release via Business Wire · release text via VC Access Online · San Francisco Business Times). Early backers Khosla Ventures and Red Glass Ventures and existing investors Hanabi and Zero Shot Fund returned; SE Ventures, Wing VC and Axiom Ventures are new. Foundry sells manufacturing as a service. It owns and runs the plant, now a 130,000-square-foot facility that is 20 times its earlier footprint, and assembles complex electromechanical hardware for defense and commercial customers. Products include drones, wire harnesses, underwater vehicles, satellites and data-center gear. AI models trained on basic assembly steps run in a grid of robotic cells, linked by an agentic system it calls FoundryOS, and every production run feeds back into the models (company-reported). CEO Adarsh Kulkarni was employee number four at Ghost Robotics and later Scale AI's first head of robotics. Co-founder Emmanuel Lamacchia was program manager for the Tesla Model Y. The president of HII Unmanned Systems is quoted in the release backing the approach. Why it matters for investors: Most robotics startups sell or lease machines and leave the factory risk to the customer. Foundry takes the opposite position: it owns the building, the cells and the output. That makes it closer to a contract manufacturer with a software edge, so the numbers that matter are utilization, price per unit assembled and capital spent per cell, not software-style gross margins. The upside is that every new product it builds adds training data no robot vendor can see. Diligence questions: How much capital does each new cell or site need, and who funds it, equity or equipment debt? How concentrated is demand in defense programs? How quickly can a cell really switch from one product to the next?

Round notes

Not investment advice. The AI Cap Table is published for information only and is not investment, legal, or tax advice. Nothing in this newsletter is an offer or solicitation to buy or sell any security. Funding data comes from public sources we cite and verify, but errors are possible; do your own due diligence and consult a qualified adviser before making any investment decision.