Oxide Computer paid income tax this spring. In October, it raised $445 million.
Startups raise money because they burn it. Oxide raised because it is selling computers at a profit. The Series D, led by Eclipse, values the seven-year-old company at $6 billion and closed with orders already ahead of what the factory can ship.
The round rests on one number. Over the past twelve months Oxide scaled manufacturing capacity by 20×, and demand still leads supply.
Profitability, then a megaround
Hardware is where startups go to die slowly. Oxide did the reverse: it reached profitability earlier in 2026 on the ordinary business of building and selling machines, not on a one-time gain. It is rare enough that the company opened its announcement with a photograph of a tax wire.
The $445 million is working capital, not survival money. Components and assembly get paid for long before a customer pays for a rack, so a growing order book drains cash even while the income statement stays green. Profit measures a finished sale. A backlog measures a promise. The distance between them is what the round covers.
Seven years to a single rack
Oxide sells one product: a rack that behaves like a small cloud region inside a customer's own data center. Compute, storage, and networking arrive pre-integrated, wrapped in software the company wrote from the bootloader up. Twenty-four sleds, each carrying an AMD EPYC processor with up to 144 cores, roughly 1.64 TB of memory, and up to 300 TB of flash.
CHRONICLE: Oxide Computer
─────────────────────────────────────────────────────────────
2019 ──── 2023 ──── 2025 ────── Feb 2026 ──── Spring ──── Oct 2026
🌱 🔧 📦 💰 ◉ PROFIT 🔥 $445M
$20M Series A Series B Series C paid Series D
seed $44M $100M $200M income tax at $6B
Funding and milestones, 2019–2026. Sources: Oxide, Axios.
The path there was slow and unglamorous. Steve Tuck and Bryan Cantrill founded the company in 2019 in Emeryville, California, against the consensus that renting infrastructure from hyperscalers would eat everything. They spent years building hardware, networking, an operating system, and a control plane as one system — the reason a rack can go from delivery to provisioning without a data-center crew.
Customers today sit in financial services, government, national laboratories, and aerospace, mostly in places where running other people's computers is not an option.
Why CPU racks, not GPUs
The rack ships without a graphics processor. For most of the AI boom, that read as a fatal omission.
Oxide's answer is that agentic systems spend most of their time outside the model. Code harnesses, tool calls, retrieval, orchestration, logging, data storage — that work runs on CPUs, memory, and disk. The GPU handles inference; everything that feeds it and acts on its output needs general compute. When AI moves from answering to executing, that general compute stops being background noise.
Oxide's Series D round
Led by Eclipse at a $6B valuation; existing backers USIT, Riot Ventures, and Jane Street, with Atreides Management and AMD Ventures new. Oxide, 2026
In August the company said it would build a compute sled that can host a graphics card, and pointed at AMD. Intel Capital, an existing backer, and AMD Ventures, new in this round, now sit on the same cap table. The rack that had no accelerator is getting one — on its own timeline, after the CPU business paid for the roadmap.
Equity to fund the backlog
Oxide could have financed the backlog with debt. It chose equity, and said why: the founders call themselves children of the Dot Com bust, cautious about shocks they cannot see coming. Debt against inventory works until the market turns. If demand cools, the components are still on the books and the lenders still want paying.
Equity absorbs that risk, and it costs the founders dilution. The Hacker News thread that greeted the round spent much of its 300-plus comments on exactly that trade — whether a profitable company should keep selling stock to fund purchase orders, or borrow against machines it has already sold.
The answer hangs on what the order book is worth. If those orders convert, equity was the expensive choice and the dilution is noise. If the AI buildout stalls, equity is the choice that keeps the company standing. Oxide picked the version it could survive.
There is a second reason the round is equity rather than debt. Oxide wanted the cap table, not only the cash. AMD supplies the processors and now holds a stake; Atreides Management studied the company for two years before writing a check. Capital that arrives with a roadmap is worth more than capital that arrives with a coupon.
The own-versus-rent math
Oxide's bet is that the pendulum has swung back toward ownership. Public-cloud prices, vendor concentration, and lock-in all push the same way. When a workload runs around the clock, owning the iron wins on cost; when it spikes twice a month, renting still wins. Oxide is chasing the steady, regulated, data-resident load that was never going to leave the building anyway.
Whether the GPU sled ships before the backlog forces buyers back to a hyperscaler
Whether manufacturing keeps pace with the order book across 2027
Whether AMD's stake turns into a shared silicon roadmap
Whether the public-sector channel, added in September, converts to named customers
The accelerating pace of on-prem cloud computing projects has required us to scale manufacturing capacity by 20x over the past 12 months, and yet demand still exceeds supply.— Steve Tuck, co-founder and CEO, Oxide Computer
What has to go right
Two things carry the story from here. The first is throughput. A backlog is a gift until lead times push buyers back to the cloud they were leaving; Oxide has to convert orders faster than it can disappoint. The second is the GPU transition. Add accelerators badly and the integrated-system pitch becomes a rack plus somebody else's problem — the exact fragmentation the company exists to remove.
The company is also, quietly, the most visible test of a larger claim: that a hardware business can be profitable, stay independent, and still scale into the AI buildout without turning into another capital-hungry neocloud. Profitable, then funded. On-prem, not rented. It is a contrarian pairing, and for now the order book is on its side.