DeepSeek wants ten billion dollars at a forty-five billion valuation, and nobody's named a lead investor. Biggest Chinese AI fundraise on record, and the cap table is still a black box. Welcome to Startup Fundraising. Today: DeepSeek's record raise, Cognition finally getting a denominator for that billion, and a DriveNets Series D with actual invoices behind it — which is already more than most of this week can say. And Anthropic files a confidential S-1 the same week DeepSeek drops ten billion. Public markets are about to price frontier AI with no P&L in the room. That's the escalation I wanted. We'll walk through all of it — but that DeepSeek cap table is the first thing I'm pressing on. Cyber News Centre writes:
In what would be the largest first-time financing by a Chinese tech startup on record, Beijing-based DeepSeek is in advanced discussions to raise $10 billion (70 billion yuan) at a staggering $45 billion valuation, the Financial Times and Bloomberg reported.
DeepSeek: ten billion dollars, forty-five billion valuation, largest Chinese AI fundraise on record. And the Cyber News Centre piece still doesn't name a single lead investor. That's the first thing I need before I can read the rest of the cap table. Liang Wenfeng is "focused on AGI, not short-term profit" — which is exactly the kind of line you say when sovereign backers are writing the checks and nobody's asking about a quarterly call. What has to be true in twenty-four months for forty-five billion to make sense? The article doesn't go there. And the timing is the weird part: Anthropic files a confidential S-1 the same week. One is asking public markets to price frontier AI with a prospectus. The other is asking sovereign capital to price it with no visible terms and no lead named. Those are very different tests landing in the same news cycle. The Anthropic filing at least drags a P&L into the room eventually — that's what an S-1 does. DeepSeek doesn't have that mechanism. "Sovereign backers rush to fund China's open-source AI champion" is geopolitics dressed up as a term sheet, and I don't know what I'm pricing. Memeburn writes:
Cognition AI just raised more than $1 billion at a $26 billion valuation as its AI coding agent Devin reportedly writes nearly 90% of the company’s internal code. The funding highlights how quickly AI coding tools are starting to handle larger parts of software development with less human involvement, with investors betting billions on that shift.
Cognition closes the billion at a twenty-six billion valuation, and Memeburn is the outlet putting the eighty-nine percent stat right up front — the number that's been doing valuation work all week. But let's be exact: Devin writes eighty-nine percent of Cognition's own internal code. Goldman Sachs, Microsoft, Dell, Cisco, and Palantir are named as enterprise customers, and that's a different evidentiary bar than a repo metric. Right — eighty-nine percent of Scott Wu's codebase is not eighty-nine percent of a Goldman Sachs production environment. Those are two different sentences, and the headlines keep collapsing them into one. The round is done; the story is still carrying the weight on whether this is an "AI replaces engineers" moment or just a very impressive internal demo. We now have four hundred ninety-two million in annualized revenue against that twenty-six billion — a fifty-three-times multiple, which we flagged yesterday. The eighty-nine percent stat adds texture, but it doesn't move the multiple. What it does do is sharpen the software supply chain question: if your AI is writing nearly all your code, who's auditing the binary output? That's a moat claim and a risk disclosure at the same time. And NVentures is in the Inherent seed this same week — a lab whose entire pitch is, basically, "we haven't shipped Faraday yet." Cognition at least has a stat. Inherent has a mission statement. NVentures apparently prices both the same way, which is a posture worth watching. PR Newswire writes:
DriveNets, a leader in large-scale networking solutions, today announced it has completed a $410 million Series D financing round, reaching $1 billion total capital raised. With more than $1B in secured business and having been cash-flow positive since 2025, the company will use the additional funding to scale inventory to support its growing AI fabric pipeline and expand its Heterogeneous AI infrastructure solutions. The funding round was led by Bessemer Venture Partners and Atreides Management.
DriveNets, Ra'anana, Israel — four hundred ten million Series D, Bessemer and Atreides co-leading, AMD and Red Dot Capital coming in new, Pitango and D1 staying in. Total capital raised hits one billion. And the PR Newswire release gives you the denominator right in the subtitle: more than a billion in secured business, cash-flow positive since 2025. That's an invoice stack behind this raise, not a narrative. Ethernet fabric for AI deployments, telecom as the original customer base, and now foundation labs and hyperscalers in the pipeline — this is the unsexy infrastructure story I've been waiting for all week while everyone else was busy pricing vibes. Worth flagging: AMD is in this round as an investor and named as a technology partner in the same announcement. That's a strategic check with integration upside baked in — Bessemer is leading, but AMD's presence on the cap table isn't just financial. And this is Israeli-headquartered, closing the same week XCENA's Korean LP geography was in the conversation. Non-U.S. capital keeps showing up in the AI infrastructure layer — DriveNets just added another data point to that thread. Kristen Winzent, writing in Gigascale Capital:
Today, we’re announcing Gigascale Capital Fund I, a $250 million fund to back founders rebuilding the physical economy for climate impact. We believe climate progress will come from better-performing systems: technologies that are cheaper to deploy, faster to scale, more reliable, more productive, and cleaner than what it replaces.
Gigascale Capital Fund I: two-fifty million, first-time fund, climate and the physical economy — clean American rare earths, nuclear microreactors, commercial fusion. The thesis is performance, not virtue: cheaper to deploy, faster to scale, cleaner than what it replaces. They point to solar's forty-gigawatt to six-hundred-gigawatt decade as the proof of concept. This is the fund I've been waiting to see all week — no AI, no "agentic" anything, just an invoice-ready answer to what it actually costs to replace physical infrastructure. Twenty-five portfolio companies over three years before closing Fund I is a real sequencing choice: they built a track record before they named the fund. Worth flagging for the concentration conversation we've been having: this week's capital story has been dominated by frontier AI mega-checks, and here's a two-fifty million first-time fund closing on a completely different thesis. That's the data point the "emerging managers getting squeezed out" thread needed. The performance framing matters here — "cheaper to deploy, faster to scale" is a metric commitment, not a mission statement. The question in two years is whether the microreactor or rare-earth bets have a revenue stack behind them the way DriveNets just showed up with a billion in secured business. Same evidentiary standard, different physical bet. From Ekemini at Ventureburn:
Inherent has raised a $50 million seed round to fund the development of “Faraday”, an ambitious AI platform engineered to redefine the scientific method for the machine learning age completely. The mega-seed round was jointly led by two of the giants of venture capital, Index Ventures and Radical Ventures, and was sprinkled with participation from NVIDIA’s V.C. wing, Nventurers.
Inherent's fifty million seed closes with Index and Radical co-leading, and NVentures participating — Ventureburn had this first, and the cap table is now fully confirmed. Credit where it's due on the sourcing. NVentures was in OpenRouter, then in this round, and Faraday still has no shipped milestone — just a mission statement about rewriting a four-hundred-year-old paradigm. That's a lot of rhetoric for a seed check. To be fair, Index and Radical don't co-lead a fifty million seed on vibes alone. But the Faraday pitch is still defined by what it hasn't done yet, and that's a harder sell the bigger the check gets. Ex-DeepMind, ex-Microsoft, Biden White House policy pedigree — genuinely elite credential stack. And Faraday's job is to ask questions humans didn't think to ask. Great. What has it asked? Because this same week DriveNets is sitting on a billion in contracted revenue, and I can read their invoice stack. Have a fundraising question, a story idea, or a correction for us? Send it our way at startupfundraising at lantern podcasts dot com. We read every note, and your feedback helps shape future briefings.
You'll find links to every story we covered today in the show notes, so if something stuck with you, they're there for a deeper read.
That's Startup Fundraising for today. Thanks for listening, and we'll be back tomorrow. This is a Lantern Podcast.