← Startup Fundraising

AI capital piles into Chai, DeepSeek and defense infra (July 15, 2026)

July 15, 2026 · 10m 48s · Listen

Four rounds today, and for once, the size is almost the least interesting part — I'm looking at who signed the checks, and on what terms. If you're just joining, the AI venture market's been bunching into fewer, bigger rounds — infrastructure, models, apps. We went from PitchBook-NVCA numbers showing AI ate most of the first-half U.S. venture dollars, to Prime Intellect's $130 million at a billion, then back-to-back mega-rounds for PixVerse and MiniMax. This is Startup Fundraising. Today: a biotech AI round, a Chinese LLM slapping a price tag on itself before IPO, and a defense startup worth $400 million on day one of existence. I've got questions. Start me with Chai. We're staying on AI venture market concentration — follow the show and you won't miss what comes next. Here's SiliconANGLE:

Chai Discovery Inc., a company that develops artificial intelligence models to predict interactions between biochemical molecules, today announced it has raised $400 million in new funding, nearly tripling its valuation to $3.8 billion. Index Ventures led the Series C round alongside Kleiner Perkins, Sequoia Capital and Dimension. New investors joining the investment included Bain Capital Ventures, Battery Ventures, Baillie Gifford, BDT & MSD, Sapphire Ventures, Avra Capital and other notable investors.

Chai Discovery: $400 million Series C, valuation nearly tripling to $3.8 billion. The detail I keep circling is Index Ventures leading. After a week of sovereign funds and strategics piling in, finally, a mega-round where we can point to a clean institutional lead. So now the AI-funding concentration story has a Sand Hill face, not a state fund. Kleiner, Sequoia, Dimension are riding alongside — but Index actually wrote the lead check and set the terms. Fine, real firm, real lead. But look at the pace — $130 million at $1.3 billion in December, and seven months later they're at $3.8 billion. Three rounds in under a year. And the headline says AI-designed antibodies 'reach Big Pharma.' Reach how? A partnership announcement isn't a purchase order. What does Chai actually get paid when a Big Pharma uses one of their models in discovery? That's the piece we can actually test. 'Reach' could mean a signed deal with milestones, or it could mean a pilot and a press release. The terms of those partnerships matter more than the $3.8 billion. Two years from now, there'd better be revenue tied to a molecule that made it into a real pipeline. Otherwise, it's just a very expensive model demo. That's how we'll know whether this round made sense. Inc42, with Anjali Jain:

VC firm Elevation Capital has launched Fund IX with a target corpus of $500 Mn to back Indian startups across consumer tech, consumer brands, fintech and financial services, enterprise AI, frontier tech and healthcare. The new fund will primarily invest in seed and Series A startups building AI-first products amid what the firm described as the ongoing AI platform shift.

Elevation Capital's Fund IX: $500 million for India, seed and Series A, aimed at what they're calling AI's app layer. And it lands less than a year after their $400 million IPO-bound fund. So two funds in under twelve months. That's a firm raising as fast as it can deploy — which tells you the LPs aren't the ones being cautious here. The number that actually jumps out: Indian AI startups pulled $676 million in the first half of 2026. That's more than four times the $162 million from a year earlier. And here's what I like — these checks are going after the app layer, not the models. Somebody sophisticated is betting value accrues to what's built on top, instead of the foundation layer everyone's overpaying for. That's a different shape from strategic corporates writing checks into their own supply chain. Here, an institutional India fund is betting on where usage shows up — without the chipmaker angle of seeding future customers. Seed and Series A, too. So they're not paying the mega-round premium — they're getting in before the valuation gets stupid. For once, I don't have a complaint. Here's Dominic-Madori Davis at TechCrunch:

DeepSeek, the Chinese large language model developer, is preparing for a 2027 IPO debut, but it could come as early as the end of this year as it also looks to raise around $1.5 billion in new funds at about a $71 billion valuation, Bloomberg reports. The news comes after the company raised$7 billion in funds just a month ago at around a $50 billion valuation in its first-ever outside funding round.

DeepSeek raised seven billion a month ago at fifty billion. One month later they want another one-point-five at seventy-one. That's a forty-two percent step-up in thirty days, per Bloomberg. And it's a cap-table story wearing a growth-round costume. If a 2027 IPO can suddenly come as early as the end of this year, the raise starts to look like a bridge with a reference price attached. Right — the calendar gives it away more than the check size. You don't compress an IPO window that hard because the business is starving. You do it to set a private mark before you file. The commercial case has substance, though. Twenty-three percent of enterprise tokens through Vercel's gateway in June, versus Anthropic's thirty-two. For a Chinese open-source shop running on Huawei chips under U.S. export controls, that's a real adoption number underneath the valuation. Fine — the usage is real. But nobody in this raise is price-sensitive. US labs, Chinese labs, same week, and the number just goes up. Tencent and a Beijing state fund are already on this cap table, so who exactly is negotiating the seventy-one? SiliconANGLE writes:

Air defense startup Singularity Defense Corp. formally launched today, announcing that it has raised $80 million in new funding at a $400 million valuation to mass-produce low-cost interceptors that can knock down the cheap drones and missiles now common on the battlefield.

Eighty million at a four hundred million valuation — for a company that formally exists as of this afternoon. That's five times the money before a single interceptor comes off a line. And read what the CEO actually said. 'We're staying pretty damn light on details.' No specs, no cost-per-interceptor, just 'much lower cost' than a Patriot. Okay — lower than hundreds of thousands is a wide field. The pitch centers on the assembly line, more than the hardware. Production team pulled from Tesla and Toyota, building automotive-style manufacturing to out-ramp any comparable U.S. system. This lands right next to the European defense escalation we've been watching — Helsing at eighteen billion, Quantum Systems at eight. Singularity's the U.S. data point: this multiple expansion has crossed the Atlantic. The whole bet is kill economics at volume. A cheap drone costs a few thousand; you have to intercept it for less, and you have to mean it at scale, not just in a demo. A day-one valuation at five times the money says the assembly line has to be real, not just good slides. From SecurityWeek:

UK-based sovereign infrastructure company Valarian announced on Tuesday that it has raised $50 million in Series A funding. The company, which previously raised $20 million in seed capital, said the latest funding comes from New Enterprise Associates (NEA), Lightbank, XTX Markets, Sequel, LitVC, and angel investors Gokul Rajaram and Nikesh Arora. The new investment will fund Valarian’s expansion and enable it to enhance its platform.

Okay, a UK sovereign infrastructure company raises a $50 million Series A, and honestly? This is the one on the slate I don't have to squint at. Valarian built ACRA — a control layer on top of Kubernetes that boxes every workload into its own enclave, and customers hold their own keys, so Valarian can't see inside once it's live. And NEA led it — another clean, named lead. Second one today if you're counting the pattern: an actual Sand Hill firm driving the round, instead of a sovereign fund piling in at whatever price. Right, and the money's for a boring reason — expansion and enhancing the platform. No 'category-defining' fireworks. After a seventy-one-billion-dollar DeepSeek number floating around today, a $50 million round for regulated-enterprise plumbing feels almost quaint. They already raised $20 million in seed, so this is a real step-up, not just a re-label. And the angel list is telling — Nikesh Arora, Gokul Rajaram. Operators who know what selling into locked-down enterprises actually takes. That's the tell for me. Air-gapped, default-deny, short-lived secrets, customer-held keys — that's a product built for people who've already been burned. In two years, regulated buyers need to actually be deploying this, not just admiring the architecture. But at least the claim is testable. If you track startup fundraising, you might also like Infrastructure Secondaries Daily: LP stake sales, GP-led continuation vehicles, and discount-to-NAV pricing, every day. It's a sharp read on private-market liquidity. Find it wherever you listen to podcasts.

What we're watching next: DeepSeek is reportedly preparing for a 2027 IPO, with a listing possible as early as the end of 2026.

You'll find links to every story from today's briefing in the show notes, so if one caught your ear, you can dig in there. That's Startup Fundraising for today. This is a Lantern Podcast.