Newsletter · 2026-09-14 · Ashutosh Agarwal

AI Infrastructure Draws the Week's Biggest Checks as Mistral Becomes Europe's Most Valuable AI Startup - The Raise - Week of September 14, 2026

The Raise for the week of September 7 to September 14, 2026. Founder and VC podcast synthesis on the week's biggest checks flowing into AI's physical infrastructure, from Crusoe's data centers and Bluecore's floating reactors to the Boring Company's tunnels, plus Mistral's €3 billion Samsung-led round that made it Europe's most valuable AI startup and the founder story of Minted's Mariam Naficy.

The Raise

Week of September 14, 2026: AI Infrastructure Draws the Week's Biggest Checks as Mistral Becomes Europe's Most Valuable AI Startup

The biggest checks this week went into the physical machinery behind AI, Crusoe raising more than $3 billion to build data centers at a $30 billion value, France's Mistral pulling in €3 billion to become Europe's most valuable AI startup, Elon Musk's tunnel company landing $3 billion from the UAE, and a two-month-old startup putting nuclear reactors on barges. Plus the founder story of the week: Mariam Naficy, who sold her first company for $100 million cash two weeks before the dot-com crash, nearly shut her second one down, and turned it into a $300 million business.

If last week was about "AI that acts," this week the money moved down a layer, into the wires, chips, reactors, and even tunnels that all of it runs on. Investors wrote enormous checks to the companies building AI's physical backbone, and Europe finally minted a genuine heavyweight of its own. Here's what the podcasts covered in the week of September 7 to September 14.

This Week's Rounds

Mistral AI: €3 billion (about $3.5 billion) in a Series D round, valuing the French startup at €21 billion (roughly $24.4 billion), led by Samsung Electronics. This is one of the largest tech equity raises in European history, and it cements Mistral as the continent's answer to OpenAI and Anthropic. (We flagged an earlier report of this round back in July at similar numbers; this week it officially closed, with Samsung confirmed as the lead and the CEO out doing interviews.) The other backers include the EU-backed Scale-Up Europe Fund and US firm PSG Equity, and Samsung now sits on the cap table alongside ASML and Nvidia, a lineup CEO Arthur Mensch called "telling about the focus we have on industrial AI." His pitch is that Europe can't simply buy its AI from America or China: "any region cannot depend, cannot decide to fully buy the technology. They need to produce the technology." The money goes to three things: scaling up training compute, building sovereign AI infrastructure with a heavy physical footprint in Europe, and expanding sales in the US and Asia. Mensch also said Mistral's annual recurring revenue (the yearly value of its subscription contracts) is on track to "exceed a billion dollars this year," and that with the new deals, "we are absolutely tracking toward that target and expect to be beating it." His whole bet is on open-weight models (AI that companies can download and run on their own machines), which he argues is what enterprise customers in manufacturing, finance, and defense actually want: "almost 99% of the use cases you have as an enterprise can be built on open source models." (Squawk Box Europe Express, "Mistral raises €3bn in funding round," Sep 8, 2026; Bloomberg Tech, "Qualcomm Signs Deal to Provide Amazon With Custom AI Chips," Sep 8, 2026.)
Crusoe: more than $3 billion, at a valuation of roughly $30 billion, co-led by Redux Management and Valor Equity Partners, with Abu Dhabi's Mubadala Capital participating. Crusoe builds large-scale data centers and rents out AI computing power, and the raise marks a massive jump for a company that started life in 2018 doing something completely different: using excess natural gas from oil fields to power cryptocurrency mining. It pivoted that spare computing muscle straight into the AI boom. Its customer list now reads like a who's-who: OpenAI, Microsoft, and Meta, and it's building the flagship Texas data center for OpenAI and Oracle's "Stargate" project. The round came right after Crusoe locked in a roughly $13 billion deal to supply AI cloud capacity to the quantitative trading giant Jane Street. As the hosts put it, this is Mubadala's playbook: "identify the future leaders and invest in them when the time is right." (Worth noting: at recording, the round was reported as being finalized rather than formally announced.) (The Smashi Business Show, "Saudi Nuclear Deal, G42 AI Push & Mubadala's $3B Bet," Sep 7, 2026.)
The Boring Company: $3 billion, at a $23 billion valuation, a 4x jump from its last round in 2022. Elon Musk's tunnel-digging company got its big new check from the UAE, which is both the lead investor and the customer for its next project, a tunnel network in Dubai. The fascinating tension the hosts drew out: after nearly 10 years, the Boring Company's tunnels are actually open for riders in just one city, the Las Vegas Loop (a network that has carried about 5 million passengers since 2021, which one host noted is "equivalent to one weekday of the New York City subway"). So why the quadrupled valuation? The bull case is cost and potential. Its recent pilot tunnels in Dubai were built for about $39 million a mile, roughly 100 times cheaper than the latest New York City subway extension, which runs around $4 billion a mile. And the machines are going driverless ("zero people in the tunnel"), which could transform the economics. As the Best One Yet hosts summed up the Musk investing pattern: "You don't invest in what Elon has built. You invest in what Elon tweets about building someday in the future." (The Best One Yet, "'Miracle Below 34th St', New York's 9/11 comeback… The Boring Company's fundraise," Sep 11, 2026.)
Temporal: reportedly raising $500 million at a valuation of at least $12 billion. (Not yet closed, this is the rumor circulating.) If it lands, it would be more than double the $5 billion valuation Temporal held just six months ago, when it raised a $300 million round. Temporal makes "durable execution" software, plumbing that lets developers run long, multi-step computer processes that can survive crashes, timeouts, and failures without having to start over. It was originally popular for things like payment processing, but AI has been a huge tailwind: AI agents are, by nature, long-running multi-step processes that fail a lot, and nobody wants to restart (and re-pay for) a job every time it stumbles. As one host described it, Temporal is "one of those companies where it's like overnight success, but 10 years in the making," years of grinding, then viral word-of-mouth as happy engineers carried the tool from job to job. (Software Engineering Daily, "SED News: The NVIDIA–Hugging Face Deal, China's Proxy Economy, the Open Weight Surge," Sep 8, 2026.)
Bluecore Energy: a $50 million seed round, led by Silverton Partners, just two months after coming out of stealth. This is one of the week's best stories. Bluecore puts small nuclear reactors on floating barges, so clean power can be shipped to wherever it's needed: ports, AI data centers, communities short on electricity and water. Founder Asante wasn't even planning to raise this year, but the calls kept coming after the company's pre-seed launch in July, and it took just eight weeks to close: "I kept having to figure out a way to make more room on the cap table. I wasn't anticipating us raising that level of capital in that period of time." Existing backers Slauson & Company, Harlem Capital, and Ripple co-founder Chris Larsen doubled down, joined by CoLab Capital, Kevin Hart's Heartbeat Ventures, and angels from Tesla, Uber, Amazon, and Google. Bluecore is the first nuclear company headquartered at a major US port (Long Beach), has already launched two floating barges, and is now in review with the Nuclear Regulatory Commission and Coast Guard. On safety, Asante's framing: "The Navy is doing this as we speak for the last 70 years and has never had an accident. We just made it smaller." His team comes from SpaceX, Rivian, and Toyota, and his mission line is a good one: "access to clean energy and also clean water should be a human right." (TechCrunch Startup News, "Nuclear startup Bluecore Energy raises $50M seed round…," Sep 9, 2026; Equity, "An ex-Anthropic researcher's doomsday warning comes at a very interesting time," Sep 11, 2026.)
Poseidon Aerospace: a $60 million Series A, led by TQ Ventures, up from an $11 million seed last year. The startup builds uncrewed cargo planes and is deliberately boring about it: no vertical-takeoff gimmicks, no hydrogen or electric engines, just fixed wings and regular combustion engines, aimed squarely at moving freight for the lowest possible cost. Co-founder and CEO David Zaganoff (ex-Amazon logistics) and co-founder Parker Tenney (ex-Lockheed Martin) built it around a contrarian philosophy: "a lot of people who start aerospace companies get nerd sniped by very cool technology… For us, we want to improve cargo." Or, more bluntly: "Can I make a really good box with wings to move things for super cheap?" New investors Hanwha Asset Management, G-Squared, and Jaws joined, alongside existing backers Starship Ventures, Draper Associates, and Drover Ventures. Its first uncrewed cargo plane, "Egret," is expected to fly by year-end, targeting defense and regional commercial cargo, and, unusually, Poseidon plans to operate its own air-cargo business rather than sell the planes. (TechCrunch Startup News, "Nuclear startup Bluecore Energy raises $50M seed round…," Sep 9, 2026.)

Founder Story of the Week

Mariam Naficy: she sold her first company for $100 million two weeks before the dot-com crash, nearly shut her second down, and turned it into a $300 million business. Now she's building her third. (The Foundr Podcast with Nathan Chan, "704: How One Pivot Turned My FAILING Startup Into a $300M Brand," Sep 10, 2026.)

In a week full of AI infrastructure, the most human story was a 30-year lesson in patience, near-death, and knowing when to ignore your own investors.

Act one: perfect timing, by luck. In 1998, while still a Stanford Business School student, Mariam Naficy co-founded Eve.com, an early online beauty retailer, with her partner Varsha. When they pitched SoftBank, the response was blunt: "but guys, you know, women don't buy things online." They built it anyway: 120 employees, and $10 million in sales in the first year (1999), making it one of the biggest e-commerce companies on the web at a time when almost nothing but software was sold online. They sold to Bill Gross's Idealab for over $100 million in cash, and the NASDAQ crashed two weeks later. She's candid that a lot of it was luck: they'd planned to sell replenishable skincare, but color cosmetics "just started flying off the shelves," and the big beauty conglomerates that controlled distribution suddenly couldn't stop small indie brands from reaching customers directly online. (There's also a wonderful aside about buying the eve.com domain from a mother who'd registered it for her five-year-old daughter, Eve Rogers, Bill Gross closed the deal by offering the little girl an honorary board seat, 1% equity, a trip to Disneyland, and free educational software for life. She said yes.)

Act two: Minted, and staring at the exit. Naficy launched Minted in 2008 to sell design-forward stationery. Her first instinct was to hedge: put established stationery brands online, the safe move her investors preferred. It generated zero sales for an entire month, and the company was running out of runway. "I was about to shut the company down and I was really despondent," she says. "Every single day I'd have to like pick myself up and keep going… it just looked so bad." The thing that pulled her through was a mindset: "you just cannot care what other people think about you… if you fail, who cares?"

Her gut had wanted something weirder all along: crowdsourcing the designs. So at night, she hired a coder off rentacoder.com (who turned out to be a college student) to build a design-competition site, and coaxed artists into entering the first contest, which drew only about 60 designs. When couples started putting their own photos on "save the date" cards (something the printing industry told her "people don't do"), she trusted the crowd over the expert. (Today nearly every save-the-date has the couple's photo; Minted was first, "because the crowd told us to.")

Act three: the pivot that saved it, and ignoring the investors again. Still short on traffic and orders, Naficy decided to try holiday cards, after a miserable two-hour trip to a stationery store with her newborn. Her investors pushed back hard: "Mariam, you are losing focus. You started with weddings. You should focus on weddings." She did it anyway. The growth hack that made it explode: she went to a magazine ad agency and offered to be their "buyer of last resort": "call me if you have extra space, give me a massive discount and I will make you an ad in an hour," landing dirt-cheap ads in front of millions of women right before print deadlines. It "took off like a rocket ship so badly that we had to sort of start suppressing demand," hitting maximum production capacity and having to restrict orders to 50 cards or more. Crucially, she'd taken a Series A to keep the company alive (largely to protect her friends-and-family investors) right before the surge. And Minted thrived through the 2008 financial crisis because it sold premium cards (a $200 holiday-card order) to wealthy customers who weren't cutting back.

The lessons she keeps. Minted now does about $300 million a year (the host guessed $400 million; she corrected him to the publicly stated $300 million), and she's now its chairwoman. Two pieces of hard-won wisdom stood out. On money: "you always have to act like you have half the financing you have," because things take twice as long and twice as much cash, and she's watched great founders fail purely because they "ran out of runway" one milestone short of the next raise. On the brutal seasonality of a business that makes most of its money in Q4: "you build the church for Easter Sunday": enough capacity for the one peak day, idle the rest of the year, which in the early years meant everyone, finance team included, pulling customer-service shifts through the holidays.

Act four: Arcade. After 16 years running Minted, Naficy is now building her third company, Arcade, an "AI-to-physical-product marketplace" where you describe a product in plain words ("a holiday tablecloth with a red grid… red botanicals in the middle of each square") and get an instant, guaranteed, actually-manufacturable price. She's raised a $25 million Series A (March 2025) and about $42 million total since a September 2024 beta launch, and built a computer-vision team that includes someone who trained foundational image models at Stability AI. The mission rhymes with her whole career: using a marketplace to build livelihoods for creative people, this time paying independent artists a cut every time their trained "model" generates a sold product: "they shouldn't have to have their future earnings taken away by AI." (The Foundr Podcast with Nathan Chan, "704: How One Pivot Turned My FAILING Startup Into a $300M Brand," Sep 10, 2026.)

Also Heard

Anthropic walked away from a $6 billion acquisition of Decart. After weeks of full due diligence on the Israeli AI startup, Anthropic reportedly said no and signed nothing. (We noted these talks a few weeks back; now they're off.) Decart builds software that sits between AI models and the chips they run on, letting workloads move freely across Nvidia, Google, and Amazon hardware, its CEO called it the "Great Equalizer," claiming an eight-times improvement in compute efficiency. The hosts' theory on why the deal died: that headline "8x" number tends to hold up in a clean lab but collapse under real-world messiness, and with a huge IPO on the horizon, Anthropic had every reason to avoid "a complex, multi-billion dollar integration distracting the engineering leadership team during a roadshow." A telling data point on how frothy prices have gotten: Decart had raised at a ~$4 billion valuation, itself a fast jump from $3.1 billion, and $6 billion would have been another 50% premium on top. (Elon Musk Podcast, "Anthropic rejects six billion dollar Decart deal," Sep 9, 2026; The Twenty Minute VC, "Jensen Huang Declares AGI Has Arrived… Anthropic Pulls From Descartes Acquisition," Sep 10, 2026.)
Oura filed to go public. The smart-ring maker filed for a NASDAQ IPO under the ticker OURA, reportedly seeking to raise around $3 billion at a market value near $16 billion. The numbers behind it are genuinely strong: about $1.4 billion in revenue, up 74% year-over-year, 5 million paid members (up from 1.5 million in early 2025), and 85% retention over 12 months. The debate among hosts: whether a brand beloved by athletes and biohackers can go mainstream, and survive Apple Watch adding more health features. (Brew Markets, "Canada's Tariffs Take Effect & Summer Box Office Boom," Sep 8, 2026; Chit Chat Stocks, "Stocks At 52-Week Lows; Meta's Agentic AI Push…," Sep 11, 2026; Run the Numbers, "How Oura Built a $1.4 Billion Business Around a Tiny Ring," Sep 10, 2026.)
Two clean startup exits. Dynatrace agreed to buy Arize (an "AI observability" company that helps businesses monitor what their AI agents are actually doing in production) for $915 million. Founded only in 2020, Arize had raised relatively little (a ~$70 million Series C in 2025), making this a tidy outcome; the hosts framed it as the kind of consolidation coming for many single-problem AI startups that are "solving an important problem, but it's hard to create a really truly large business" on it alone. Separately, the team behind open-source database startup InstantDB joined OpenAI in what's being described as an acqui-hire, as OpenAI keeps assembling its own agent-infrastructure stack. (Software Engineering Daily, "SED News: The NVIDIA–Hugging Face Deal, China's Proxy Economy, the Open Weight Surge," Sep 8, 2026.)
Chime bought its own bank. The fintech Chime announced a $590 million acquisition of Stride Bank, its longtime banking partner, to speed up product launches in a category where waiting on third-party approvals slows everything down. CEO Chris Britt also raised Chime's guidance, to 30% top-line growth and a 17% profit margin. (Squawk on the Street, "10AM Hour: Wells Fargo, Chime & Charter CEOs 9/9/26," Sep 9, 2026.)
The mega-round tally kept climbing (with the usual caveats). Two names we've tracked got fresh, bigger numbers this week, flagged so you're not double-counting. Cognition, the maker of the AI coding tool Devin, is now reported to have completed its round at roughly $2 billion at a ~$48 billion valuation (we covered it last week as "~$1 billion at $46–47 billion"). And Anthropic is reportedly filing its IPO paperwork in late September, targeting a listing in mid-October at a headline valuation as high as $2 trillion, while, notably, posting what was described as its first quarterly operating profit of $559 million. As ever, the revenue figures attached to these AI names vary wildly from podcast to podcast, so treat the specific numbers with a healthy dose of skepticism. (The AI Daily Brief, "AI Model Month Is Off to a Blistering Start," Sep 9, 2026; Rich Habits Podcast, "Trump's $5K Bribe, Anthropic's $2T IPO, & Meta's Muse," Sep 11, 2026; Elon Musk Podcast, "Why AI labs gamble with human extinction," Sep 10, 2026.)
A twist on last week's story. We covered "Town," the ~$1 billion AI-assistant startup, just last week. This week, one of its lead backers, Index Ventures, reportedly pulled out of the round. A reminder that even the hottest deals of the moment can wobble before the ink dries. (The Twenty Minute VC, "Jensen Huang Declares AGI Has Arrived… Index Pulls Out of Town…," Sep 10, 2026.)
Quick hits. Reusable-rocket startup Stoke Space raised "another billion" (TechCrunch Startup News, Sep 9, 2026); quantum-inspired AI company Multiverse Computing was discussed around a $570 million Series C at a ~$1.7 billion valuation, though that round was announced a few weeks earlier (The New Quantum Era, Sep 7, 2026); and insurance-brokerage roll-up Hub International filed to go public at a $29 billion valuation, a test case for whether public markets will validate a decade of private-equity deal multiples (The Insurance Guys Podcast, Sep 9, 2026).