❯ AI inference cloud company Groq raises $350 million, valuation halved to $3.5 billion
[CHIPS FOR CLOUD] Groq has closed a $350 million funding round at a post-money valuation of $3.5 billion, led by investment firm Disruptive, with Nvidia planning to participate. What it now sells is compute rental: it installs Nvidia GPUs across 13 data centers globally and sells inference capacity by usage to developers and enterprises. The in-house LPU chip line is no longer its main business. The company says the platform has more than 6 million developers, enterprises, and AI-native companies.
[A HOLLOWED-OUT YEAR] As of last September, Groq’s valuation was still $6.9 billion; in less than a year, it has been cut by nearly half. What happened in between is rare even by chip-industry standards: in December 2025, Nvidia secured a license to Groq’s core inference technology at a reported cost of $20 billion, while also hiring away founder and CEO Jonathan Ross, president Sunny Madra, and roughly 90 percent of the engineering team. The money went to existing shareholders. The corporate entity stayed; the people and the technology left. This June, Groq first raised $650 million to kick off its transformation, and two months later took this $350 million. Groq itself disputes that this is a down round, saying it is repricing “the Groq that exists after the Nvidia licensing deal.” Now in the CEO seat is Alex Davis, chairman of Disruptive.
[WHAT REMAINS] With the chips stripped away, Groq genuinely has three things left: 13 data centers spanning North America, Europe, the Middle East, and Asia-Pacific; current installed capacity of 54 megawatts, with a company target of more than 200 megawatts by 2027; and a distribution base built up from more than six million developers. That hand dictates that it can only push in the direction of a new-style cloud, with competitors becoming companies like CoreWeave that specialize in renting out AI compute — no longer Nvidia. The problem is that the financial model for this business is inherently unattractive: capital expenditure is extremely heavy, typically held up by debt, the GPUs it buys depreciate quickly, and no matter how fast revenue grows, it may not convert into free cash flow. CoreWeave has already demonstrated that difficulty. Groq’s financials are still not public, and outsiders cannot see its unit economics.
[RESETTING THE VALUATION] In the same week, Groq’s valuation was halved because it sold chip technology to Nvidia, while Etched’s valuation doubled because it delivered racks into a customer’s data center. The two events are two sides of the same rule: the value of inference hardware lies not in the design blueprint, but in machines that are powered on and running. This $3.5 billion round is no longer buying a chip company; it is buying a compute sublessor that does business using someone else’s hardware. The valuation framework has accordingly shifted from semiconductors to infrastructure. What has truly rewritten expectations is for the other chip teams challenging Nvidia: the best outcome may be absorption through a licensing deal, rather than holding out until mass production and an IPO. Existing shareholders got their money back via the $20 billion licensing fee, the company left behind was repriced, and more people will likely factor this exit path into their models from now on.
▪ SIGNALNvidia used $20 billion to buy the technology and the people, leaving behind a shell that has pivoted to renting out GPUs. The optimal path for challengers is shifting from mass production and IPO to being absorbed by a giant through a licensing deal.
❯ AI Chip Startup Etched Raises $700M Series D, Valuation Doubles to $21B
[ORDERS FIRST] Etched has closed a $700 million Series D at a post-money valuation of $21 billion, led by quantitative trading firm Jane Street. The company sells its proprietary Sohu inference chip as a complete rack — cabinet, software, and liquid cooling included — so customers can run large-model inference after moving it into their own data center. Founded in 2022 by three Harvard dropouts, Etched has raised a cumulative $1.9 billion and holds more than $1 billion in customer contracts.
[FOUR JUMPS IN 8 MONTHS] Last December its valuation was $5 billion; in July it raised $300 million at $10.3 billion; a month later it leapt to $21 billion — more than quadrupling in eight months. The turning point is clear. Etched only emerged from stealth at the end of June, with a public track record of $800 million raised and $1 billion in orders — but zero machines delivered. What actually changed the valuation curve was last month’s delivery: the first rack went into Jane Street’s data center. After testing, Jane Street called the results “satisfactory” and immediately connected the machine to its live production load. A customer test-drives the hardware, then turns around and leads the round — in chip startups, that is about as hard a vote of confidence as it gets. The money is backing not a roadmap but a machine already working inside someone else’s data center.
[THE TWO BETS] Etched’s wager rests on two pillars. First, low-voltage inference (LVI): it pushes compute-unit operating voltage below half that of competing AI chips, buying several times the compute density per unit area. The company claims sparse trillion-parameter models can sustain more than 80% of peak throughput. Second, cluster-scale memory (CSM): a mix of HBM and SRAM plus a proprietary low-latency interconnect lets every chip in the rack share one memory pool, purpose-built for long-context and multi-trillion-parameter mixture-of-experts models. Its first chip, Sohu, hard-wires Transformer math directly into silicon; the company’s number is 500,000 tokens per second on Llama 70B, roughly 20x an eight-GPU H100 system. The A0 tape-out runs on TSMC’s N4P process. Its 400-plus engineers come from Nvidia, Google’s TPU team, Broadcom, SK Hynix, and TSMC. Among inference-chip challengers, Groq has pivoted to cloud, Cerebras mostly sells chips and cloud services, and Etched sells complete racks from day one rather than letting customers assemble their own systems. For Nvidia, the headache is that these specialized machines only compete with it for inference orders.
[WHAT CAPITAL IS BUYING] A $21 billion valuation against $1 billion in contracted orders — a contract multiple of just over 20x — sits in extremely aggressive territory for a hardware company. But the pricing logic of this round is not complicated: investors are paying for inference capacity that is already installed, not for a chip roadmap. Over the past two years, most inference-side challengers died at the same spot: great numbers on paper, but customers would not move production workloads over. Etched cleared that hurdle with a single rack inside Jane Street’s data center. The real pressure now falls on inference-chip companies still at the sample-and-white-paper stage; the reference point has shifted from benchmarks to installed systems, and the fundraising bar has risen with it. What to watch next is delivery cadence: whether $1 billion in contracts becomes revenue depends on TSMC capacity and rack yield — not on releasing another benchmark.
▪ SIGNALA chip company’s most expensive valuation jump came from a customer plugging the machine in, not from another benchmark run. Pricing power in inference hardware is shifting from performance claims to machines on the ground.
❯ AI chip-IP firm Velaura closes $110M Series A, valuation tops $1B
[POWER PLAY] Velaura AI has raised a $110M Series A at a post-money valuation above $1B, led by Seligman Ventures. Rather than building whole chips, it sells digital chip IP and a companion design platform: customers integrate its blocks into their own AI accelerators to get lower power consumption at the same compute. Its flagship product, Titan Core, targets conventional accelerator designs, with the company claiming 2x to 4x performance per watt.
[PLATFORM PIVOT] Velaura is a Silicon Valley company whose founding date has not been publicly disclosed; it previously existed mainly as an IP vendor. The change comes from the demand side: the AI data center bottleneck has shifted from compute to power. Per-rack power keeps climbing, and electricity bills plus cooling are starting to determine deployment scale. That is what turned energy efficiency from an engineering metric into a procurement criterion. Velaura only launched Titan Core as a standalone platform this year, with the fundraise following right behind. The investor roster shows who it wants to sell to: besides the lead, new backers include Capricorn Investment Group and Prosperity7 Ventures, while existing shareholders Mayfield, Maverick Silicon, MARA, Premji Invest, Samsung Catalyst Fund and StepStone Group all followed on. Samsung and crypto miner MARA respectively sit at the two ends of the spectrum — advanced process nodes and power-intensive deployments.
[SILICON PROOF] The toughest hurdle in the IP business is convincing customers you can ship at volume. Velaura’s credential: more than 30 million ASICs. Its underlying technology has already shipped at commercial scale across multiple advanced process nodes, with proven yield and reliability — not simulation data. That sets it apart from most low-power startups, which only have lab results. The team comes from Apple, Nvidia, Google, Qualcomm and Marvell, and co-founder and CEO Rajiv Khemani has spent decades in the semiconductor industry. It is betting on two markets at once: hyperscale data centers on one side, and so-called physical AI — robots, drones and autonomous systems — on the other. The latter is even more power-sensitive than data centers, because the battery is only so big. Proceeds from the round will go toward accelerating product commercialization, expanding engineering and customer teams, and co-developing with strategic customers.
[POWER PINCH] The $1B valuation on an IP company that has yet to deliver at scale is premised on energy efficiency becoming the hardest constraint in AI infrastructure. Capital is starting to pay separately for performance per watt — two years ago, that would not have gotten its own line item. Those that benefit are upstream suppliers that can push the power curve down; the pressure falls on compute operators that scale by stacking more cards and drawing more power. When power supplies can’t keep up, operators whose chips draw less power can pack more compute into the same rack. The data center capacity yardstick has to change: it used to be counted in GPUs, and from now on it will be counted in available power.
▪ SIGNALThe ceiling on compute expansion is shifting from chip supply to power supply. Energy efficiency has gone from an engineering metric to a line item on the procurement list.
❯ AI video firm Higgsfield completes $400M Series B at $5.4B valuation
[REVENUE FIRST] AI video and image generation platform Higgsfield has completed a $400 million Series B at a $5.4 billion valuation, led by growth-stage fund DST Global. Users type an idea into its web-based workspace, and the platform directly produces finished video and images: Cinema Studio, aimed at directors, handles storyboards and multi-scene final edits, while Marketing Studio, aimed at marketing teams, mass-produces ad-ready assets. Founded in San Francisco in 2023, the company has raised more than $450 million to date, with this round alone representing the majority of that sum.
[8-MONTH 4X] Back in January, Higgsfield’s Series A raised just $50 million at a $1.3 billion valuation; eight months later, the valuation hit $5.4 billion — up more than fourfold. Two things happened in between. The first was the May launch of Supercomputer, an agentic product that can automatically complete multi-scene visual production end-to-end; within three months of launch, users in that line grew 42x. The second was the opening of the enterprise front: advertising, film/TV, fashion retail, finance, and even pharma began paying. Company disclosures show annualized revenue has reached $700 million — higher than the size of this round. At least 18 institutions participated, with existing shareholders such as Accel and Menlo Ventures all adding capital, and Goldman Sachs Alternative Investments growth equity, Intel Capital, and NTT DOCOMO Ventures also on the list.
[WHY IT WINS] Higgsfield is not the company with the best models; it’s the first to turn generative tools into a production workflow. Creation tool Runway and enterprise talking-head video firm Synthesia each own a segment, but Higgsfield splits product lines directly by job function: directors use Cinema Studio for films, marketing teams use Marketing Studio for assets, and both lines share the same generation backend. What enterprises buy is schedulable production capacity. It has also pulled away on scale — over 30 million users across 238 countries and regions, 20 million generations per month or more, and 390 Fortune 500 companies using it. The ability to close enterprise deals comes from the team’s track record: CEO Alex Mashrabov’s previous company AI Factory was acquired by Snap in 2019, where he led the scaling of consumer generative visual products; CTO Yerzat Dulat runs the technology. The round’s proceeds are earmarked for R&D, global infrastructure, AI talent, and overseas markets, with compute the biggest line item: video is the most compute-intensive segment in AI, and at 20 million generations per month, costs can’t be squeezed — the gross margin on $700 million in revenue wouldn’t hold.
[BUDGET SHIFT] A $5.4 billion valuation against $700 million annualized revenue comes to a revenue multiple under 8x, cheap for today’s AI companies — provided the $700 million holds. What capital is buying in this round is retention, not generation quality: underlying model capabilities are being leveled month by month, anyone can access comparable models, and the layer that keeps collecting revenue is the one that locks workflows into enterprise processes. The substance of 390 Fortune 500 customers is that ad-asset production has been pulled in-house from outsourcing agencies. That is also the most fragile spot: Higgsfield doesn’t train its own base models. If upstream model vendors move down into marketing-asset tools, the pricing power of this middle layer will be the first to slip. Ad production firms need to recalculate — budgets are moving away from time-and-materials billing.
▪ SIGNALA company that doesn’t train its own models reached $700 million in annualized revenue by inserting generation into the daily production schedules of marketing and film. Money in visual production is flowing from capacity providers to workflow contractors.
❯ AI Finance Software Company Rillet Raises $100M Series C at $1B Valuation
[LEDGER] AI-native ERP company Rillet has closed a $100 million Series C at a $1 billion valuation, led by ICONIQ. It rewrote the corporate general ledger: transaction data streams into the ledger in real time via native integrations, and AI agents handle reconciliation and journal entries directly inside the ledger, while humans retain approval rights and every change leaves an audit trail. The company came out of stealth in 2024 and now has more than 600 customers, including publicly traded companies.
[PACE] This is Rillet’s third round of funding in fourteen months, bringing total funding to more than $200 million. The pace is this intense because demand is moving faster than product iteration — its new annual recurring revenue doubled in the past three months. It replaces the legacy systems Oracle Fusion, SAP, Workday, Microsoft Great Plains, and NetSuite, as well as the tangle of spreadsheets and plug-ins that grew up around them. Most of those systems are built on architectures from ten or twenty years ago: data lives in one place, work happens in another, and only at month-end close do the two get reconciled. AI agents can’t get inside that structure; they’re stuck outside as assistants. Rillet’s approach is to replace the ledger itself, giving agents a place to stand. Disclosed customers include Mercor, Function Health, and Temporal.
[REPLACEMENT] The selling point is the close. Traditional finance teams spend one to two weeks each month doing the month-end close. Rillet argues for a continuous close: data flows into the ledger in real time, current-period numbers are always visible, and the concentrated month-end cycle is flattened. Founder and CEO Nicolas Kopp says “finance agents need more than access to data — they need to work inside the general ledger.” He predicts that within two or three years, every company will run finance this way. The moat is switching costs. ERP is one of the hardest systems in a company to replace; once the general ledger moves over, audit, tax, and consolidated reporting all follow, making renewal nearly the default. That also explains why Sequoia, a16z, Bain Capital Ventures, Oak HC/FT, Battery Ventures, FirstMark, Scale Venture Partners, and Creandum have invested three rounds in a year — betting on a window to grab installations. The ERP layer only changes once per generation.
[SHIFT] A $1 billion valuation isn’t expensive in the ERP space — when Oracle acquired NetSuite, it was $9.3 billion. Capital is pricing whether AI can get into core systems. Over the past two years, most enterprise AI spending has gone to edge cases: customer service, documents, code completion, where a failure has limited impact. The general ledger is another story — one wrong entry leaves a mark on the audit report. Rillet’s $1 billion valuation means the market accepts that AI agents can operate inside an audited system, as long as approval and audit trails are in place. Under pressure are the companies that bolt plug-ins onto old ERPs. Their position rests on the assumption that legacy systems can’t be replaced — and that assumption is loosening.
▪ SIGNALEnterprise AI spending is moving from edge cases into audited core systems. A generational shift in the general-ledger layer locks in for a decade.
❯ Voice input company Wispr closes $280M Series B at $2 billion valuation
[SPEECH, NOT TYPING] Voice input company Wispr has closed a $280 million Series B at a $2 billion valuation, led by Menlo Ventures. Its product, Flow, is a cross-app dictation layer: on iOS, Android, and Windows, users speak into any input field and Flow converts speech directly into polished written text. The company has raised $361 million in cumulative funding, and more than 60 billion words have been written on the platform.
[TEN MONTHS LATER] The previous round was a $25 million Series A extension led by Notable Capital in November 2025, when cumulative funding stood at $81 million. In less than ten months, that figure has jumped to $361 million. In between, it elevated itself one layer beyond a dictation tool. Enterprise adoption has been rapid: the company says people at nearly every Fortune 500 company and in more than 10,000 businesses are using it. This kind of tool typically enters through employees installing it themselves, with IT procurement taking over once it sticks. Existing investors Notable Capital, NEA, Neo Ventures, 8VC, and MVP Ventures all re-upped, with Acrew, Forerunner, Goodwater, and Peak XV joining as new backers, plus a slate of athletes and celebrities.
[IN-HOUSE MODEL] Wispr also previewed Canto, its first self-developed speech recognition model — the dividing line marking the shift from calling on outside models to building in-house. The company’s numbers: in noisy real-world environments, word error rate drops from above 30% to 5%–10%, roughly a 4x improvement. The model supports multiple languages and mid-sentence language switching, and also draws on users’ own vocabulary lists and contacts. Its quality metric is zero-edit rate — the share of dictation that needs no corrections at all. Wispr expects Canto to reduce the edits required in daily use by about another 30%. The product line is also branching out: meeting notes tool Flow Notetaker is already live, and the company has set up the Wispr Advanced Interfaces Lab, led by former Amazon Alexa researcher Ariya Rastrow, focused on getting systems to understand intent and deliver results directly. Proceeds from this round are mostly earmarked for model R&D and expanding coverage.
[THE INPUT LAYER] A $2 billion valuation for a dictation tool looks absurd on its face — what capital is buying is the position of the input layer. The keyboard is the gateway to all software. Whoever stands between the input box and the application can see what users are trying to do in every scenario — that’s the conviction underpinning Wispr’s push toward understanding intent and delivering results directly. The risk is just as obvious: OS vendors already build dictation themselves, and Apple and Google could bake this capability into their systems at any time. Wispr’s room to maneuver lies in cross-platform reach and enterprise-side manageability. Voice’s positioning needs a re-rating — it’s moving from accessibility features and in-car scenarios to everyday input at the desk.
▪ SIGNALThis round isn’t a bet on transcription accuracy — it’s a bet on the input box. Whoever catches speech first sees the user’s intent first.
❯ Satellite constellation company Muon Space closes $250M Series C led by Eclipse
[PACKAGE MODEL] Satellite company Muon Space has closed an oversubscribed $250 million Series C led by Eclipse Capital. It doesn’t sell individual satellites — customers order a fully operational constellation, with mission design, satellite platform, payload, software, and in-orbit operations delivered as a package. The company calls this model Mission Foundry. Founded in 2021 in San Jose, California, it has now raised more than $386 million in cumulative equity financing, per the announcement.
[SCALE-UP] The previous rounds were a 2024 Series B and a 2025 Series B extension — and over the past year-plus, Muon has mainly been proving it can mass-produce. It launched 7 satellites in the first half of this year, bringing the total in orbit to 11 across six launches, with a 100% mission success rate. Over 50 satellites are in development, 13 of which are already on the launch manifest for the coming year. The real change is on the ground: a new San Jose factory just came online with a designed annual capacity of 500 satellites — ten times the previous level. Two programs are already running: FireSat, a global wildfire-monitoring constellation developed with Earth Fire Alliance and Google.org, named one of Time’s Best Inventions of 2025; and Vindlér 2.0, an RF data and analytics constellation for Sierra Nevada Corporation.
[FOUNDRY MODEL] Space has traditionally been built one mission at a time — every customer’s constellation meant years of design from scratch. Muon has turned that into a replicable platform, consolidating simulation, design, manufacturing, launch coordination, and in-orbit operations under one roof, and compressing delivery timelines from years to months. CEO Jonny Dyer: “Space infrastructure needs to scale like cloud infrastructure.” Customers span defense, government, and commercial — a dual-use structure that is winning orders most smoothly right now, with stable government budgets and growth on the commercial side. The investor lineup is telling: besides lead investor Eclipse, Google, Salesforce Ventures, Wellington Management, I Squared Capital, and Toyota’s Woven Capital all came in — a roster spanning cloud and software, infrastructure funds, and an automaker. One use of proceeds stands out: on-orbit AI compute. The more data a constellation collects, the less feasible it is to downlink everything; putting inference on satellites is moving from concept to engineering roadmap.
[VALUATION ANCHOR] Strictly speaking, this isn’t AI funding — but it converges on the same issue as the other rounds in this briefing: wherever data is produced, compute must follow. Capital is paying for the data sources themselves: wildfire, RF, Earth observation. These are physical-world data that model training and real-time inference cannot capture — and satellites are the only collection point. The beneficiaries are integrated companies that can build, launch, and operate satellites; the pressure falls on component suppliers covering only one segment — once full-package delivery takes hold, the middle layer’s pricing power gets squeezed out. Delivering on capacity is the only test this money faces: between a designed annual capacity of 500 and the 11 now in orbit lies a full order of magnitude.
▪ SIGNALThe space business is shifting from custom engineering to batch manufacturing. Whoever can deliver constellations in batches like server racks earns infrastructure-company valuations.
❯ Smart Ring Company Happy Health Raises $75M to Target Sleep Apnea
[DIAGNOSTIC RING] According to a company announcement, Happy Health closed a $75 million Series A round led by ARCH Venture Partners and OpenLoop. Its product is a clinical-grade smart ring, the Happy Ring, worn while sleeping to collect physiological data; AI interprets it and directly produces a diagnostic conclusion for obstructive sleep apnea. The company was founded in Austin, Texas in 2019.
[TWO CLEARANCES] Both lead investors have backed the company since its founding in 2019. The product has received two FDA clearances: routine biometric monitoring in September 2024, and home sleep testing in June 2025 — the latter making Happy Ring the first smart ring cleared for multi-night sleep testing. The nine months between them marked the step from a consumer wearable into a medical device.
[THREE NIGHTS] Traditionally, diagnosing sleep apnea requires an overnight stay at a sleep center, wired up with electrodes, and appointments are often booked weeks out. The company says wearing the ring for three nights is enough to produce results, with 98% accuracy, while also identifying other sleep issues such as insomnia. Founder and CEO Dustin Freckleton is a physician by training; he suffered a stroke at 24 and later discovered it was caused by sleep apnea. The company is built around his own medical history. That is the difference from consumer smart rings: the latter offers trend references, while Happy produces diagnostics that can go into a medical record. The new funding will advance clinical validation and expand the platform from sleep to at-home monitoring of other chronic conditions.
[HARDWARE-MEDICINE CONVERGENCE] Capital is paying for AI that can be billed to insurers. Consumer wearables have amassed plenty of physiological data over the years, but without diagnostic clearance they remain lifestyle products whose revenue ceiling is hardware margins. Once FDA clearance is obtained, the payer shifts from consumer to insurer, and the unit economics of the same ring are completely different. The sleep-center business needs revaluation — a diagnostic process that requires dedicated facilities and technicians is being replaced by a ring worn to bed.
▪ SIGNALThe watershed for wearables is whether they can bill. The moment diagnostic clearance is obtained, the payer switches from consumer to insurer.
OUTLOOK
[BILLINGS ONLY] Eight deals totaling $2.265 billion: three bets on silicon and compute, three on application-layer revenue, and two outside AI — satellites and medical devices. It looks scattered, but the logic is uniform: capital’s premium this cycle goes only to invoices already issued. Etched’s $1 billion contract, Higgsfield’s $700 million annualized revenue, Rillet’s doubled new annual fees, and Happy Ring’s two clearances are all verifiable proof; Groq, lacking such evidence, had its valuation cut in half. The exceptions are Velaura and Muon, which are selling capacity not yet realized. The beneficiaries are mid-tier companies that already have paying customers; the pressure is on peers with only demos and white papers. The next funding will continue to chase installed base and renewal numbers. The number to watch now is capacity fulfillment rate, not how fast valuations climb.