The First Footsteps Through the Corn

Solace Care

Twenty-five thousand covered lives can make a Pre-Seed company look established. One new country can make it a beginner again.

Solace Care raised a €2.1M Pre-Seed round, about $2.45M at the August 25 ECB rate, led by Spintop Ventures. Plug and Play, Further Than Capital, Wave Ventures, and Nordic insurance leaders and angels also participated.

Valtteri Korkiakoski, founder and CEO, and Josef Karakoca, founder and CTO, are building for the work a life-insurance payment does not finish. Before a loss, policyholders can organize wishes, documents, and instructions. After a loss, families receive guidance through banks, insurers, pension providers, tax authorities, subscriptions, and the emotional work around it.

The product is sold through insurers and brokers. That gives Solace Care distribution a consumer app would spend years buying. One partnership can place the service in front of thousands of policyholders at once.

It also gives the company a harder operating calendar. An insurer relationship must survive procurement, security review, legal scrutiny, integration, localization, and patient follow-through before the first family receives help.

Solace Care says it moved from its first policyholders in mid-2025 to more than 25,000 covered lives across Sweden, Finland, and Norway. That is company-reported reach, not audited revenue or retention, but it shows the distribution model has crossed into live use.

The investor list fits that challenge. Spintop brings early-stage software experience. Plug and Play brings an insurance network. Several backers have worked inside Nordic carriers, where they learned how trust is evaluated, budgeted, and bought.

The new capital will deepen Nordic insurer and broker partnerships, open the Netherlands and the UK, and add senior product and commercial hires. None of those countries is a copy-and-paste market. The institutions, legal steps, public registries, language, and expectations around death all move differently.

That is the real test hiding inside the round. Solace Care is not only expanding software. It is translating a sensitive service without making families feel like they have entered another workflow.

Europe is aging, insurers are digitizing, and the administration after death is still scattered across institutions that rarely coordinate. Solace Care wants to turn that work into part of the promise families already bought.

The €2.1M gives the team more room to enter new countries. Each country will decide whether 25,000 covered lives were the beginning of a repeatable system or the advantage of an early home market.

Mundo AI

A transcript can preserve every word and still delete half the conversation.

It misses the pause that turns certainty into doubt, the laugh that changes the meaning of a sentence, the glance across a room, the background noise, and the sequence of movements that tells a person what is actually happening.

Mundo AI is building for that missing half. The Vancouver-founded company raised a $20M Series A led by GreatPoint Ventures, with Y Combinator, E12 Ventures, and Next Frontier Capital participating. A previously unannounced $4M seed brings total disclosed funding to $24M.

Jason Liao, Naijide Anwaer, Garreth Lee, and Kenneth Wu founded Mundo in 2024 after studying together at the University of British Columbia. Y Combinator lists Jason Liao as Founder and CEO, Mundo as an active Winter 2025 company, and the team at 30 people.

Mundo's current frame is broader: the data layer for “perceptual intelligence.” The company develops datasets, evaluations, and applied research across audio, video, and emerging modalities for AI teams trying to make models understand real-world signals.

That includes natural speech-to-speech interaction, fine-grained video understanding, and data types where the training and evaluation playbook does not exist yet. Reasoning gets the headlines because a benchmark can print a score. Perception is messier because the useful evidence often lives between the words.

Mundo and Y Combinator say leading AI labs already use the company's datasets and evaluations. Those customers are unnamed, so treat that as company-reported traction, not a public customer ledger. Revenue, valuation, and board changes were not disclosed either.

The new capital is expected to expand research, engineering, and operations. Collecting real-world sensory data is expensive, consent and quality matter, edge cases multiply, and an evaluation is useful only when it measures the behavior a model needs.

GreatPoint Ventures is backing a company that sits below the model layer, where better intelligence may depend on how precisely the world is represented before training begins. Y Combinator, E12 Ventures, and Next Frontier Capital are joining a bet that the next multimodal advantage will be built in the data and evaluation system, not discovered after deployment.

Mundo now has to turn a wide research frontier into repeatable infrastructure. Perception touches voice, video, robotics, agents, and any system expected to operate around people. The obligation is to prove richer data produces better behavior without sanding the human context off the world it is trying to capture.

Starcloud

Starcloud has to finance launch capacity before the spacecraft that need it are ready. The $250M extension turns rocket supply into part of the capital stack, because orbital GPUs are useless on a factory floor.

Starcloud has raised a $250M extension to the $170M Series A it announced in March. Manhattan West led the new capital at a $2.3B post-money valuation, with NVIDIA, Cisco Investments, Cedar Capital, Goanna Capital and Standard Capital joining returning backers Benchmark, EQT, Soma Capital, NFX and 776.

That makes the combined Series A $420M and brings company-reported total capital raised to approximately $450M. The obligation attached to it is much larger.

Philip Johnston, Ezra Feilden and Adi Oltean have already crossed one boundary. Starcloud-1 launched in November 2025 carrying an NVIDIA H100, and the company reports that it ran Google's Gemma and trained an LLM in orbit. Modern AI compute can leave Earth and still work.

Now the company has to make that result repeatable enough to become infrastructure.

The extension is intended to expand manufacturing, advance Starcloud-3, support engineering with NVIDIA and secure launch capacity. Starcloud says it is moving into a 100,000-square-foot Washington facility. Starcloud-2 is planned for 2027, while Starcloud-3 is intended for SpaceX Starship.

Those plans put Starcloud inside a supply chain it cannot control alone. A terrestrial data-center operator can buy land, negotiate power and replace a failed server. Orbital compute adds rockets, launch integration, radiation, thermal rejection, communications and hardware replacement to the operating model.

Capital can reserve launch capacity. It cannot make a new rocket arrive on schedule.

That distinction matters because AI infrastructure on Earth is colliding with power availability, water, land and permitting. In orbit, solar energy is abundant and some data can be processed near the spacecraft that generates it, reducing downlink demand.

The trade is not free. Launch cost, spectrum, reliability and the inability to service hardware like a conventional rack all move into the economics. Starcloud has proposed a network of as many as 88,000 spacecraft, but the FCC record documents a plan, not approval and not a deployed constellation.

This round is more interesting than another large AI valuation. Manhattan West and the syndicate are financing the distance between a demonstration and an operating system for manufacturing, launch and replacement.

Starcloud-1 proved compute could run in orbit. The next spacecraft must prove customers should pay to keep enough of it there.

RunwayVC

A venture fund can buy ownership. It cannot buy permission to test a robot beside a live turbine, drilling operation, or factory line.

RunwayVC has reached a €40M first close for RunwayVC Fund II. Aker ASA remains the cornerstone investor, joined by Halliburton, Aker BP, Aker Solutions, KLP, Investinor, industrial families, and private technology and finance investors.

That LP roster is more than a fundraising slide. These organizations control industrial environments where an impressive demo meets safety review, procurement, integration, and maintenance.

Managing Partner and Co-Founder Tor Bækkelund and Senior Partner and CTO Sagar Chandna have built the firm's second vehicle around that boundary. RunwayVC invests from pre-seed through Series A across industrial AI and software, robotics, automation, and autonomous systems, then tries to bring founders closer to the operators who know what the technology must survive.

Fund I gives the model a record. RunwayVC reports 24 investments, 23 follow-ons, and 2 exits since 2022. Its portfolio companies have raised more than NOK 2B from outside investors. Those are company-reported figures, not a victory lap disguised as an audited return.

Fund II plans about 20 investments over the next 3-5 years, primarily in Norway and the Nordics, with selected deals elsewhere in Europe and the US.

The first 2 disclosed investments make the thesis tangible. Minerva Humanoids is building rugged humanoid robots for hazardous industrial work. HIVE Autonomy adds intelligence to machines that customers already own. One asks industry to trust a new worker. The other asks it to trust new judgment inside familiar equipment.

An LP who could become a customer may shorten the route to a pilot, supply better operating context, or show a founder why technically elegant software will never survive the buying process.

It can also pull attention toward the problems an existing network already understands. Proximity may sharpen judgment. Familiarity can flatter it.

RunwayVC still has to prove that access becomes repeatable commercial advantage rather than a collection of warm introductions. Its companies still have to work safely, repeatedly, and economically when a machine leaves the demo environment and enters somebody else's operation.

The €40M first close gives RunwayVC a wider industrial network and more capital to test that proposition. The next evidence will not arrive in a fund announcement. It will arrive when founders earn permission to enter live sites, survive the standards waiting there, and win customers beyond the room that made the first introduction.

Switchboard Health

Switchboard Health spent 4 years helping patients reach specialty care. Now it is buying responsibility for what happens after the referral.

Switchboard acquired Livara Health and closed an oversubscribed equity financing of more than $5M. First Trust Capital Partners, Route 66 Ventures, A1 Health Ventures, Allumia Ventures, and Martin Ventures participated. No lead, valuation, exact amount above $5M, or acquisition price was disclosed.

That missing price tag matters less than the change in job description. Switchboard started as a navigation layer. Its software helps providers and health plans match patients with specialists, automate authorizations and scheduling, track referrals, and retrieve consult notes so primary care teams do not lose the patient after the handoff.

Livara moves the handoff inside the company. It combines MSK care planning, virtual delivery, and navigation through physicians, physical therapists, and psychosocial providers. Switchboard can now help determine where a patient should go and directly deliver care for eligible MSK cases.

Derek Baird, co-founder and CEO of Switchboard, is putting a clinical operating model behind the referral infrastructure. Livara CEO Rob Cohen and founder Dr. Kamshad Raiszadeh bring a program built around value-based MSK economics, including arrangements that can assume responsibility for total cost of care.

There is evidence worth respecting and a limit worth keeping. A 2024 Validation Institute review found Livara and Carelon program participants had 43% lower MSK spend than matched nonparticipants in the post-program year, with more than 75% of the difference tied to lower surgery use. The review also warned that unmeasured differences between the groups could explain some of the result.

Switchboard says patient volume rose 500% over the past 9 months. That is company-reported growth, but it explains the deal logic: Livara gets a distribution channel inside active referral workflows, while Switchboard gains a care model with outcomes history.

This is also Switchboard's second capability acquisition in under a year. Conduce Health added predictive analytics and precision specialist matching in September 2025. Livara adds the care delivery that follows the match.

The financing gives Switchboard room to integrate those pieces, and the obligation arrives with it. Routing a patient creates software value. Delivering the care puts clinical outcomes, unit economics, and trust on the same scoreboard.

Switchboard is no longer selling a better handoff. It is building the company that has to answer for what happens next.

Keenable

The next search customer may never see a results page. It will ask a question, pull evidence from dozens of pages, decide what matters, and move before a person finishes scanning the first 10 blue links.

Keenable.ai is stepping into that change with a $26M Seed round led by Accel, with Conviction and business angels participating. Co-founder and CEO Andrey Styskin and co-founder and Chief Scientist Matthias Petri are building independent web-search infrastructure for AI labs, inference providers, and agents that need current source material inside their reasoning loops.

Google has closed its Custom Search JSON API to new customers, and Microsoft retired the legacy Bing Search APIs in 2025. Both now steer buyers toward narrower or bundled alternatives, creating room for independent infrastructure without guaranteeing who will win it.

Styskin and Petri know the cost hiding behind that choice. Styskin previously led search and ads at Yandex and later directed web infrastructure at Amazon AGI. Petri was a Principal Applied Scientist at Amazon AGI, where Keenable says he built web grounding for Alexa.

Keenable says that index already spans more than 100B documents, with p95 latency below 250ms in US East and production use at several unnamed AI labs and inference providers. The names are undisclosed, and the coverage and performance figures are company-reported. That distinction matters when the product is supposed to become the evidence layer beneath somebody else's AI. For AI builders, retrieval becomes a procurement and reliability decision, not merely a developer convenience. That is the real product boundary.

One public integration makes the mechanism easier to see. Keenable partnered with voice-AI company Gradium so an agent can retrieve live web information without breaking a conversation. It shows where search moves when the query becomes one step inside a larger action.

TechCrunch reports that Keenable has 15 engineers across the United States and Europe and plans to double headcount by year-end. The capital buys time to keep an expensive index fresh while proving that independent retrieval can compete on quality, latency, cost, and trust.

Accel and Conviction are backing a market opening created partly by the incumbents' own incentives. Keenable still has to turn a 100B-document claim into repeatable customer proof, name enough adoption to establish durability, and show that an independent index can stay independent without becoming economically fragile. Agents may consume the web differently from people. The infrastructure still has to earn every query.

Liner

Luke Jinu Kim is moving Liner from the browser tab where a student checks a source into enterprise systems where an unsupported answer becomes somebody's operating problem.

Liner raised a $36.1M Series C led by LB Investment. InterVest, Atinum Investment, and CJ Investment returned, while Korea Development Bank, KB Securities, Daishin Securities, STIC Ventures, and Helios Private Equity joined the round.

Liner started in 2015 with an almost quaint request: highlight the web. It spent the next decade turning those user signals into an AI search and research layer built around source traceability.

Foundation models compete to generate more capable answers. Liner works on the evidence between question and answer, showing which source supports a claim and where the model asks for trust it has not earned.

The consumer footprint gave Liner room to build. It reports 14M+ registered users across 220 countries and territories. Its products include Liner Scholar, Liner Write, and Liner Finance.

That matters because Liner is not asking enterprise buyers to start with a blank research demo. It is carrying habits built in the consumer product into higher-stakes workflows.

The enterprise move raises a harder standard. Liner says its system scored 95.3 on OpenAI's SimpleQA benchmark, but SimpleQA tests short, fact-seeking questions. That supports a reliability argument. It does not prove an AI agent can integrate cleanly, protect sensitive data, and keep working when the demo ends.

Some proof is moving beyond the slide deck. Liner says its search technology is integrated into Saudi-backed HUMAIN's enterprise platform. It also formed a smart-building partnership with Samsung C&T and launched an AX Division for enterprise, public-sector, and university work.

The Series C will fund R&D, infrastructure, global hiring, consumer monetization, and enterprise expansion. That is capital for the reliability layer around the product: the people, support, and integrations that determine whether traceable AI becomes usable infrastructure.

Luke Jinu Kim and co-founder Brian (Chanmin) Woo built Liner around the belief that information becomes more valuable when a person can inspect why it deserves attention. LB Investment and the rest of the syndicate are now underwriting the enterprise version of that belief.

The next milestone will not be another benchmark screenshot. It will be a company letting Liner's evidence layer sit inside a consequential workflow, then keeping it there because the answers remain inspectable when the stakes stop being academic.

Hike Medical

Custom medical devices are still losing days in the spaces between a fax, a payer rule, a phone scan, and a factory floor. Hike Medical wants one company accountable for the whole trip.

The San Francisco company disclosed $22.5M raised across its Seed and Series A financings. Saga Ventures led through Max Altman, with Indicator Ventures, Fifth Down Capital, RiverPark Ventures, Orthofeet, Sam Blond, Jerod Mayo, and others participating.

That wording matters. Hike did not announce one new $22.5M Series A, and it did not disclose the split between the rounds or a valuation. The capital record is combined. The operating obligation is current.

Aadi Bhanti grew up inside three generations of orthotics and prosthetics work. The co-founder and CEO saw a device order move through referrals, missing documentation, payer approvals, clinical evaluation, scanning, fabrication, fitting, billing, and delivery. Every handoff created another place for time and money to disappear.

Hike started with custom insoles and pulled software into the physical work. Hike Clinical lets a provider evaluate, document, scan with a phone, and order in one workflow. Hike Lab 3D prints the device in Peoria. Hike Intelligence handles referral intake and the paperwork required to get a case approved and paid.

The company says that integration reduced production from weeks to days and cut remake rates from 1 in 15 to 1 in 400. Those are Hike-reported results, not an independent audit. They still explain why the company is widening the bet.

The new capital is intended for engineering and sales hiring in San Francisco, manufacturing hiring in Peoria, and expansion beyond insoles into the broader durable medical equipment market. CTO Nik Varshney is building the technical layer while COO Jerry Tang carries the scaling problem across software and physical operations.

This is the part healthcare AI often avoids. Reading a fax is useful. Generating a note is useful. Neither gets a brace, prosthetic limb, diabetic insert, wheelchair, or cranial helmet onto a patient by itself.

Hike is betting that administrative automation becomes more valuable when it is connected to the clinical decision and the factory that has to produce the right object. That means owning more complexity, more compliance, more inventory, and more ways to disappoint a patient if the machine slips.

The next test is not whether Hike can print another insole. It is whether the same operating system can absorb new device categories without losing the quality, turnaround, and judgment that made integration worth financing.

Airway Therapeutics

The help keeping an extremely premature infant alive can also become part of the injury those lungs have to survive.

Mechanical ventilation and supplemental oxygen are necessary inside neonatal intensive care. Fragile, underdeveloped lungs can pay for that necessity with inflammation and damage that contribute to bronchopulmonary dysplasia, or BPD.

Airway Therapeutics has closed a $50M equity round to test whether zelpultide alfa can change that equation. The structure matters: $26M had already been raised through a SAFE, while $24M came through the Series E-2. Approximately 94% of existing investors participated, including Cincinnati Children's and a syndicate of large family offices that the company did not name.

This is not a fresh $50M victory lap. It is capital being assembled around a long, expensive proof schedule.

Zelpultide alfa is a recombinant version of human surfactant protein D, or SP-D. Airway engineered it to do more than help manage the physical surface of the lung. The molecule is designed to modulate excessive inflammation, improve pathogen recognition and clearance, and support surfactant homeostasis.

The thesis begins with a biological gap. Very premature infants can lack enough SP-D while needing the respiratory support that exposes fragile lungs to more stress. Airway is trying to replace that missing protection early enough to prevent BPD instead of managing the damage after it takes hold.

ClinicalTrials.gov lists ZELA as a 366-participant, randomized, double-blind, placebo-controlled Phase 2b/3 trial. The Phase 2b portion compares 4 mg/kg and 6 mg/kg doses with an air-sham control, with up to 7 intratracheal administrations while the infants remain intubated.

The primary outcome is severe enough to keep the language honest: grade 2 or grade 3 BPD or death at 36 weeks postmenstrual age.

Marc Salzberg, MD, Chairman and CEO, and Alan S. Wolk, CFO and COO, now have to move 2 clocks together. One is clinical: finish dose selection and reach the interim analysis expected by the end of Q2 2027. The other is operational: complete the chemistry, manufacturing, controls, and analytical work needed for BLA readiness.

That second clock is where promising biologics learn whether the science can survive production and become a consistent, regulator-ready product.

Airway reports favorable Phase 1b safety and tolerability, but the pivotal trial has no posted results. The $50M gives the company a defined chance to produce them.

The round buys runway. The infants, investigators, and manufacturing record will decide whether the runway leads anywhere.

Emerald AI

A fine-tuning job and an inference request can occupy the same data center without carrying the same promise to the customer. One can wait. The other may be the product.

Emerald AI built a $1.05B valuation around that difference. The company raised an oversubscribed $150M Series A to scale software that decides when eligible AI workloads can slow, pause, or move as the grid tightens, then proves the response happened.

Emerald Conductor coordinates computing demand and onsite energy against grid conditions. The utility gets a resource it can plan around. The operator gets a possible path to faster power access. The AI customer keeps service requirements that cannot disappear because a transformer is having a bad afternoon.

The technical proof is real but bounded. A peer-reviewed Phoenix test reduced power use by 25% for 3 hours on a 256-GPU cluster while maintaining defined quality-of-service requirements. Emerald AI says it has completed 5 demonstrations and reached commercial, multi-megawatt, full-data-center deployment.

That still leaves the expensive questions. Most customers are unnamed. Revenue, pricing, and contracted megawatts are undisclosed. A controlled cluster does not prove every production workload can flex, and modeled grid capacity is not booked business.

Varun Sivaram founded Emerald AI in 2024 after years across energy technology, policy, and strategy. Chief Scientist Ayse Coskun and its operating team are working where power systems and computing systems stop being separate conversations.

Energize Capital and DCVC co-led the round. NVIDIA, Siemens, Aramco Ventures, Salesforce Ventures, GE Vernova, RWE, In-Q-Tel, Radical Ventures, Energy Impact Partners, Lowercarbon Capital, and others participated. Energize Managing Partner John Tough is joining the board.

That cap table is not decoration. Chipmakers, utilities, grid-equipment companies, energy producers, and venture firms all need the same interface to work, even though they price failure differently.

The capital will fund deployments, hiring, utility programs, and standards for verified flexible load. Emerald AI says its approach could unlock more than 100GW on the existing U.S. grid, but that figure is modeled system opportunity, not capacity under contract.

The next proof arrives inside an operating agreement. A utility has to know when it can call the load. An operator has to know which workloads remain protected. A customer has to know who owns the consequences when the model meets the hardest hour of the year.

Software can move the work. Trust decides whether the grid is allowed to ask.

Hivemind Digital Group

A board seat changes the meaning of a venture check.

M&G Investments led Hivemind Digital Group's $17M strategic round and named Alex Seddon, its Head of Impact and Private Equity, to Hivemind's board. The governance attached to it tells the sharper story.

Matt Zhang founded Hivemind Capital in 2021 after 15 years in financial services, most recently running global structured-products trading at Citi. He built the firm around a conviction that digital assets could become an institutional asset class. This round pushes that argument beyond managing crypto exposure and into the infrastructure used to issue, manage, and access tokenized assets.

CPIC Investment Management (HK), ZA Bank, FalconX, and Sonic Boom Ventures joined the round. That is a useful mix: an insurance asset manager, a digital bank, institutional crypto-market infrastructure, and venture capital. It also cited unnamed board members and former executives from Man Group, Apollo, Coinbase, Citi, First Citizens, and GoldenTree.

The investor list is not decoration. Tokenization has spent years producing pilots and forecasts. Institutions still have to solve the work that survives the demo: governance, custody, compliance, liquidity, reporting, and the operating discipline that keeps markets from becoming expensive group projects.

Hivemind says it manages $1B in client assets across 6 fund strategies and 3 global offices. Those figures are company-reported. M&G plc reported £371B under management as of March 31, 2026. A $17M check gives it exposure; a board seat gives it a closer view of what works, what breaks, and where institutional standards collide with onchain ambition.

The capital will support tokenization infrastructure, investment opportunities across asset classes, international partnerships, and the financial plumbing around next-generation asset management. Hivemind has not disclosed a valuation, ownership stake, first deployment asset class, or timeline. That restraint matters because tokenization forecasts can outrun operating proof with impressive efficiency.

A Ripple and Boston Consulting Group report projected tokenized real-world assets could reach $18.9T by 2033. That is possibility, not arrival. The market will be earned with legal rights, reliable settlement, credible liquidity, and buyers who return after the pilot loses its novelty.

Hivemind now has capital from institutions that understand both sides of that divide. The next proof lives in specific assets, repeatable issuance, measurable use, and infrastructure that survives contact with the institutions it was built to serve.

Celera Semiconductor

Maverick Silicon has now put $50M behind Celera Semiconductor across 2 rounds in 13 months. The second check is financing the harder move from showing that analog design can be automated to proving that automation can support a repeatable semiconductor business.

Celera closed a $30M Series B on August 25, fully funded by Maverick Silicon. The same investor supplied the company's $20M Series A in August 2025. Concentrated conviction brings a concentrated question: can Celera turn design speed into dependable delivery?

Analog chips sit inside systems that sense, regulate, convert, or move power. Their behavior depends on device physics, layout, process variation, temperature, and expert judgment. Digital design has layers of automation. Analog development still requires specialists to work through stubborn tradeoffs.

Celera is trying to make more of that work computable. Its Nesto technology stores digital twins of analog functions, while its software flow moves from specifications through synthesis, verification, layout, and manufacturing release. The company claims 10x faster development.

In March 2025, Celera said its software designed a high-voltage step-down converter from specification to manufacturing release in days. That was meaningful, company-reported evidence. One sample does not settle the commercial argument.

The Series B follows Celera's June acquisition of SiliconGate. The Portuguese company brought teams in Lisbon and Porto plus a reported library of 900+ silicon-proven IP designs. Celera plans to expand that group while widening the power-management applications it can address.

Patrick Brockett leads Celera as CEO. Dr. Helen Song, who joined from MediaTek in 2025 to lead product design, is now CTO. Calum MacRae remains a verified founder, though the current announcement no longer identifies him as CTO.

Automation does not erase analog expertise. Celera's wager is that scarce engineering judgment can be captured, reused, and searched by software, giving specialists more throughput without pretending physics became a dropdown menu.

The $30M will fund design automation and customer projects. The next evidence will be less theatrical than “AI-designed chip”: named adoption, repeatable performance across product families, manufacturing qualification, and customers returning with the next specification.

Maverick Silicon has bought Celera more time, talent, and capacity. It has also raised the burden of proof. Fast design starts the conversation. Reliable silicon, delivered again and again, decides whether the category changes.

Natural

Natural spent its first year arguing that AI agents will become financial actors. Upper90 just financed the balance-sheet question behind that thesis.

Natural secured an up-to-$100M credit facility from Upper90 on August 19, adding debt capacity beside more than $40M in equity. That wording matters. This is not another $100M equity round, and the announcement does not say the full facility was drawn. It is potential capacity behind payments and credit products.

Kahlil Lalji, Eric Wang, and Walt Leung incorporated Natural in August 2025. The San Francisco team then announced a $9.8M Seed round, a $30M Series A led by Forerunner, and a product map stretching from wallets and transfers to cards, billing, direct payment rails, and credit.

Six products were live when Natural announced its Series A in July: Wallets, Vaults, Pay, Request, Transfer, and Connect. Credit was scheduled for Q4. An API can tell an agent how to move money. It cannot erase settlement timing, credit exposure, fraud, compliance, or the need to fund the space between an instruction and final cash.

Upper90 has built its model around that space. The firm leads with credit for technology companies with predictable revenue or collateral, then scales facilities with asset performance. For Natural, the partnership turns a forecast about autonomous commerce into a harder assignment: underwrite the activity, control the risk, and make capital available without pretending software repealed finance.

The market is giving the thesis real company. Visa is studying live agentic-payment activity. Mastercard has launched machine-payment infrastructure and run controlled agent-led transactions. Natural is betting the winner must join identity to wallets, ledgers, settlement, compliance, observability, and credit.

That obligation is still mostly ahead. Natural has not disclosed the facility's pricing, maturity, covenants, collateral, initial draw, or loss assumptions. It has not published audited transaction volume or credit performance. Its Credit product was still on the roadmap in the latest detailed launch schedule.

The facility measures conviction before it measures proof. Upper90 is giving Natural room to finance a category that is still forming, while Natural accepts the burden of showing that agent activity can become underwritable payment volume.

The clean demo is no longer the hard part. The hard part is deciding which agent can spend, whose money is at risk, what happens when the transaction fails, and whether the system can answer without making a human clean up the balance sheet later.

Mantel

A molten salt can capture carbon in a lab. A commercial project has to survive the pipe rack, steam header, procurement schedule and an operator who cannot afford a bad installation.

That distance is where Mantel's new $18M Series A extension has to earn its keep.

Constellation and Azimut Investments S.A. joined existing investors, bringing Mantel's total funding to $50M. The capital will support projects in power generation, oil and gas, and pulp and paper as the company moves into commercial deployment.

CEO Cameron Halliday began testing molten borates at MIT after other high-temperature materials degraded. In 2019, one formulation absorbed more than 95% of the CO2 through repeated cycles. Cameron Halliday co-founded Mantel with COO Danielle Colson Rapson and CTO Sean Robertson in 2022 to turn that result into industrial equipment.

The chemistry attacks a costly weakness in conventional carbon capture. A typical system consumes steam to regenerate its capture material. Mantel works at the high temperatures already present inside boilers, kilns and furnaces, then recovers heat as useful high-pressure steam.

Mantel reports 97% lower energy losses, a CO2 stream above 99.9% purity and a projected cost below $50 per tonne at scale. Those are company claims and design targets, not audited commercial outcomes. The next proof belongs inside working plants.

The D-1 project at Kruger's Wayagamack mill in Québec is rated above 2,000 tonnes of CO2 per year and 1 MW of steam. M-10, with an unnamed Canadian oil and gas producer, is designed for about 60,000 tonnes of CO2 and 150,000 tonnes of high-pressure steam annually.

Mantel is the capture partner for TerraSpark's proposed 1.6 GW energy campus in West Virginia. A project that large does not care how elegant the chemistry looked on a bench. It cares about fabrication, integration, permitting, commissioning, uptime and who carries the risk when one link slips.

That makes the investor mix more interesting than a climate-tech logo wall. Constellation understands power assets and reliability. Azimut brings capital-markets and project-finance experience. Mantel needs both because repeatable infrastructure is built with more than a better molecule.

The $18M does not settle the carbon-capture argument. It finances the part where the argument becomes measurable: install the equipment, return useful steam, capture the promised tonnes and prove the next site does not require starting over.

Carbon capture has spent decades asking industry to believe the curve. Mantel now has to make the construction schedule believe it too.

MobileX

MobileX built a wireless brand around giving customers more control over what they pay for. Now the company is taking on a capital partner with control of its own.

CONX has made an undisclosed strategic investment that gives it a controlling interest in MobileX Global. CONX chairman Charlie Ergen and CEO Jason Kiser will join the board, while Peter Adderton remains the founder and CEO named in the announcement.

That governance change matters more than the missing check size. This is not a routine minority investment stapled to a growth plan. CONX is entering the boardroom while MobileX tries to scale a model built against one of wireless's most durable habits: selling customers more data than many of them actually use.

MobileX attacks that habit with an AI Forecaster that studies usage for up to 10 days and recommends a plan. Customers can accept the recommendation or build their own talk, text, and data combination. The promise is simple enough to fit on a phone bill: software handles the complexity so the customer pays closer to actual need.

The operating model is harder. MobileX rides wholesale network access, so its edge has to come from pricing, software, distribution, and service. The company says the new capital will accelerate customer acquisition, expand its independent dealer and retail footprint, deepen the proprietary AI platform, and explore global connectivity.

It already has reach. MobileX says customers can start through its apps, website, Amazon, Walmart.com, more than 3,700 Walmart stores, and 5,000 independent dealer locations nationwide.

An app can personalize a plan. Thousands of retail doors have to explain it, activate it, support it, and make money from it without turning the offer back into the kind of confusing bundle MobileX was built to challenge.

CONX can remove capital as the easy explanation for why growth is hard. MobileX now has to show efficient customer acquisition, durable retention, productive dealers, and plan economics that survive scale.

The announcement does not disclose the investment amount, valuation, ownership split, subscriber count, revenue, or profitability. Earlier reporting put a value on the deal, but the primary announcement keeps the terms private. The honest record is control, board representation, and a larger mandate.

That makes this a clean wireless test. MobileX is betting that AI can make a mass-market service feel individual. CONX is betting that the model can become materially larger.

A company that sells control to customers now has to prove what it can build after handing control to a capital partner.

Light Links

Firouz Vafadari is asking factory and defense buyers to make an unusual trade: keep the mobility of wireless, but stop depending on radio.

That bet now has $6M behind it. Light Links, Inc. closed a Pre-Seed round led by Outlander VC, with Anorak Ventures, Output Capital, Mana Ventures, Crosscourt Ventures, and other dual-use and defense investors participating.

Light Links built Wi-OW around invisible infrared light. The network diffuses laser light across a room, letting reflections carry data when a direct path is blocked. Standard Ethernet connects it to existing infrastructure.

The company reports up to 25 Gbps of symmetrical throughput and round-trip latency measured in microseconds. It says each device receives a guaranteed transmission window instead of competing for attention and retrying after a collision. Those are company claims, not independent benchmark results, but they define the business Light Links wants to build.

A robot does not need a network that is usually fast. It needs the control message to arrive while the machine is moving, the sensor is reporting, and the next action cannot wait for a retry.

NIST has documented the problem: industrial wireless must survive interference, multipath, congestion, and demands for predictable latency. Mobility makes wireless useful. Variability keeps it from owning more of the factory.

Light Links says Wi-OW is replacing Wi-Fi in live operations for an unnamed Fortune 50 company. The customer and results are undisclosed, so this is company-reported traction, not an audited victory. Still, replacing an installed network is a harder signal than winning a demo.

Defense sharpens the argument. Radio can be detected, jammed, or restricted around sensitive ordnance. Wi-OW keeps data on light inside the space, and Light Links is developing optical identification for drones. Certification and repeatable field performance remain the proof.

Vafadari serves as founder and CEO, Heinz Willebrand as CTO, J.J. Garcia-Luna-Aceves as CSO, and Jeremy Chan-Youker as COO. Outlander financed the move from optical performance to production hardware and operational trust.

Light Links must now manufacture enough systems, avoid making every installation a custom project, and prove in live deployments that infrared behaves predictably around dust, movement, obstructions, sunlight, and maintenance crews.

Wireless won the enterprise by removing the cable. Light Links now has to prove that light can remove the uncertainty without putting the cable back.

Nevvon

A caregiver can finish every required lesson, change agencies, and be told to start over before seeing the next client. The worker moved. The proof of the work did not.

Nevvon is trying to separate those 2 things. The home care compliance and training company announced an undisclosed strategic investment from BVP Forge and launched TrainOnce, a product designed to let completed training follow caregivers across agencies.

Repeated training costs the caregiver time, gives an agency another onboarding task, and can delay the first client visit. In a workforce defined by movement, the industry's recordkeeping still behaves as if every employer is an island.

Nevvon reports more than 57M completed lessons across more than 650,000 caregivers, compliance and audit coverage in all 50 states, and more than 300% U.S. caregiver-network growth over 5 years. Those company-reported figures explain why BVP Forge is backing portability now rather than another course catalog.

James Cohen, Nevvon's co-founder and CEO, built the company from the administrative pain of operating home care. Tal Arad, co-founder and COO, and Uri Sevilla, CTO, helped turn that pain into mobile, multilingual training and integrations with the systems agencies use to stay audit-ready.

TrainOnce raises the harder question that sits beyond product design: who accepts the record? Nevvon says completed compliance training can count wherever a caregiver works next without lowering the compliance bar. Making that promise durable across agencies, payors, self-directed programs, and rules that change by state is where the real company gets built.

BVP Forge brings private-equity capital and its ForgeEdge operating program to that test. Partner Navid Oreizy framed Nevvon's position around the collision between complex regulation and the need to keep care moving. The investment amount, ownership structure, valuation, and board terms were not disclosed. Existing management will continue leading the business.

The U.S. Bureau of Labor Statistics projects 17% employment growth for home health and personal care aides from 2024 to 2034, with about 765,800 openings each year. More workers and more self-directed arrangements create more chances for a valid training record to become stranded inside the wrong system.

Nevvon now has capital to expand the platform, deepen its work with payors and self-directed care programs, and reach more caregivers across the United States. The next proof is not how many lessons the platform can serve. It is how many organizations will trust one completed lesson enough not to make the caregiver take it again.

Standard Metrics

Renewal is the ordinary vote a software customer gets. Standard Metrics has persuaded many of its customers to make a second one and finance the company behind the product.

Standard Metrics raised a $20M Series B led by 8VC, with Salesforce Ventures, Spark Capital, January Capital, and others participating. The company has now raised roughly $50M and says most of that capital came from customers.

That relationship matters because Standard Metrics is trying to become the shared operating layer between private-market investors and portfolio companies. The product has to collect inconsistent data, preserve the trail back to source documents, and turn it into reporting, benchmarks, and answers an investment team can trust. The answer still has to survive review.

Co-founder and CEO John Melas-Kyriazi started Standard Metrics in 2020 after seeing the reporting problem from the investor side: spreadsheets in different formats, KPI definitions that drifted, and quarterly updates trapped in email and PDFs.

AI makes that mess more dangerous before it makes it useful. An agent can answer faster, but speed is worthless when the underlying metric is stale, mislabeled, or impossible to audit. Standard Metrics spent its first chapter building the data foundation. This round is financing the move from collecting portfolio information to making it queryable and operational.

The company says it now supports more than 150 investment firms managing over $400B, alongside more than 10,000 portfolio companies. It also reports approximately 20X business growth since its $23.7M Series A in 2022. Those figures are company-reported, but they clarify what 8VC and the syndicate are underwriting: a network with enough structured private-company data to support AI analysis, benchmarking, and deeper portfolio intelligence.

The capital will expand the platform, improve AI document processing, deepen the AI Analyst and MCP capabilities, and broaden performance insights. One customer reportedly cut quarterly reporting from 15 days to 2 by connecting Standard Metrics data to its AI workflows. Nobody buys infrastructure because the demo knew the word agentic. They buy it because the quarter closes, the auditor stops asking for another file, and an investment committee can see what changed.

The open question is whether customer-backed conviction can become a durable standard across firms that disagree on definitions, workflows, and data ownership. The cap table says buyers want it. The next proof is whether the numbers keep agreeing under real operating pressure as the network gets much larger.

Imprint

Imprint has spent years making co-brand cards feel like software. Its financing is a reminder that the receivables still behave like finance.

Imprint secured $2B of debt capacity since April, split between $1.5B of incremental warehouse funding and a $500M AAA-rated asset-backed securitization. This is not a $2B equity round, and the announcement does not say the entire capacity has been drawn.

The warehouse piece is a map of how quickly the capital stack has widened. Bank of Nova Scotia, Royal Bank of Canada and TD Bank Group formed a syndicate for a new $1B facility. Imprint also doubled an existing line from $500M to $1B and added Citi beside Mizuho, Truist and HSBC.

Then came PRNT 2026-A, Imprint's second receivables securitization. The deal was marketed at $300M, upsized to $500M and drew $2.35B of orders. Imprint says launch coverage reached 4.7x, compared with 1.7x for its $300M debut ABS in October 2025.

That demand matters because Imprint is asking institutional capital to underwrite more than a clean cardholder app. Buyers are underwriting receivables, credit performance, servicing, data, controls and the operating discipline behind every swipe.

Founder and CEO Daragh Murphy, co-founder and chairman Gaurav Ahuja, and CTO Will Larson are building one stack across applications, servicing, rewards, payment networks and digital wallets.

The product can feel configurable. The funding cannot be casual.

Imprint partners include Booking.com, H-E-B and Shell. More programs create more receivables. Warehouse lenders finance those assets as they accumulate; securitization moves seasoned pools into capital markets. Each layer creates growth capacity and another audience that expects evidence, not product theater.

Imprint says the new structure doubled its lending partners and reduced its funding-cost margin by 23%. The company did not disclose facility pricing, advance rates, covenants, utilization or ABS spreads, so the operating proof will arrive through portfolio performance rather than the headline.

This follows Imprint's $150M Series D at a $1.2B valuation in December 2025. Equity funded platform expansion across credit, debit, secured cards, flexible financing, loyalty and automation. Debt now has a different job: carry the receivables without making growth more expensive than the loyalty it creates.

Co-brand finance is becoming a software contest at the front end and a capital-markets contest underneath. Imprint just bought more room in both directions. The next statement that matters will be written in utilization, credit quality and repeat access to institutional demand.