The Things You Refuse to Hand Away

Antioch

A warehouse does not reset because an engineer changed 1 line of code.

The robot still needs the same floor, lighting, sensors, and failure to show up again. Software can rerun a bad build before lunch. Physical AI often has to reconstruct the scene and hope the edge case cooperates.

That is the tax Antioch is trying to remove.

The New York company raised a $32M Series A led by Greylock Partners, with A*, Category Ventures, BoxGroup, and Icehouse Ventures participating. Saam Motamedi joined the board. Antioch says the round and its $8.5M seed bring total funding to $40.5M.

The capital will support product development, engineering hiring, deeper simulation, and adoption by more physical AI teams. The list sounds ordinary until you see the operating job underneath it.

Antioch calibrates high-fidelity simulations to customer hardware and real-world data, then runs thousands of scenarios in the cloud. A perception update, planner change, sensor configuration, or redesign can be tested against weather, terrain, and known failures before it reaches the machine.

The point is a verifier that can tell a team whether a proposed change improved the physical system, what regressed, and which failure deserves the next engineering cycle.

That gets more valuable as models improve. Faster intelligence creates more combinations to validate. If testing remains tied to field time, better models can help a team reach the same constraint sooner.

An August SEC filing identifies Alex Langshur as CEO. Co-founder Harry Mellsop worked on computer vision for Tesla Autopilot. Michael Calvey also came from Transpose, while Collin Schlager and Colton Swingle brought experience from Meta Reality Labs and Google DeepMind.

The customer evidence is early but concrete. Amazon's Ring says Antioch simulations closely matched physical results, including scenarios withheld from calibration. Antioch also integrates NVIDIA simulation tools, with Nebius providing cloud infrastructure.

Now the harder work becomes commercial. Robotics teams have spent years building test rigs, field programs, and internal simulation tools because they do not trust abstractions with production consequences. Antioch has to earn a place inside that loop by making each customer's failures repeatable enough to guide the next release.

A $32M round can buy compute, talent, and runway. The product earns its keep when one bad field event stops being a story somebody remembers and becomes a test every future build has to pass.

GenHealth.ai

Healthcare administration has a way of making skilled people serve as human adapters. One screen holds the patient record, another holds payer rules, the fax machine still has a vote, and payment waits for someone to carry the facts across all three.

That handoff is what GenHealth.ai just financed. The Boston healthcare AI company raised a $16.5M Series A led by Flare Capital Partners, with Craft Ventures and Obvious Ventures returning alongside Eniac Ventures, InHealth Ventures, Epsilon Health Investors, and ARTIS. The company reports $30M in total funding.

Ricky Sahu, founder and CEO; Ethan Siegel, co-founder and CPO; and co-founder Eric Marriott came out of 1upHealth knowing the plumbing beneath healthcare data. Their bet is operational: healthcare needs software that carries work through, not another dashboard showing where it stopped.

GenHealth.ai's agents handle intake, eligibility, prior authorization, billing, denials, and custom workflows inside EHRs, payer portals, fax, phone, and billing tools. Staff teach them existing flows; humans stay involved for judgment and exceptions.

That design sells responsibility, not novelty. A summary can be impressive and still leave a nurse, biller, or operations lead clicking through the queue. An agent earns its place when the approval moves, the claim gets filed, the denial is worked, and the patient reaches care sooner.

The early numbers deserve attention and attribution. GenHealth.ai says customers collect more than 30% more. Fierce Healthcare reports revenue quadrupled in six months and agents are projected to take more than 75M actions next year. Guidehealth says intake and prior-authorization productivity rose 4x, with nearly $1.2M in projected annual savings.

Those are company and customer reports, not audited outcomes. They explain why Flare partner Victor Lanio backed software built to complete work inside systems providers already own. In healthcare, changing the interface is easier than accepting accountability for the result.

The capital will expand engineering and go-to-market teams, extend agents into more provider and health-plan workflows, and improve the model beneath them. GenHealth.ai has grown past 20 employees since its 2023 seed.

Every new workflow brings a payer rule and exception path that must remain visible when automation meets reality. The market will learn less from the label “AI employee” than from how often the work arrives completed, documented, and ready for the person who still owns the decision.

Piston

Fuel fraud has an accomplice that fits in a wallet.

A physical fuel card can move money without knowing whether the person holding it belongs in the truck, whether that truck belongs at the pump, or whether the gallons make sense for the tank. By the time the fleet finds the mismatch, the fuel is gone and an operator is reconstructing the truth from a statement.

Vikram Sekhon and Shivam Shah knew that job from running fleets. They did not start Piston Technologies Inc to make the plastic smarter. They removed it and rebuilt authorization around the operating event.

Piston has raised a $15M Series A led by FPV Ventures, with existing investors Spark Capital and Pear VC participating. The company reports $22.5M in total capital raised.

Each Piston transaction can connect a driver, device, vehicle, station, pump, fuel type, telematics signal, and fleet policy before fuel flows. That gives the fleet a chance to stop the wrong purchase instead of documenting it beautifully after settlement.

The other side of the network matters just as much. Independent fuel merchants need commercial volume, workable point-of-sale integrations, and economics that do not disappear into layers of intermediaries. Piston negotiates site-specific rebates and routes fleet demand toward the locations those operators serve.

Piston says payment volume grew 8x year over year, its merchant network expanded 40x, and customer retention held at 98%. The platform is now deployed at more than 2,000 stations across 48 states, with integrations certified across systems covering more than 95% of U.S. merchant fuel sites. Those are company-reported numbers, but they explain why FPV Ventures is underwriting a much larger build.

The capital will support national network expansion, product development, and senior hiring. Piston is also extending its controls through Piston Guard, which evaluates transactions for anomalous behavior, and Piston Analytics Agent, which helps operators interrogate spending patterns without another afternoon lost to spreadsheets.

The hard part is no longer proving that a card can be replaced by an app. Piston has to make a two-sided network feel dependable on an overnight route, useful to the fleet manager at month-end, and profitable to the station owner every day.

That burden now follows every new region Piston enters. Coverage has to expand while each authorization stays attached to the driver, vehicle, merchant, and policy that made the network worth building in the first place.

Vheda Health

The first institutional check arrived 13 years after Vheda Health started doing the unglamorous work of getting high-risk members to answer, engage, and act.

That timing is the story behind Agora’s $47M investment. Shameet Luhar and Philip Rub founded Vheda Health in 2013 after seeing the same expensive gap from the payer side: a health plan can identify risk perfectly and still lose the outcome if the member never receives the right intervention at the right time.

Vheda built around that handoff. Its platform combines clinical and behavioral data, connected devices, an app, and human care coordination for Medicaid, Medicare, and Special Needs Plan populations.

The company says its programs now produce more than 80% average monthly engagement, a 3:1 average ROI, and more than $975M in savings for health-plan partners. It also reported 2,001% revenue growth and 4,049% membership growth since 2021. Those are Vheda’s figures, not an independent guarantee, but they explain why a first-time institutional investor would look past the usual early-stage theater.

Agora is making Vheda its inaugural investment. The San Francisco private equity firm was founded in 2026 by Neil Vangala, formerly a managing director at Eir Partners, and generally invests $30M to $60M in healthcare and life-sciences technology companies. Its stated preference is founder-owned or venture-backed businesses that need a new path to growth.

Vheda fits that thesis almost too neatly: founder-built, commercially established, deeply specialized, and still carrying the operating constraints that come with serving government-sponsored health plans. The $47M will fund national expansion, new product and analytics capabilities, and targeted acquisitions. That gives Vheda more routes to growth, along with more ways to dilute the discipline that made the company investable.

Healthcare analytics can find risk all day. The value arrives when data reaches a care team, the care team reaches a member, the member takes action, and the health plan can measure what changed. Vheda’s wager is that this loop can scale without turning the human part into a line item and the analytics into another dashboard waiting for attention.

Agora’s first deal is backing 13 years of evidence that health plans may pay for that connection. Vheda’s next chapter will be written in new markets, acquired capabilities, and the stubborn distance between identifying a member’s risk and changing the member’s trajectory.

Catalyst Power Holdings LLC

The interesting word in Catalyst Power Holdings LLC's new $15M credit facility is revolving. A single energy project can earn a loan. A revolving line asks a lender to believe the next project can be underwritten, built, operated, and followed by another.

East West Bank has provided Catalyst Power with a facility of up to $15M running through 2029. The capital is intended to finance the first phase of a commercial-scale combined heat and power, or CHP, and solar pipeline.

That accounting matters. This is debt availability, not an equity round, and the announcement does not say Catalyst Power has drawn the full $15M. It discloses the operating bet, not the pricing, covenants, project count, or generation capacity.

CHP produces electricity where a customer uses it, then captures heat that conventional generation would discard. That heat can support hot water, space heating, industrial processes, or cooling. The U.S. EPA says CHP systems typically reach 65-80% total-system efficiency. Catalyst Power says its systems operate in the 70-80% range.

The customer proposition is less technical: lower energy costs, greater budget certainty, and less dependence on the grid. Catalyst Power builds, owns, and operates assets at customer sites, then blends onsite output with market-based retail electricity.

The customer does not need to turn an energy upgrade into a capital request before the first kilowatt moves. Catalyst carries the development job, so it also needs a balance sheet and lender willing to carry it repeatedly.

Founder and CEO Gabriel Phillips has spent the company's story connecting retail-energy knowledge with distributed assets. Head of Distributed Generation Lucas Miller now has a financing instrument tied directly to that work. Existing backers DRW Holdings and BP Energy Partners remain part of the capital stack, while East West Bank enters through debt.

Catalyst Power reports more than 8,500 customers across 12 states and 41 utility territories. Those figures are company-reported, and the new release does not identify which projects the facility will fund. The next evidence will come from signed sites, installed systems, recovered heat, customer savings, uptime, and the discipline required to service the line.

A revolver can turn a pipeline into a repeatable financing motion. Catalyst Power has until 2029 to show how many operating assets can move through it without the underwriting, construction, or customer handoff losing voltage.

TAR

TAR's founders moved from creator software into a world of land, steel, batteries, backup generation, and customers counting time in megawatts.

That choice just attracted a $120M Series A for TAR, led by Spark Capital at a company-reported $1B post-money valuation. Returning investors Buckley Ventures and Align Fund participated, only 87 days after TAR announced a $27M seed.

AI data centers can secure chips, land, and customer commitments before they secure enough electricity to operate. Grid interconnection queues can run for years, while compute buyers measure delay in forgone capacity and contracts that cannot start.

TAR is trying to own that sequence. Its self-contained systems combine renewable generation and batteries with natural gas backup. It handles site selection, engineering, procurement, logistics, construction, commissioning, and operation.

Vertical integration sounds clean in a deck. In physical infrastructure, it means every delayed transformer, permit, shipment, foundation, control system, and commissioning decision lands back on the same team.

That is the bet inside Spark Capital's check. TAR says it can move faster because it removes the coordination drag between separate developers, contractors, suppliers, and operators. Its deployment automation and robotics are meant to shift more work into repeatable manufacturing and reduce field labor.

The company says it is executing a utility-scale deployment with one of the largest neoclouds, developing a dedicated project campus, and finishing TAR Terminal One in West Texas. The customer is unnamed, and TAR has not disclosed commissioned capacity, revenue, uptime, or project economics.

The market pressure is measurable. The IEA projects data-center electricity consumption rising from 485 TWh in 2025 to 950 TWh by 2030, with data centers contributing about half of U.S. electricity-demand growth through the decade.

TAR is hiring across 40+ roles in engineering, robotics, power systems, operations, supply chain, and project management. Patrice Becker and Leonhard Soenke are building the organization that has to turn investor speed into construction speed without asking customers to lower their reliability standard.

The AI infrastructure stack is moving farther into land, equipment, energy markets, and local approvals. TAR has raised enough capital to own more of that stack. Now every promised megawatt has to arrive as infrastructure, not a slide.

Fullscript

Fullscript began around a kitchen table because an integrative practitioner was spending more time managing a supplement dispensary than treating patients. Brad Dyment, Kyle Braatz, and Chris Wise turned that frustration into software. Fifteen years later, two private equity firms are preparing to buy control of what grew from it.

On September 10, 2026, Altas Partners and L Catterton agreed to acquire a majority stake in Fullscript from HGGC and Snapdragon Capital Partners. Terms are undisclosed. Closing is expected in Q4 2026, subject to approvals and customary conditions. This is a pending ownership agreement, not a new venture round.

What changed between the kitchen table and the cap table is the product’s place in care. Fullscript began by making clinician-directed supplements easier to recommend and fulfill. A $240M strategic growth investment from HGGC and Snapdragon followed in 2021.

The next moves pulled Fullscript deeper into the practitioner’s day. The 2024 Rupa Health acquisition added lab management. A 2026 Oura partnership brought wearable signals beside lab results and treatment history. University Hospitals then put Fullscript inside Epic’s medication workflow.

Journeys lets providers combine labs, provider-reviewed AI drafts, supplements, lifestyle protocols, and follow-through without handing the patient relationship to a direct-to-consumer subscription. Fullscript reports 135,000+ practitioners supporting 10 million patients annually. Those are company figures; the progression is visible.

Co-founder and CEO Kyle Braatz still leads Fullscript. Kurtis Funai, the fourth employee, is CTO. Their challenge under new majority owners runs through the product itself: make personalized care feel consumer-grade while leaving clinical judgment with the practitioner.

Altas brings a selective healthcare and services investment model. L Catterton brings consumer-health and wellness experience. They are underwriting a company where patient expectations, practitioner trust, clinical data, and commerce keep colliding.

Price, valuation, exact stake, and governance remain private while the agreement awaits closing. HGGC and Snapdragon helped Fullscript move beyond dispensing. Altas and L Catterton are proposing to own the stretch where supplements, labs, wearables, EHR context, and adherence have to behave like one useful workflow.

That clinical handoff now carries more patient data, with the practitioner still expected to make sense of it.

Epsilon Health

An imaging center can buy another AI tool and still have the same staffing problem on Monday morning. Someone has to fit it into the reading queue, the clinical handoff, and the responsibility for the final report.

That is the problem Epsilon Health chose to own. The company emerged from stealth with $27.6M led by AlleyCorp, joined by Uncork Capital, Renegade Partners, SemperVirens VC, and Jack Altman. Epsilon calls it Series A funding.

The number needs one honest footnote. Axios reported a $20M Series A one day earlier, while Renegade lists a prior seed investment. Public materials do not reconcile the difference, so $27.6M is the later first-party figure and no missing tranche is invented.

Founder and CEO Rustin Rassoli grew up around his father's imaging practice, where a delayed read was never abstract. Epsilon's answer was to become the practice instead of selling another point solution into one.

AI sits inside the work of board-certified radiologists. Production cases and physician corrections feed the system while a radiologist owns the final interpretation. Epsilon improves technology and service together, but inherits the staffing, integration, quality, and clinical-accountability work vendors usually leave to customers.

Epsilon says it has served more than 250,000 patients, processes 2,500+ imaging studies daily, and is on track to interpret roughly 1% of daily U.S. X-rays in 2026. It also handles more than half the volume for one unnamed large outpatient provider. Those are company-reported metrics, not an audited victory lap.

CMO Roi Bittane previously served as CMO and Chief of Strategy at Envision Radiology. Head of Machine Learning Arjun Karpur brings experience from DeepMind and Google Research. Their premise is that radiologists and technologists must improve one system before models matter at practice scale.

ACR has documented the pressure: imaging demand is growing faster than the radiologist workforce. The harder work is connecting an algorithm to a signed report, a waiting patient, and tomorrow's reading queue.

Epsilon will fund expansion, hiring, clinical partnerships, and infrastructure. The next proof will live in quality, turnaround time, radiologist capacity, and whether production data improves care without turning physician oversight into marketing copy.

Epsilon now has the money to test whether owning the whole clinical handoff can let scarce radiologist judgment travel farther while keeping responsibility exactly where patients need it.

Motive

Motive spent 2025 preparing to sell shares to public investors. Its next $1.3B+ will be judged one customer cohort at a time.

Motive has secured more than $1.3B in growth financing from General Catalyst's Customer Value Fund. The company also withdrew its S-1, replacing an immediate IPO timetable with a capital structure built around the economics of acquiring and expanding customers.

That distinction matters. This is not a conventional venture round, and Motive did not publish the agreement's pricing, draw schedule, or accounting treatment. General Catalyst describes CVF as a way to pre-fund sales and marketing while earning a capped return from the customer value created by that spend. If the customers do not produce revenue, General Catalyst says the company does not repay the capital from its own pocket.

Motive enters that agreement with scale. The company reports ARR above $600M and 30% year-over-year growth. ARR from customers spending more than $100K grew nearly 60%, with net revenue retention above 120%. Nearly 100,000 customers use Motive across transportation, construction, energy, field service, manufacturing, agriculture, retail, waste services, and the public sector.

Shoaib Makani and Obaid Khan founded the company in 2013, when KeepTruckin was helping fleets handle electronic logs. The name changed as the operating problem widened. Motive now connects driver safety, vehicles, equipment, maintenance, workforce workflows, and fleet spend through one platform.

The technology story belongs in the field. Motive's Physical Operations Graph connects data across workers, vehicles, equipment, and spend. Its AI models look for risks and operating events where a false positive can waste a manager's time and a missed event can cost far more. CTO Amish Babu leads that technical system, while CFO Chirag Shah now has an unusually large financing tool to divide between product invention and repeatable growth.

Pranav Singhvi, Managing Director at General Catalyst and co-head of the Customer Value Fund, is joining Motive's board. The seat makes sense. A financing structure tied to customer value turns retention, expansion, and go-to-market discipline into board-level capital questions.

Motive can now keep investing without asking an IPO window to cooperate. The next chapter will be written in customer economics: operators adopting more of the platform, staying, expanding, and generating the returns that let the financing follow Motive deeper into the physical economy.

Inspiren

A senior-living alert earns its keep in the next few minutes. Staff need the event, urgency, response, and the record that follows after the room goes quiet again.

That handoff is the business Inspiren is financing with a $70M Series C led by NewView Capital. Insight Partners, Primary Venture Partners, Scale Venture Partners, Vintage Investment Partners, Lightbank, Avenir Growth Capital, Story Ventures, and Camber Creek also participated, bringing the Brooklyn company’s disclosed funding to $225M and its reported valuation to $550M.

Founder and Chief Clinical Officer Michael Wang built the company as a former Green Beret and cardiothoracic nurse. CEO Alex Hejnosz began as an EMT before co-founding CipherHealth. CTO Dominique Simoneau-Ritchie leads the team putting edge AI into rooms without treating a resident’s life like surveillance footage.

Inspiren’s pitch is bigger than fall detection. AUGi watches for falls, bed exits, and chair exits using privacy-protected imagery. Sense covers bathrooms without cameras. eCall connects help buttons and pendants to alerts, context, and two-way voice. Mobile and web tools carry the response into documentation, care planning, staffing, and analytics.

Point solutions can tell an operator that something happened. Inspiren is trying to own what happens between detection and the next clinical or operating decision.

That distinction lands where the math gets less forgiving. CDC data show 14M U.S. adults 65 and older reported a fall in 2020. By 2030, Census projections put 1 in 5 Americans at retirement age. More residents will need support while caregivers remain spread across rooms and shifts.

NewView says Inspiren is in more than 400 U.S. communities. It reports company-supplied results of up to 68% fewer falls, 50% faster response times, and 54% fewer hospitalizations. Those are directional results, not independent clinical proof, but they explain why a growth investor sees more than a sensor business.

The capital will expand go-to-market, advance the hardware ecosystem, and deepen the models Inspiren uses to detect, predict, and support action. The commercial obligation grows with it. Operators are deciding whether one system can earn trust across residents, families, caregivers, clinical leaders, and the record that follows every alert.

The next chapter will be written inside those handoffs, where a model’s output becomes a caregiver’s decision and a caregiver’s decision becomes part of somebody’s care.

Intermezzo

Siddharth Ram and Kumar Ramanathan spent decades making financial and workforce systems scale. Then they chose one of the few business problems where being almost right lands on somebody else's bank account: global payroll.

At Intuit, Siddharth Ram was Chief Architect for Small Business and QuickBooks. Kumar Ramanathan led QuickBooks platform engineering and served as a chief architect. They later reunited at Velocity Global, where payroll complexity became the job.

That history led them to build Intermezzo around an unglamorous truth. A polished HCM interface can still depend on local providers, spreadsheets, email, approvals, and experts carrying country-specific rules in their heads. Every jurisdiction adds tax logic, deductions, filings, and deadlines.

Intermezzo has raised a $10M Seed led by Crosslink Capital. UKG Ventures, CloudPay, Character, Behind Genius Ventures, and angel investors participated. Character led an earlier pre-seed, but its amount and Intermezzo's total funding remain undisclosed.

Intermezzo offers HCM companies, employer-of-record providers, and payroll businesses one API for international payroll. It handles gross-to-net calculations, payslips, filings, compliance reporting, accounting sync, and on-demand payroll runs under the partner's brand.

The interesting part is where Intermezzo puts AI. Agents turn source material into structured payroll rules, while a native engine applies them to calculations and filings. Behind Genius Ventures describes an LLM intake layer around a deterministic computational core, plus a layer that explains the result.

Payroll is not a creative-writing exercise. Regulation can change without turning every paycheck into a probabilistic suggestion. Intermezzo says its platform can reduce a 20-day cycle to 24 hours with accuracy above 99%. Those are company-reported claims, and the execution burden now grows with the cap table.

The investor mix makes this more than a seed announcement. UKG and CloudPay live inside workforce and payroll infrastructure. CloudPay calls the relationship a strategic partnership, putting distribution, domain knowledge, and customer expectations close to a new payroll engine.

The $10M gives Intermezzo room to encode more jurisdictions, deepen the platform, and earn trust inside products that cannot treat payday like a beta test. Siddharth Ram and Kumar Ramanathan are building the layer beneath the clean payslip, where every rule change has to become the right number for a real person.

Positron AI

More than 50 racks of Positron AI’s first product are running inside Oracle Cloud Infrastructure. That deployment is carrying a much more expensive message: the company shipped, customers stayed, and investors are willing to finance the next architecture all the way to production.

Positron announced $875M in Series C financing at a $5B post-money valuation. The package includes a $375M Series C at a $3.5B pre-money valuation and a Series C-1 of up to $500M, language worth preserving when the headline number is this large.

Mitesh Agrawal is leading the commercial scale-up as CEO. Co-founders Thomas Sohmers, CTO, and Edward Kmett, Chief Scientist, built the company around a specific argument: inference economics are increasingly decided by memory capacity, usable bandwidth, power, and how quickly hardware reaches a data center.

Atlas gave that argument something better than a benchmark slide. Positron says more than 50 racks are being deployed at Oracle, with Parasail using the capacity and Jump Trading and i3d.net named as production customers. That production proof is doing real fundraising work.

The new capital funds the harder handoff. Positron plans to tape out its Asimov chip on TSMC’s N3P process by the end of 2026, bring up a 2 MW-plus engineering and emulation facility, secure LPDDR5X memory supply, and ramp Titan systems into production in the second half of 2027.

Asimov is designed around 288 GB to 2.304 TB of memory per chip. Titan will combine 4 to 8 chips for multi-trillion-parameter models, long context, video, and persistent agents. The company’s performance and efficiency targets remain projections until production systems face independent customer workloads, but the design choice is clear: stop treating memory as an accessory to compute.

New Enterprise Associates (NEA), Atreides Management, LP, Valor Equity Partners, Andra, SemiAnalysis, and Jim Clark led the financing across its 2 tranches. Forest Baskett, Gavin Baker, Thomas Jermoluk, and Dylan Patel are joining Positron’s board, putting technical, capital-markets, and operating scrutiny at the same table.

Positron now has the money to buy time, memory, engineering capacity, and a path through production. Asimov still has to tape out, Titan still has to ship, and the economics still have to hold when buyers put their own workloads on the racks. The next chapter will be measured in wafers, systems, megawatts, and customers who reorder.

Graph AI

Patient safety loses the plot when the people making clinical judgments spend their day moving evidence between systems.

Adverse-event reports arrive from channels that were never designed to share one regulatory clock. The work is repetitive. The responsibility is not.

Graph AI has raised a $13.3M Series A led by Insight Partners, with returning investor Bessemer Venture Partners participating. Bessemer led the company's $3M seed in October 2025, and the new round will support expansion across the US and Europe.

CEO and co-founder Raghavendra Parvataraju is building around a useful constraint: AI in pharmacovigilance has to move quickly without making the evidence, reasoning, reviewer, or handoff disappear. CTO and co-founder Vijay Ponukumati is turning that constraint into inspectable infrastructure.

Graph Safety starts where the mess enters. /intake validates and routes incoming safety material. /nucleus connects evidence, clinical context, reviewer decisions, and submission-ready case states. Graph AI says both are live.

Next, /report will automate aggregate safety reporting, while /signal is being developed with design partners to surface patterns across cases. Graph AI is connecting the safety lifecycle before another point solution creates another human handoff.

That architecture matters because speed alone can become a compliance problem wearing a demo smile. A model may produce an answer in seconds. A safety organization still has to reconstruct why it was produced, which evidence survived the trip, where a human intervened, and whether the result can stand up in an inspection.

Graph AI says live deployments cut case processing from more than 3 hours to under 10 minutes and operating costs by up to 66%. Those are company-reported outcomes; customer names and counts remain undisclosed. The useful signal is less movement around the expert, not less judgment in the decision.

Insight principal Richard Matus described a market split across tools and service providers. Bessemer partner Nithin Kaimal returned after the seed to finance the next stage of an AI-native challenger to a services-heavy workflow.

The Series A gives Graph AI more room. It also gives the company a harder proof obligation across new customers, jurisdictions, validation programs, and product modules. Patient-safety teams will remember the software that returned time to their experts and left the evidence intact when somebody asked how the decision got made.

KAVIA AI

Tata Elxsi took the least theatrical route into KAVIA AI's cap table: it used the product first.

The engineering and design company began working with KAVIA in July 2025 across internal and customer-facing programs. Roughly 1 year later, it invested through STEP.UP. The undisclosed check carries something most funding headlines cannot buy: operating context.

The announcement calls the investment part of KAVIA's seed financing. It does not disclose amount, stake, valuation, full syndicate, or whether the complete seed round closed. KAVIA previously reported a $1M pre-seed in September 2024. Precision matters because the relationship tells a stronger story than a padded total.

Labeeb Ismail built KAVIA AI around a problem enterprise engineering teams keep rediscovering. AI can produce code quickly. A company still has to know what the change touches, whether it preserves architectural intent, how the work was tested, and who accepted it across repositories and release gates.

KAVIA's Enterprise Knowledge Graph connects code, specifications, tests, documentation, and dependencies across repositories. Spec Builder turns intent into requirements and validation plans. CodeWiki preserves system knowledge. VS Code, CLI, Git, and customer-controlled deployment bring that context into existing workflows.

That product thesis gets more interesting beside Tata Elxsi. Its teams work across automotive, media, communications, healthcare, energy, and manufacturing. In those industries, brownfield software is embedded in products, customer obligations, compliance controls, and release schedules with memories longer than the newest model.

KAVIA's leadership matches that sale. Founder and CEO Labeeb Ismail previously led large engineering organizations at Comcast and Sky. Co-Founder and COO Anita Ganti has run product-engineering businesses across Wipro, Flex, and Texas Instruments. Their buyers can tell a fast demo from a system they can defend.

KAVIA says the financing will advance the platform, expand deployments, and increase customer control over infrastructure and models. Tata Elxsi brings engineering depth and a route into programs where reliability has a price tag.

Strategic money can become decorative when a partnership begins and ends with the announcement. This relationship came with a year of use already on the clock. Its next chapter belongs inside the codebases, approval chains, and release environments where both companies now have to make that conviction repeatable.

Mach Industries

Three months is barely enough time to install a production line. Mach Industries used it to add $600M to the same Series C and more than double its valuation.

The defense manufacturer announced a $600M continuation at a $3.7B valuation, following the $300M tranche it closed in June at $1.8B. The combined Series C now stands at $900M. Ribbit Capital, Infinite Capital, Bedrock, and Sequoia Capital are continuing the partnership.

That calendar is the story hiding inside the cap table. Private capital can move with a few signatures. A defense production system has to add people, equipment, suppliers, testing, quality controls, and facilities before a new dollar becomes a machine somebody can field.

Ethan Thornton founded Mach Industries in 2023 and remains CEO. The company is building long-range strike and counter-UAS platforms while bringing more of the machinery underneath them in-house: airframes, engines, solid rocket motors, energetics, autonomy, and manufacturing.

Mach's portfolio spans Viper, Glide, Stratos, Dart, and Pike. It is also developing Atlas with Whisper Aero under a Defense Innovation Unit contract. In June, Mach said it had manufactured more than 250 aircraft and designed more than 1,000 solid rocket motors. Those company-reported figures explain what investors are paying to accelerate.

The company has been assembling the factory around the product. Its Huntington Beach headquarters includes 115,000 square feet of engineering and manufacturing space. Mach Propulsion adds jet-engine capacity, while the May acquisition of Exquadrum brought solid rocket motors, energetic systems, and testing closer to the platform teams.

Vertical integration can shorten the handoffs between design, propulsion, testing, and production. It also gives Mach a much larger operating obligation. Five vehicle programs create five schedules, supply chains, qualification paths, and customer conversations before the portfolio reaches repeatable scale.

The September financing pushes Mach's publicly reported funding above $1B and raises the standard of evidence. Investors repriced the company from $1.8B to $3.7B in one quarter. The record will be written in production cadence, contract execution, and systems delivered.

Mach is financed for the transition from fast prototype shop to defense manufacturer with its own propulsion, energetics, and factory network. That is where the company can earn the valuation investors just moved forward on the calendar.

Heron Power

A 286,000-square-foot factory can become a cash obligation years before it becomes productive capacity.

Heron Power is living in that interval. The company secured a $60M facility from J.P. Morgan and TriplePoint Capital while converting a Morgan Hill warehouse into Heron Factory One. Mass production is targeted for late 2027.

This is debt capacity, not another equity round. Heron raised a $38M Series A in 2025 and a $140M Series B in February 2026. The facility adds lenders, repayment obligations, and a financing clock to a story previously told through venture conviction.

Heron plans to invest more than $100M in Factory One, hire more than 600 people, and design the site to produce up to 10,000 Heron Links a year, which the company says equals more than 40 GW of annual capacity.

Heron Link is the reason for that ambition. Drew Baglino and the team are building modular, software-controlled power electronics to replace several pieces of conventional grid equipment for data centers, solar, and battery storage. Heron says fewer conversion steps can shrink footprints and deployment time.

Heron says it has identified more than 40 GW of customer interest across data centers, storage, renewable generation, and other large power loads.

Interest still has to become qualified equipment, signed delivery schedules, and repeatable factory output. A company can have a sharp product thesis and a loud market signal while the factory clock keeps charging rent, payroll, equipment, and execution against a production date that has not arrived.

Heron also appointed Zach Kirkhorn to its board. The former Tesla CFO invested in Heron’s Series A and advised on financial and operational topics. Adding him as institutional credit enters the capital stack pairs Baglino’s manufacturing background with someone who has watched factory expansion collide with the balance sheet.

Public materials do not disclose the facility’s rate, maturity, covenants, collateral, lender split, or amount drawn. Those details will determine the price and flexibility of the runway. The announcement establishes capacity to borrow, not a free $60M beside the production line.

Heron now has the capital structure and boardroom built for the crossing. The work left is physical: qualify the product, convert customer interest into signed schedules, bring Factory One online, and make each production milestone arrive before the financing clock becomes the loudest machine in the building.

Buildcheck AI, Inc.

One drawing says the duct fits. Another says the fire wall stays intact. The field discovers they cannot both be right.

That gap between disciplines is where a routine design issue starts learning expensive words: RFI, change order, delay, rework. Buildcheck was created to find the contradiction while it is still sitting inside the drawings, when a correction is a revision instead of a crew waiting on an answer.

The company has raised a $12M Series A led by Telescope Partners. WND Ventures, the venture arm of DPR Construction, participated, while Uncork Capital, Salt VC, and Xfund returned. The round follows Buildcheck’s $5.9M seed by 9 months and brings company-reported total funding to $18M.

Alexander Michalatos, co-founder and CEO, came to the problem after working across construction, including hospital design-build projects. Co-founder and CTO Alex Gureiev built the proprietary computer-vision models behind the platform. Co-founder Andrei Molchynsky brings the owner and real-estate side of the work. The 3 met at Stanford after seeing the same coordination failure from different seats.

Buildcheck reports more than 150,000 sheets reviewed for over 110 paying construction organizations, with more than 4x growth since last year and customer ROI of 10x to 35x. Those are company figures. WND adds operating diligence: DPR is a customer and a builder putting the product against live design risk.

The next version of Buildcheck stretches beyond automated checks. Custom Checks converts a contractor’s QA/QC manuals into review agents. Diffs isolates meaningful changes between revisions. Code Reviews and Value Engineering look for regulatory trouble and design choices that may fail the budget before either reaches the jobsite.

That expansion matters because construction software adoption usually breaks on context. A generic model can know a code book and still miss how a contractor reviews, prices, assigns, escalates, and closes an issue. Buildcheck is trying to make the firm’s judgment repeatable without pretending the machine should make the final call.

The capital is financing a larger product, more customers, and a team spanning AI engineering and construction expertise. Buildcheck will earn the next phase in the space between “the model found something” and “the project team trusts it enough to act.” Every drawing set is another chance to make that handoff before concrete, steel, schedule, and margin make the disagreement permanent.

Verily Health Inc.

Stephen Gillett is turning Verily Health's partner list into an ownership map.

NVIDIA spent the past year working inside Verily's technical stack. CU Healthcare Innovation Fund II sits closer to the clinical institutions expected to use it. Both are now putting capital behind the same bet: healthcare AI needs a trusted path from fragmented data to decisions that can survive research and care.

Verily Health announced the investments on September 9 as an extension of its March fundraising. NVIDIA is a new investor. CU Healthcare Innovation Fund is adding to its position. The incremental amount, valuation, security type, and ownership percentages were not disclosed.

That accounting matters. Verily raised $300M in March in a round led by Series X Capital, with Alphabet, UCHealth, the University of Colorado Anschutz, and others participating. The new capital extends that round. It should not be mistaken for a second $300M raise.

The more interesting story is what each investor already knows about the work.

Verily Pre is tackling the part of healthcare AI that never looks good in a demo. Records, genomic files, clinical notes, permissions, and data-use policies arrive in different formats with different consequences. Refinery, Exchange, and Workbench standardize that data, govern access, and support model deployment.

NVIDIA already supplies part of that machinery. Verily Workbench supports NeMo, Parabricks, RAPIDS, and NVIDIA GPUs. The companies also built Forecast 1.0, a model combining genomics and electronic health records for long-term health-risk prediction. Verily says the NVIDIA stack improved performance and efficiency.

The CU side supplies a different pressure. Its healthcare fund is tied to an academic medical ecosystem where models meet researchers, clinicians, governance, and operating reality. Verily has worked with UCHealth, CU Anschutz, and RefinedScience on oncology data extraction, reporting more than 95% accuracy and 30x faster extraction in the initial project.

Founder Andrew Conrad built Verily out of Google X. Today, Chairman and CEO Stephen Gillett and CTO Scott Burke are narrowing its focus: prepare the data, govern the work, deploy the intelligence, and connect it to Verily Me.

NVIDIA can make the models move faster. The CU ecosystem can test whether the work fits healthcare. Verily now has both investors waiting at the handoff where technical capability becomes clinical responsibility.

Enigmata, Inc.

An enterprise AI pilot can die in the meeting where somebody asks where the plaintext goes.

The model may be useful. The data may be valuable. But once patient records, transactions, claims, contracts, or copyrighted archives leave their protected environment, every new copy becomes another security, compliance, and ownership problem.

Enigmata, Inc. has raised $6.5M in Seed funding led by Blockchange Ventures to commercialize Enigmata Cipher, a patent-pending cryptographic transform built to keep that copy from becoming the price of admission.

Co-founders Scott Searle, CEO, and John Brownlee, PhD, CTO, bring cryptography, secure-computing, AI, and IP experience. Peter Baer, co-founder and CMO, brings the commercial-risk background that matters when a buyer's legal team has as much influence as its technical team.

Cipher transforms records, documents, datasets, and model features into protected data that can still support training, analytics, search, retrieval, and RAG. Source values remain hidden until an approved policy allows a specific reveal.

That changes the sales conversation. A hospital may have data that could improve cancer detection. A bank may have transaction histories that could sharpen fraud models. A publisher may have an archive worth licensing. Custody, consent, and confidentiality cannot become footnotes to the demo.

Enigmata says its internal benchmarks matched the accuracy of raw-data training while completing the work 8% to 10% faster on existing enterprise hardware. Those are company-reported results, not independent proof. Selected enterprise design partners now become the place where integration, key management, audit controls, and repeatable performance meet the people paid to distrust shortcuts.

Blockchange general partner Matt Immerso is backing privacy as a condition for AI adoption. That fits a firm raised around cryptography and digital ownership. Enigmata's longer play is secure data licensing, where institutions could let models learn from protected datasets under enforceable terms without surrendering the underlying assets.

The idea is clean. Enterprise evidence is never that polite. The $6.5M gives Enigmata room to build the benchmarks, integrations, policy controls, and audit trail that turn protected computation from an architecture slide into a purchase order.

The next useful customer story will begin with a dataset that used to be too dangerous to move and end with a security team willing to let the work continue.

Ayar Labs

An AI rack planned for 2029 starts dictating Ayar Labs' spending in 2026.

Ayar Labs has added $150M to its Series E, bringing primary capital raised this year to $650M. The extension will support product development, validation, manufacturing readiness, high-volume production, and a new Bengaluru Design Center.

The useful number is not only $150M. It is the end of 2027.

Mark Wade, CEO and co-founder, told Reuters that Ayar Labs needs its optical components qualified for volume production by then so customer products can ramp in 2028 and 2029. A customer's roadmap pulls spending, hiring, testing, and supply-chain decisions years into the present.

Ayar Labs is building co-packaged optics for AI scale-up. Its TeraPHY optical engine sits close to accelerators and switches inside the package, while the SuperNova external light source supplies the laser. The goal is to move more data with less power and fewer reach constraints than copper can tolerate as AI systems spread across racks.

Vladimir Stojanovic, CTO and co-founder, helped build the research foundation behind that architecture. Now the company has to make foundry processes, advanced packaging, lasers, fiber alignment, reliability testing, and rack integration behave like one manufacturing system. Clever physics gets the meeting. Predictable qualification gets designed into somebody else's product.

That is why Wiwynn matters. The cloud infrastructure manufacturer invested earlier this year and is working with Ayar Labs on an optically connected rack-scale system. NVIDIA lists Ayar Labs among the optical interconnect partners in its NVLink Fusion ecosystem. Those relationships sit along the path from an optical engine to a deployable AI rack.

A second capital story runs beside the raise. Antero Peak Group at Artisan Partners led a separate $225M secondary purchase from early employees and investors, joined by Sequoia Global Equities, ARK Invest, and Greycroft. Reuters reported that transaction valued Ayar Labs above $5B. The secondary created liquidity for existing holders; it did not add operating cash to the company.

The Bengaluru center, overseen by VP of Silicon Engineering Sankara Venkateswaran with Arun Balachandar as site lead, adds end-to-end silicon product development beside teams in the United States and Hsinchu. By 2027, all three regions will be working against products customers expect to ship a year or two later.

The racks are still roadmaps. Their deadlines are already consuming capital.

Flow Control Holdings

The next AI rack can hold a fortune in accelerators and still lose the day to an old industrial problem: moving liquid through metal without leaks. Hoses, fittings, valves, and manifolds become part of the compute stack when heat stops being a mechanical detail.

That is the layer Blackstone is buying with Flow Control Holdings.

Funds affiliated with Blackstone Capital Partners and Blackstone Energy Transition Partners have entered a definitive agreement to acquire the Cincinnati company from Audax Private Equity. Audax will retain a minority stake, and the transaction is expected to close in Q4 2026. The price and other financial terms were not disclosed.

FCH supplies coolant hoses, stainless-steel fittings, valves, strainers, custom fabrications, and engineered assemblies used inside coolant distribution units, in-row manifolds, and secondary fluid networks serving OEMs and hyperscalers.

None of that sounds like the glamorous side of AI. It sounds like machining tolerances, material choices, welds, couplings, inspection, and production schedules. That is exactly why it matters. A model can be patched after deployment. A coolant connection earns trust before it is installed beside expensive hardware.

Scott Kerns and the FCH team spent the Audax years building toward this moment. Audax invested in the company in 2022. FCH says the next four years included new production and distribution capacity plus 10 acquisitions across data-center cooling, food, beverage, and pharmaceutical markets.

FCH took expertise developed around sanitary and high-purity flow systems and pushed it into data centers. Steel & O'Brien contributes stainless-steel machining and fabrication. Ace Sanitary contributes specialized coolant hoses. The customer buys a component, but the requirement is repeatable thermal infrastructure.

Bilal Khan and Mark Zhu framed the deal around capacity expansion and liquid-cooling demand. Blackstone has said its AI infrastructure exposure now reaches beyond data centers into power, cooling, equipment, LLMs, and neoclouds.

Capital is following AI from the rack into the industrial supply chain, where growth depends on factories, skilled labor, quality systems, and components that have to work the first time.

FCH still has to translate ownership into more capacity without giving up the reliability that made the asset valuable. As rack density climbs, that work will happen in ordinary-looking parts carrying an extraordinary amount of consequence.

Archy

Every dental appointment creates a second shift nobody booked.

After the patient leaves, somebody still has to finish the note, check coverage, submit the claim, chase the payment, answer the message, and make tomorrow's schedule hold together. That work decides whether the practice gets paid and the team gets home.

That is the operating pressure behind Archy's $50M Series C, led by JMI Equity with TCV, Entrée Capital, Bessemer Venture Partners, CRV, and Alven returning.

Jonathan Rat and Ben Kolin started Archy in 2021 after Jonathan Rat helped his dentist wife manage the technology at her practice. Practices were asking separate systems for scheduling, charting, imaging, insurance, billing, payments, and patient communication to behave like one business.

Archy built the shared platform first. Now it is financing the action layer on top of it.

Archy Revenue handles claims and collections. Archy Scribe turns appointments into clinical notes and supports hands-free perio charting. Archy Verify checks eligibility. Archy Connect manages communication and scheduling. Archy Insight turns practice data into useful answers. The company says some agents are live while others remain in development.

An AI assistant can sound smart with partial context. An agent following up on a claim or writing into a patient chart needs the right appointment, codes, history, insurance record, payment state, and escalation path before it earns the right to act.

Archy says it now serves 1,000 practices across 45 states, processes more than $300M in payments annually, and employs 130 people. Those figures are company-reported, but the progression is clear: the platform has moved from replacing server-bound software to carrying consequential work inside live practices.

The financing gives Archy room to develop that agent portfolio. It also raises the obligation around accuracy, auditability, and trust. A scheduling mistake is disruptive. A clinical or revenue-cycle mistake can become a patient, compliance, or cash-flow problem.

JMI Equity is backing a vertical-software premise: the system that already holds the workflow may be best positioned to automate it. Point tools can bring clever features. The platform sees the handoffs between them.

For dental teams, the promise is fewer evenings reconstructing work the practice already completed once. Archy's next chapter will be measured in how much of that second shift disappears without making the people responsible for the outcome lose visibility or control.