The Work Still Available

Rebuild

A restoration contractor can reach a flooded home before the water stops moving. The paperwork can still make the family wait.

Damage has to become photos, moisture readings, scope notes, line items, carrier rules, revisions, and finally an estimate that gets approved and paid. Somewhere in that handoff, urgency turns into administration.

Alex Toporek grew up around his father's restoration business and watched experienced operators lose hours to that translation. John Reim brought construction-technology experience to the same problem. They built Rebuild around a simple commercial truth: the estimate is where field evidence becomes revenue, margin, and permission to start the work.

Rebuild has now raised $13M+ in a Series A led by Asymmetric Capital Partners, with Gutter Capital, 25madison, Lightbank, and Green Egg Ventures participating. The product captures photos, moisture readings, voice notes, and job documentation, then turns that record into Xactimate-ready estimates. Rebuild says estimates can be generated in less than 2 minutes. Its voice assistant asks follow-up questions so field teams catch details before the office has to reconstruct the room from memory.

That is a more useful AI wedge than another chat box looking for a department. Restoration contractors do physical work under financial rules written by a concentrated group of insurance carriers. The estimate sits between the two worlds, which makes accuracy, completeness, and speed part of the business model rather than a software demo.

Rebuild reports that customers are cutting claim-cycle times by more than 50% and increasing reimbursements by over 10%. The company says revenue grew nearly 5x over the past year and that its software is used by operators inside Servpro, ServiceMaster, and Belfor networks. Those are company-reported results, but they explain why the investor group is funding a wider move.

The roadmap runs beyond estimating into documentation, routing, negotiation, and payments. The new capital will expand Rebuild's New York team in engineering, sales, and customer success as the company shifts from founder-led inbound demand toward larger restoration networks.

The opportunity is not to make disaster recovery sound futuristic. It is to make the administrative handoff less indifferent to the people standing in a damaged home. Rebuild now has the capital to move deeper into that handoff, where contractor cash flow, insurer approval, and a family's return home all share the same file.

Deeper Analysis 👇 https://devcuration.com/articles/rebuild-raises-13m-ai-restoration-software

Fin.com

The word global disappears quickly when a payment reaches the last mile. A stablecoin can cross a blockchain in seconds, then meet the licenses, banks, liquidity, compliance, currencies, and local rails that decide whether anybody can actually use it.

That unfinished handoff is the business behind Fin.com.

Nabeel Alamgir and Mustafa Dar have emerged from stealth with a $20M Seed round led by Expa and Garrett Camp. Coinbase Ventures, Tenet Fund, founders of Figure, Mesh founder Bam Azizi, Second Sight Ventures, and sovereign and royal family offices across the Gulf and Africa also participated.

The financing closed in August. The company announced it on September 15 after building a footprint that, according to Fin.com, was already profitable, had grown ARR more than 50x since the start of 2026, employed more than 200 people across six offices, and processed billions in payment volume across 51+ countries.

Its customers remain unnamed, but Fin.com says their platforms collectively reach 825M users. The infrastructure company does not need the consumer to know its name if the customer's product works when money crosses the border.

Fin.com combines pay-ins, payouts, virtual accounts, SWIFT, wallets, stablecoin settlement, conversion, and compliance in one layer. Its public coverage spans 30+ countries and 40+ currencies.

Instead of entering every country from the outside, Fin.com acquires licensed local operators, keeps their banking relationships and market knowledge, then connects them into a shared network. Seven acquisitions are complete. The company is targeting twelve by year-end and says it plans to pursue a bank acquisition within six months.

That is a private-equity operating move financed inside a venture-backed payments company. The $20M is financing the regulated handoffs that make a global API credible where the last mile is still stubbornly local.

Stablecoins can improve settlement without solving the whole payment. Somebody still has to convert the value, screen the transaction, source liquidity, reconcile the records, and deliver local money through institutions people already use.

Fin.com is betting that the company that finishes those jobs market by market can become the infrastructure layer customers stop thinking about. The capital now has to turn seven acquisitions into one operating experience while every new corridor brings another regulator, another bank, and another definition of done.

Deeper Analysis 👇 https://devcuration.com/articles/fin-com-raises-20m-global-stablecoin-payment-rails

Barrell Lithium

The most important customer for Barrell Lithium's Series C may be the operator the company has not selected yet.

Barrell Lithium closed an oversubscribed round funded entirely by existing investors. The amount, terms, backers, and valuation were not disclosed. What the company did put on the table is a position of more than 62,000 net mineral acres across prospective areas of the East Texas Smackover.

That acreage has been the visible return on 3 financings. Barrell Energy reported 44,000 net acres after Series A, 52,000 after Series B, and now more than 62,000 after Series C. Private capital has helped Kirk A. Barrell and the team buy a larger seat in a lithium-brine fairway before the technical work decides which seats matter.

The next handoff changes the job. Barrell Lithium plans to begin an operator-partner process in Q4 2026 covering additional leasing, drilling, facility infrastructure, and commercial Direct Lithium Extraction.

An acreage map can establish control. It cannot establish a recoverable resource, choose the extraction system, build production and injection wells, finance a processing plant, or prove commercial economics. Barrell Lithium has not published a Barrell-specific resource estimate, recovery rate, production target, or project valuation.

The Series C is financing leverage for a negotiation in which both sides bring something the other needs. Barrell Lithium brings assembled mineral rights, local land execution, geological work, and a parent company that says its portfolio businesses have completed $477M of transactions across 890,000 acres. An operator would bring the drilling program, extraction technology, infrastructure discipline, and balance sheet that turn subsurface optionality into an operating plan.

East Texas is earning attention because company and industry testing has reported high lithium concentrations and permeable reservoir rock. Barrell cites regional results of 600 to more than 800 milligrams per liter. Those numbers describe the promise of the fairway, not certified economics for Barrell's acreage.

Barrell Lithium's operator process was scheduled for Q3 in the Series B announcement. It is now slated for Q4. No reason was given, but the sequence is clear: 3 rounds built the land position, and the next partner must help build the evidence.

More acreage gives Barrell Lithium a stronger opening conversation. The operator meeting is where the geology, technology, capital, and development obligations finally sit at the same table.

Deeper Analysis 👇 https://devcuration.com/articles/barrell-lithium-closes-series-c-for-east-texas-smackover

Mithrl

Patent filings were never on Mithrl's roadmap. More than half a dozen now belong to customers that used its platform.

That is a useful way to understand the $20M Series A Mithrl announced this week. Obvious Ventures led the round, with Headline, AGI House, and several unnamed pharmaceutical executives participating. The financing follows a $4M Seed led by Bonfire Ventures in 2024.

Vivek Adarsh, Co-Founder and CEO, and Shara Balakrishnan, Co-Founder and CTO, started with a scientific backlog. Biologists could generate mountains of data, then wait weeks or months to decide which target, biomarker, or program deserved the next experiment.

Frontier models make the hypothesis pile taller. They do not automatically make the next decision defensible.

Mithrl is building the layer in between. Mithrl-1 combines a proprietary biomedical world model with agentic model routing and context orchestration inside each customer's environment. Peer-reviewed literature, public datasets, partner data, and proprietary science can meet without forcing every team into the same application.

That design matters because a useful biological answer needs receipts. A gene-expression change is not the same as protein activity, and an inference is not an observation. A plausible hypothesis can still consume months of lab work and a meaningful piece of a drug program's budget.

Mithrl says partners have made some program decisions up to 32 times faster. It also reports 16 times more primary evidence per answer and a 0.96 scientific-correctness score versus 0.60 without its knowledge system. Those are company-reported results, not an independent audit.

The patents may be the sharper signal. Customers did more than finish an analysis sooner. They found something worth protecting and owned the resulting intellectual property.

That moves Mithrl into a harder business. A repeatable platform has to emerge from custom work that respects each program's biology, data, governance, and standard of proof.

The capital will support product, engineering, go-to-market hiring, and deployments as Mithrl launches its Biomedical World Model. The company says it already serves top-10 pharma, clinical-stage biotechs, and genomics partners, though those customers remain unnamed.

Mithrl now has roughly 24 to 28 months of runway, according to Vivek Adarsh's comments to Axios. The commercial handoff is already waiting: turn unexpected discoveries into a system more scientific teams can trust with the decisions that determine which programs keep moving.

Deeper Analysis 👇 https://devcuration.com/articles/mithrl-raises-20m-biopharma-ai

38 Degrees North

A community solar subscriber may never see the project financing. The lender list decides whether the credit reaches the utility bill.

38 Degrees North has closed a $260M credit facility with Apterra Infrastructure Capital to build approximately 85 MW of community solar in New York and Illinois.

That is the clean version. The operating version is land control, permitting, interconnection, equipment, contractors, subscriber enrollment, tax structures, and a project queue that can age faster than a developer can finance it.

Apterra served as Joint Lead Arranger, Administrative Agent, and Sole Bookrunner. Stifel and National Bank of Canada were also Joint Lead Arrangers. BankUnited acted as Collateral Agent, with Farmer Mac, Amalgamated Bank, and Siemens Financial Services participating as lenders.

The facility was oversubscribed. That says something useful about institutional appetite for this borrower and this pool of projects. It does not finish the solar plants, connect them to the grid, or guarantee the savings subscribers eventually receive.

Ryan Bennett, a founder and Managing Director of 38 Degrees North, put the pressure in the right place: New York and Illinois have active community solar programs, project queues are lengthening, and permitted pipelines become more valuable when the developer has capital ready for construction.

38 Degrees North says it now has more than 100 projects and 500 MW under construction or operating, with another 250-plus MW planned for construction during the next 24 months. The company has raised more than $1.7B in capital across its platform and projects.

This $260M is debt, not a venture round. It follows an undisclosed-size HoldCo facility from EIG in August, plus earlier growth equity from S2G Investments, Kimmeridge Energy Management, and Climate Adaptive Infrastructure.

That sequence reveals the business underneath the announcement. HoldCo capital can help acquire and advance projects. Construction debt takes over when panels, equipment, contractors, and site work start consuming cash before operating assets produce revenue.

The customer experience stays simple because the capital stack is not.

Now 38 Degrees North has to turn 85 MW of financed opportunity into operating assets across two state programs, each with its own utilities, rules, timelines, and interconnection pressure. The next proof will arrive project by project, then show up quietly on somebody's bill.

Deeper Analysis 👇 https://devcuration.com/articles/38-degrees-north-260m-community-solar-credit-facility

Transient.AI

Permission is becoming a product feature in capital-markets AI.

Nasdaq Ventures has made a strategic investment in Transient.AI, joining NEXT Investors in the company's Series A. The amount, updated round total, valuation, and terms were not disclosed. That keeps the accounting clean: this is a strategic participant in an existing financing, not permission to recycle the May round as fresh money.

The more interesting detail sits outside the cap table. Transient.AI says Nasdaq is also a client. If that relationship reaches meaningful production use, Nasdaq is evaluating the company from both sides of the table: as an investor looking for strategic value and as a market operator asking whether the software can live inside regulated infrastructure.

That is a much harder test than making an AI demo look intelligent.

Transient.AI is building a control layer for agents working across research, trading, sales, portfolio management, operations, and back office. Its Declarative Agentic Framework is designed to connect legacy systems to AI while enforcing authorization, policy, sandboxing, real-time oversight, and traceability inside the institution's own environment.

FINRA's 2026 oversight report makes the buyer's anxiety unusually clear. Agents can act beyond their intended authority, touch sensitive data, create decisions that are difficult to trace, and move faster than a human review process designed for another era. Financial institutions do not need another policy PDF explaining those risks after deployment. They need controls operating while the agent is working.

Sreej Menon, Chairman and CEO, has built the company around experience inside global markets and regulatory technology. CTO Michael Ponniah carries the engineering handoff after work spanning electronic trading at Credit Suisse and Citigroup, then large-scale systems across AWS, Alexa AI, and Amazon Last Mile.

Nasdaq Ventures' own investment criteria emphasize strategic alignment, commercial traction, experienced teams, market infrastructure, and workflow technology. Transient.AI lands directly in that overlap.

The capital is expected to support enterprise expansion across North America, Europe, and Asia-Pacific, including London, Singapore, Tokyo, and Hong Kong. The check buys room. The Nasdaq relationship raises the standard of evidence.

Transient.AI now has to show that governance can keep pace with the agents it is designed to control, across the desks and systems where one unauthorized action becomes somebody's market event.

Deeper Analysis 👇 https://devcuration.com/articles/transientai-gets-nasdaq-backing-for-governed-market-ai

Factory

Five months is barely enough time for a large enterprise to finish approving a new software platform. Factory used it to move from a $1.5B valuation to $5B.

The company has raised $200M from Blackstone, Khosla Ventures, Sequoia Capital, Insight Partners, Evantic Capital, Sound Ventures, NEA, Mantis VC, and Clearlake. Factory did not name a lead investor or label the round. It did say the financing lifts total funding above $400M and will accelerate research, product, and global go-to-market.

That timing is the story behind the number.

Matan Grinberg and Eno Reyes founded Factory in 2023 around Droids, AI agents built to take on software-engineering work. The product has since widened into something more ambitious: one governed system connecting requirements, tickets, code, tests, reviews, security, deployment, and production feedback.

Enterprises already have plenty of tools that can generate code. The expensive question begins when an agent needs permission to touch real systems, switch between models, operate inside a security boundary, and leave enough evidence for an engineering leader to defend the result.

Factory is building around that handoff.

The platform can run in the cloud, on-premises, or inside fully air-gapped environments. Factory Router selects models at the task level; the company says it cut token spending by more than 60% while maintaining frontier performance. Agent Effectiveness is meant to connect AI spending with measurable value instead of another dashboard full of activity.

Factory says hundreds of thousands of developers use the platform, with Nvidia, Blackstone, Royal Bank of Canada, Palo Alto Networks, Adobe, and T-Mobile among the companies building software factories they can control, trust, and measure.

Those are company-reported claims, but they explain why investors are assigning enterprise-infrastructure economics to a category that began with coding demos.

The $200M gives Factory more room to improve the agents. It also finances the slower work around them: customer implementation, governance, workflow design, global distribution, and the accumulation of trust inside organizations where a bad software change becomes an operating event.

Factory's valuation reset happened in 5 months. Its real calendar still belongs to the enterprises deciding how much autonomy they can safely put into production, one permission, workflow, and measurable result at a time.

Deeper Analysis 👇 https://devcuration.com/articles/factory-raises-200m-at-5b-enterprise-ai-agents

TeRAM

A startup that says its first customer production is 3 years away is doing investors a favor: it is naming the calendar before the pitch can hide it.

TeRAM has emerged from stealth with $37M in Seed equity financing. Primary Venture Partners, B Capital, Hyperion, and SemiAnalysis Capital co-led the round, with Alumni Ventures and Lightscape Partners participating. The financing closed on September 10.

That is when TeRAM is targeting initial customer production for custom 3D SRAM built to integrate directly with AI compute chips. The company is still validating the technology. No public benchmark, valuation, manufacturing partner, customer name, process node, or commercial yield has been disclosed.

AI infrastructure has spent years making compute more powerful. During inference, that compute still waits for model weights and context to move through the memory system. Bandwidth, capacity, packaging, heat, and power decide how much of an expensive accelerator stays useful after the demo ends.

TeRAM wants to design those variables together. The company says it can tailor capacity, bandwidth, form factor, and packaging around a customer's architecture, placing 3D SRAM closer to the compute that needs it. That is a meaningful technical ambition. It is also a brutal manufacturing assignment.

Charlie Cheng, founder and CEO, has lived inside memory IP before, including leadership at Kilopass and Lexra. Harry Luan, founder and CTO, is a memory technologist and inventor of VLT. Founders Valery Axelrad and Bruce Bateman add modeling, device, and memory-circuit experience. TeRAM says the group brings more than 80 years of combined semiconductor-memory work.

The hiring plan reads like the real use-of-funds memo: technology development, memory architecture, SoC development, package integration, epitaxy, etch, device modeling, circuit design, CAD, and patent counsel.

Software companies can patch a weak release next week. A memory startup can spend years discovering that physics, packaging, yield, or economics disagreed with the deck.

The $37M does not certify TeRAM's technology. It funds the work required to earn the first production commitment: validate the device, integrate it with customer compute, build the team, and survive the calendar between a convincing architecture and repeatable silicon.

AI keeps compressing the time between a new model and the demand it creates. TeRAM now has 3 years to prove that memory engineering can meet that demand on a fabrication schedule investors are willing to finance.

Deeper Analysis 👇 https://devcuration.com/articles/teram-raises-37m-3d-sram-ai-memory

TypeSafe AI

Software has spent decades teaching machines to reject the wrong type. TypeSafe AI is now asking developers to confront the uncertainty inside a valid answer.

The San Francisco model lab has raised approximately $40M in Seed funding led by DCVC and emerged from two years of stealth with Jev, its first public model for machine-native, composable AI. Forbes reported a $200M valuation, citing a person familiar with the deal.

Co-founders Diogo Almeida, Sasha Sheng, and Erik Gafni are building around an uncomfortable mismatch. Models optimized to talk with people can produce brilliant text. Software does not want another persuasive paragraph each time it must classify, route, score, extract, or continue a workflow.

Jev gives developers a different object: predefined typed values with probabilities and confidence scores. TypeSafe calls it a System One Model trained with Reinforcement Learning for Calibrated Decisions.

Production software needs somewhere to put uncertainty. A low-confidence decision can move to review. A high-confidence one can cross a threshold and trigger code. The application keeps control instead of hoping a free-form answer behaves the same way twice.

Type safety is useful discipline, not magic. A correctly shaped output can still carry the wrong judgment. TypeSafe's bet is that calibrated confidence makes that risk visible enough to test, monitor, constrain, and escalate.

TypeSafe says Jev can return cited workloads in less than 100 milliseconds and be up to 100x faster and less expensive than other frontier models. It lists input pricing of $0.042 per million tokens. Those are company-reported results for structured decisions, not a universal LLM benchmark.

That narrower job may be the point. The most valuable AI call inside a claims system, security workflow, data pipeline, or customer operation may be a tiny judgment made millions of times, not a long answer that wins a demo.

DCVC is financing the model, training, serving layer, and developer interface behind that thesis. Other investors, named customers, deal terms, and a use-of-funds plan were not disclosed.

Almeida helped make models better at responding to people. TypeSafe is pursuing a different question: how should a model answer when software is listening?

The $40M gives the team room to find out. Developers will decide where Jev belongs each time they set a confidence threshold, route an exception, or leave a human attached to the call.

Deeper Analysis 👇 https://devcuration.com/articles/typesafe-ai-raises-40m-machine-native-ai

Profound

Marketing teams rarely run out of ideas. They run out of people who remember the approved language, the last campaign decision, and why the brand stopped doing something 6 months ago.

Profound just raised $180M at a $1.8B valuation around a sharper version of that problem: if AI is going to do more of the work, it needs a memory worth trusting.

Sequoia Capital and Kleiner Perkins co-led the Series D. Lightspeed Venture Partners, Khosla Ventures, Saga Ventures, Evantic, and South Park Commons participated less than 7 months after Profound's $96M Series C.

James Cadwallader, co-founder and CEO, and Dylan Babbs, co-founder and CTO, started Profound in August 2024 to show marketers how brands appear inside ChatGPT, Gemini, Perplexity, and other answer engines. Once a team saw the gap, the work of fixing it still lived somewhere else. Profound is now trying to own that handoff.

AI Marketer scans brand data and AI answers, surfaces opportunities, and deploys agents across research, content, competitive analysis, and campaigns. Context Manager supplies company knowledge, meeting transcripts, communication threads, and working patterns so the software does not meet the brand as a stranger every morning.

The company says it serves more than 1,000 enterprise brands, including Walmart, Comcast, The Estée Lauder Companies, Zoom, and Figma. More than one-third of the Fortune 100 use the platform, and Profound told TechCrunch that revenue tripled in 6 months. Those company-reported figures help explain why prior investors returned to lead a round of this size.

The financing brings Profound's total funding above $335M. It will expand AI Marketer and an applied AI lab in New York and San Francisco to evaluate frontier models, build marketing benchmarks, and post-train models for the domain.

That is the more interesting company Profound is becoming. AEO analytics tells a CMO what the machines say. An agentic workbench has to decide what deserves a response, carry context into the work, earn approval, ship it, and show whether anything changed.

Generation gets cheaper with every model release. Brand memory, judgment, permissions, and accountability do not. Profound is building where those costs gather as its software moves from describing the marketing operation to participating in it.

The enterprise decision is whether the organization can trust the context that taught the agent what the brand means, who can approve the work, and what happens when that memory changes faster than the campaign.

Deeper Analysis 👇 https://devcuration.com/articles/profound-raises-180m-ai-marketing-workbench