The Stewardship Test

Atoms

Atoms just raised $1.7B in equity to automate the part of the economy software has spent decades politely avoiding: the physical world. Andreessen Horowitz led the investment, Ben Horowitz is joining the board, and the investor group includes Bain Capital, Uber, Fifth Wall, Chemistry, A*, K5 Global, Abstract, SV Angel, and Alpha Square Group. This is not a cap table built for a clever demo. It is built for machines, facilities, deployments, and the stubborn expense of making intelligence survive contact with reality.

Founder and CEO Travis Kalanick spent roughly eight years building toward this structure. City Storage Systems and CloudKitchens became the operating foundation. Now Atoms spans food, mining, and transport.

The pitch is simple enough for a whiteboard and difficult enough to justify $1.7B in equity: treat atoms more like bits. Understand the physical world, predict what happens next, then influence the outcome through software, sensors, AI, robotics, manufacturing, and operations.

The more useful signal is that Atoms is not arriving with vision alone. Pronto, the core technology platform behind Atoms Mining, says its autonomous haulage system is being deployed across more than 100 trucks at more than a dozen operations under a global agreement with Heidelberg Materials. At one Texas quarry, the company reports hauling more than 2 million tons in fewer than eight months. Lab37 is attacking food production from another angle with Bowl Builder, an automated assembly system the company says can reduce labor costs by as much as 50%.

Those are company-reported metrics, not proof that industrial AI has been solved. The commercial question is whether Atoms can turn separate operating businesses into a repeatable platform without allowing integration complexity to overwhelm the thesis.

Uber's participation adds another layer that no spreadsheet can fully capture. The company Travis Kalanick co-founded is backing the next argument for software in the physical world. The stronger signal is strategic: platforms built around movement increasingly want exposure to the machines doing the moving.

This investment reflects growing conviction that intelligence becomes more valuable when it extends beyond software into physical operations. The harder challenge is translating multiple operating businesses into one scalable industrial platform without losing the operational discipline that made each business work independently.

The valuation, the allocation of proceeds, and the first public Atoms Transport product remain undisclosed. That is where attention belongs next. Raising $1.7B creates the opportunity to build. The market will ultimately decide whether Atoms can turn that opportunity into productive machines operating at industrial scale.

1872

1872 just raised a $15M Seed round to turn one of manufacturing's hardest problems into a software problem with very large pieces of steel attached. The financing, made through private funds advised by The O.H.I.O. Fund, arrived alongside the public opening of 1872's first factory in Cincinnati's Camp Washington neighborhood. The goal is to rebuild heavy fabrication as one closed-loop system.

CEO Dan Summers led SpaceX's Raptor Integration & Test Engineering team from zero to a 24-hour production rate. CTO Michael Grant built core flight software infrastructure for Starship and worked on console during its early flight campaigns. Head of Operations Brian Mongilio led a Merlin engine production team of more than 70 engineers and technicians.

1872 starts with CAD and manufacturing constraints, then uses Factory OS to coordinate sourcing, cost estimates, stock availability, production planning, machines, robots, and material movement. Its domain-specific AI systems are being trained around the work of welders, buyers, material planners, and logistics professionals. Path Robotics supplies adaptive welding through Obsidian, its physical AI model for real-time welding decisions. The objective is not one clever robot. It is getting the factory to behave like a single coordinated system.

That is a meaningful wedge. Complex frames, skids, enclosures, and modular infrastructure for energy, utility, and industrial customers represent high-mix, high-consequence work. Conventional automation prefers repetition. 1872 is targeting variation, scale, and the difficult coordination between a 3D model and a finished assembly. Its first facility is intended to validate the operating model, begin customer production, and target full autonomy in 2027. Those are forward-looking objectives, but they are anchored to a functioning factory.

The O.H.I.O. Fund is also making a regional investment thesis. Ohio has deep industrial talent and customers who understand what delayed steel deliveries do to a project schedule. The American Welding Society expects the United States to need 320,500 new welding professionals by 2029. Automation is unlikely to replace the craft, but it can shift skilled workers toward tasks that require judgment while machines absorb repetitive and hazardous work.

The investment is ultimately a bet that manufacturing's biggest gains will come from coordinating entire production systems rather than automating isolated tasks. If 1872 can connect engineering, procurement, fabrication, robotics, and logistics into a single operating layer, the company will be solving a much larger problem than welding alone.

Cincinnati now becomes the proving ground. The market will determine whether 1872 can compress lead times from months to weeks while delivering the precision that heavy fabrication still demands.

Elio

Elio just raised $21M to work on a problem the AI industry keeps trying to solve with more compute: machines can only reason over what their sensors capture. The round was led by Innovation Endeavors and Xora, with Kevin Weil and Scribble VC participating. Existing investors UpWest and Resolute Ventures also returned. Bizportal reports this as a Series A, bringing Elio's total funding to $29M.

Elio is not building another model trained on the same old pixels. It is rebuilding the sensing layer so AI can determine what information it needs and adjust what the optics capture in real time.

A conventional sensor captures one fixed view of the world, usually designed around what the human eye can see. Elio says its dynamic optics process light through micromirror layers before it reaches the sensor, while software continuously adjusts the system. The goal is not a prettier image. It is a more useful measurement.

Elio says a single reconfigurable module can support modes that traditionally require separate sensors, including depth, heat, fine detail, and low-light imaging. If that capability scales into production, sensing begins to behave less like fixed hardware and more like software that improves after deployment.

The applications make the thesis tangible. In microscopy, Elio says researchers can study living cells over time without destroying a sample to capture a single measurement. In semiconductor inspection, the system is designed to identify defects inside stacked layers. Robotics and defense introduce another challenge: maintaining accurate sensing while conditions continually change.

Founder and CEO Nadav Grossinger and CTO Nitay Romano have spent more than two decades working at the intersection of optics and AI. Elio says the pair founded Pebbles Interfaces, joined Meta following its acquisition, and spent seven years developing AR and VR sensing technologies. That experience does not guarantee success, but it explains why this team is tackling the physics that most software startups never attempt.

Innovation Endeavors has consistently argued that AI's next frontier extends into the physical economy, where better models provide limited value if a system cannot accurately measure the world around it. Xora brings additional deep-tech expertise, while the continued participation of UpWest and Resolute Ventures suggests existing investors see the opportunity expanding rather than narrowing.

The investment reflects a broader shift in where AI infrastructure is evolving. Chips, models, and data centers remain essential, but the systems that determine what becomes data may prove just as important. If programmable sensing reaches commercial scale, the companies controlling that conversion layer could shape what AI models know long before inference begins.

The $21M financing is only the starting point. The larger question is whether programmable sensing becomes a foundational layer of the AI stack or remains a specialized technology reserved for a handful of industries.

Paper

Paper just raised a $34M Series A led by Accel and ICONIQ. Coding agents are compressing software production while design teams are trying to keep context, judgment, and taste from becoming collateral damage.

Founder and CEO Stephen Haney is building Paper around a blunt technical choice: the canvas renders in HTML and CSS. Instead of asking designers to work in one representation, engineers in another, and agents to translate between them, Paper wants the visual workspace and the shipped product speaking the same language.

That architecture is the real story. Paper Desktop and its MCP connection let agents read from and write directly into the canvas, while teams can pull production context into design work and push decisions back toward code. The traditional handoff becomes a continuous feedback loop.

Paper says ARR increased 25x in the month following the Desktop launch, thousands of teams now pay for the product, and the company spent four consecutive months near the top of Ramp's fastest-growing companies list. Lovable, Harvey, Vercel, PostHog, Quartr, and Y Combinator are among its customers. Those are company-reported figures, but they suggest agent-native teams are actively searching for a different kind of design platform.

The investor roster fits that thesis. Accel and ICONIQ led the round, with Designer Fund, Michael Grinich, Anton Osika, and engineers and designers from Anthropic and OpenAI participating. Basecase, David Hoang, and Guillermo Rauch also returned. Paper says the financing brings its total funding to $38.5M.

The business lesson is not that AI makes design faster. Speed is already becoming a commodity. The harder problem is preserving intent while more people and more agents contribute to the product. When code can appear in minutes, the valuable platform becomes the one that helps teams compare options, retain context, and decide what is actually worth shipping.

Paper now has the capital to deepen its editor, expand its agent infrastructure, and invest further in AI-native workflows. The next challenge is proving that a connected canvas can become core operating infrastructure instead of another destination where polished mockups wait for engineering.

The broader category is about more than adding AI to design software. Mature platforms will continue shipping agent features. Paper's opportunity is to show that the underlying architecture matters just as much as the intelligence layered on top. If design, code, and agents can finally operate inside the same system, the workflow itself becomes the competitive advantage.

Trooly.AI

Trooly.AI has raised nearly $10M in Seed funding, and the interesting part is not another AI company collecting another check. It is what Trooly believes has become expensive now that machines can generate almost anything: reliable input from real people.

BlueRun Ventures led the round, with Hillhouse Venture Capital and Wang Huiwen participating. The funding will support Trooly's core product, multimodal voice-agent technology, its global respondent network, and expansion among international companies where bad decisions arrive with adult-sized invoices.

Co-founders Zhen Wang, CEO, and Hao Sun, CTO, built Trooly after working on Museland at Zulution AI. Instead of stretching long-form AI conversation technology into endless companionship, they directed it toward a narrower commercial problem: deep qualitative interviews that help product teams understand what customers actually mean.

The platform helps teams plan studies, recruit and screen real respondents, conduct long-form interviews, ask follow-up questions, and turn conversations into structured insights. Trooly says it supports more than 400 teams, 50+ languages, and a network of more than 180M potential respondents across 25 countries.

Those are company-reported figures, but the thesis is easy to understand. Traditional qualitative research can be slow, expensive, and too episodic for product decisions. Synthetic personas are faster, but a simulated customer is still a simulation wearing a useful haircut.

Trooly is making a sharper bet: real users remain in the loop while AI handles more of the planning, recruiting, interviewing, probing, and analysis. The software is not trying to replace the customer's voice. It is trying to remove the machinery that keeps companies from hearing it in time.

That is why this seed round lands differently from another broad AI for everything pitch. Trooly is attaching AI to a defined workflow, a visible budget, and a measurable business consequence. Product teams, consumer brands, game studios, and investors already pay for insight. The larger opportunity is determining whether deep qualitative research can evolve from an occasional project into permanent operating infrastructure.

The financing also reflects where application-layer AI investment continues to move. Investors are increasingly rewarding products that compress expensive workflows without eliminating the human source of value. As generative models become more capable, the companies that own trusted inputs may prove just as valuable as those producing the outputs.

The real test now is whether AI-assisted research can consistently deliver better decisions at the speed modern product teams demand. If it can, trustworthy customer understanding may become one of the most defensible layers of the enterprise AI stack.

Abstract

Abstract just raised $25M to challenge one of cybersecurity's most expensive habits: sending every scrap of telemetry into one giant system, paying to keep it hot, then waiting for detection after the data has landed. This is an argument over who controls the economics and architecture of the SOC.

Cheyenne Ventures and AVP co-led the Series A extension, with Olive Hill Ventures joining and Crosslink Capital and Rally Ventures returning. The round brings Abstract's disclosed funding to $48.5M across an $8.5M Seed, a $15M Series A, and this extension. Abstract says the financing was completed at three times its previous valuation, though the valuation itself remains undisclosed.

The old SIEM bargain was never abstract. Centralize everything, index everything, and watch costs rise with volume while analysts sort through the noise. Abstract is pulling that model apart with a composable platform that separates collection, detection, retention, and AI-enabled operations, allowing security teams to replace one layer without rebuilding the entire stack.

Its detection fabric evaluates data in motion before normalizing and routing it into OCSF, ECS, and CIM. Astro AI works across detection, triage, investigation, and response. This is not a chatbot bolted onto legacy architecture. It starts by making the underlying data cleaner, more portable, and more actionable before AI operates on top of it.

Abstract reports 380% year-over-year ARR growth, 264% net revenue retention, a tripled customer base, and 40 strategic hires. Those are company-reported figures, but they help explain why expansion-stage capital is arriving now. Investors are backing evidence that enterprises want greater control over data, storage costs, and vendor flexibility.

CEO and co-founder Colby DeRodeff brings more than two decades of SIEM experience. COO and co-founder Chris Camacho contributes the perspective of both buyer and operator. CTO and co-founder Aaron Shelmire adds deep expertise in detection engineering. Early materials also identify Ryan Clough as a founding product leader, although his current role has not been publicly confirmed.

The new capital will expand in-stream detection, extend Astro AI across security operations, and grow the go-to-market organization. Security teams do not need another product that adds cost while calling complexity progress. They need infrastructure that lets them preserve what works, replace what does not, and avoid being locked into yesterday's architecture.

The broader shift is architectural rather than incremental. As security data volumes continue to grow, the companies that separate intelligence from storage economics may gain an enduring advantage. The question is no longer whether AI belongs inside the SOC. It is whether the underlying architecture gives AI something worth reasoning over in the first place.

WATT Fuel Cell

WATT Fuel Cell just closed a new equity financing, and the most important number is not the one the company kept private. The amount, valuation, round series, and identities of the new and existing investors were not disclosed. What WATT did disclose is where the capital is going: more people, more production capacity, and more customer deployments. That makes this less of a victory-lap financing and more of an execution round.

WATT builds solid oxide fuel cell systems in Mount Pleasant, Pennsylvania. Its WATT HOME platform combines a fuel cell, battery storage, and an inverter to keep essential residential loads running during outages, using natural gas or propane infrastructure homeowners already have. Fuel goes in, electricity comes out without an engine hammering away next door. That much is clear.

The clearest test is in West Virginia. WATT and Hope Gas announced a program for 7,250 WATT HOME systems over three years, with 750 planned for 2026, 2,500 in 2027, and 4,000 in 2028. In December 2025, WATT reported its first 2 kW installation at a Hope Gas facility and initial shipments supporting residential deployments.

Climate hardware companies can spend years collecting pilot programs like participation trophies. WATT now has a disclosed deployment program large enough to force the hard questions into the open: Can manufacturing keep pace? Can installation and service scale? Can field performance translate utility interest into repeatable economics?

Founder and CEO Caine Finnerty established WATT in 2010. The company's current leadership page also identifies Sarah DeWald, Wenhua Huang, and Paul DeWald as founders with operating responsibilities across finance, fuel cell technology, and engineering. Building a fuel cell company for 16 years is not the fast version of entrepreneurship. It is the version where materials science, certification, manufacturing, utilities, and customer trust all get a vote.

WATT has also disclosed strategic backing over time from energy and backup power companies, including EQT, Generac, and Peoples Gas. Those organizations should not be interpreted as participants in this financing because WATT did not identify the current investors. Their historical involvement still helps explain the opportunity: distributed energy sits at the intersection of fuel supply, grid resilience, hardware manufacturing, and homeowner behavior.

The financing now shifts the conversation from technology validation to operational execution. Deployment commitments create expectations that prototypes never face. Success will depend less on proving the chemistry works and more on proving the company can manufacture, install, support, and scale thousands of systems with consistent reliability.

The financing answered one question by providing the capital to expand. The more important question is whether WATT can convert a 7,250-system deployment commitment into a durable residential energy business with economics that hold beyond its first large customer.

Coverwatch

Coverwatch just raised $4.5M in pre-seed funding to take on a business insurance design flaw: the people shopping your policy can make more when your premium costs more. That is a strange incentive in a market built around protection. CoFound and Restive Ventures led the round, with Kfund and Liquid 2 Ventures participating. The capital will fund product development and team expansion, but the bigger bet sits underneath: rebuilding the brokerage rather than decorating an old process with an AI button.

The San Francisco company combines licensed insurance expertise with software that assesses risk, identifies coverage gaps, compares more than 60 carrier partners, and supports the entire policy lifecycle. Its flat-fee model is not tied to premium size, giving Coverwatch a cleaner answer to one of the buyer's most important questions: whose economics are actually being optimized?

Coverwatch says clients save 20% to 40% on average while improving coverage, with tailored quotes delivered in 24 to 48 hours. Those are company-reported figures, but they clarify the thesis. AI creates value by expanding the search, reading policy documents, tracking changing exposures, and giving licensed experts more time to apply judgment.

Co-founders Miquel Llobet, CEO, Kevin Wu, COO, and Wilmer Yan, CTO, built the company around firsthand frustration. Miquel Llobet previously founded Y Combinator-backed HOA management platform Aldara, where he experienced the complexity of commercial insurance firsthand. Together, the founding team brings operating and technical depth to a market where shallow expertise often becomes an expensive mistake.

The company's first wedge was homeowner associations, where renewing last year's policy can quietly become an expensive decision. Coverwatch has since expanded into venture-backed technology companies, ecommerce businesses, and consumer brands, where risk profiles evolve much faster than an annual renewal cycle.

That expansion is the larger signal. Vertical AI is moving beyond workflow automation toward full-stack services that own both the customer relationship and the business outcome. In insurance, software cannot simply summarize a policy. It has to navigate licensing requirements, carrier relationships, coverage decisions, claims, and the financial consequences of getting those decisions wrong.

The opportunity extends beyond making brokers more efficient. If AI can continuously monitor changing business risk while incentives remain aligned with the customer rather than premium size, the brokerage itself begins to operate differently. The competitive advantage shifts from selling policies to becoming an ongoing risk management platform.

The real question is no longer whether AI can read insurance paperwork. It is whether aligned incentives, broad market access, and continuous risk intelligence can make the traditional brokerage model explain why it still deserves the account.

Workstreet

Workstreet just took strategic growth capital from Coalesce Capital. Financial terms were not disclosed, but the deal says plenty about where cybersecurity services are headed. Workstreet has built an AI-native security and compliance business around a hard operating truth. Growing companies need credible security programs before they can assemble every specialist in-house, and enterprise buyers do not lower the bar simply because a company is moving quickly.

The company combines experts with AI-supported delivery across governance, risk and compliance, privacy, vCISO services, penetration testing, vulnerability management, and more than 35 frameworks, including SOC, ISO, FedRAMP, and CMMC. The transaction announcement says Workstreet serves more than 1,000 customers.

Co-founders Romeen Sheth, CEO, Travis Good, MD, CISO, and Ryan Rich, CTO and Head of AI, are building for a market where compliance cannot wait for audit season. It has to operate continuously and withstand scrutiny from customers, regulators, and boards asking increasingly difficult questions.

That pressure is not theoretical. NIST Cybersecurity Framework 2.0 places greater emphasis on governance, SEC cybersecurity disclosure rules have raised the stakes around reporting, and FedRAMP 20x is pushing federal cloud assurance toward persistent, machine-readable evidence.

That is why this transaction reads differently from the typical AI plus services pitch. Workstreet is using software to extend expert capacity, compress repetitive work, and move security closer to the speed of the businesses it protects. The objective is to deliver more expert judgment to more customers without every engagement starting from zero.

Coalesce Capital manages more than $1.8B in regulatory assets under management and focuses on human capital-driven, technology-enabled services. That investment thesis aligns with Workstreet's strategy: mission-critical work, recurring complexity, specialized expertise, and technology that scales service delivery without pretending human judgment disappears.

The founders and management team are investing alongside Coalesce, keeping the builders aligned with the company's next phase. The transaction provides capital and operating support to expand talent, technology, and go-to-market capabilities.

The broader opportunity extends beyond compliance. As AI automates more routine work, the firms that combine software with trusted expertise may become the new default for outsourced security operations. The competitive advantage will belong to the organizations that scale judgment rather than simply reducing labor.

The next question is whether AI-native expert services remain a differentiated category or become the operating model every serious cybersecurity services firm is forced to adopt.

TruArc Partners

TruArc Partners just closed Fund V and its parallel funds with $1.2B in capital commitments. The target was $1.1B. Fund IV closed at $840.1M. In a market where limited partners have rediscovered the word selective, exceeding the target is a vote cast with actual capital.

The strategy remains focused. TruArc is not trying to own every sector with a pulse and a pitch deck. The New York private equity firm concentrates on middle-market specialty manufacturing and business services, backing companies that make, move, maintain, or quietly keep commerce functioning.

That strategy comes with defined parameters. TruArc targets businesses generating $10M to $50M of EBITDA and enterprise values between $100M and $500M. Equity investments generally range from $50M to $150M, with capacity reaching $200M through co-investment. This is capital built for control investing and operational execution.

Fund V is roughly 43% larger than its predecessor. More capital creates greater capacity for platform investments and add-on acquisitions, but it also raises the burden of proof. A larger fund makes every exception to the investment strategy more expensive.

TruArc's approach is built around sector expertise and execution. The firm works alongside management teams to drive organic growth while pursuing acquisitions that expand products, services, sales channels, and geographic reach. In fragmented markets, disciplined buy-and-build strategies can transform regional operators into businesses with durable competitive positions.

The investor base includes family offices, asset managers, funds of funds, and insurance companies across multiple geographies. TruArc said returning investors participated strongly while new institutions joined the fund. Reinet Investments disclosed a €298M commitment in May 2025, providing one visible data point within an otherwise private limited partner base.

The close comes as private equity firms face a more demanding fundraising environment. Limited partners are asking harder questions about liquidity, specialization, and whether an operating strategy can consistently translate into portfolio performance. TruArc's answer proved compelling enough to secure another allocation.

Co-Managing Partners Ogden Phipps II, John Pless, and Alan Mantel lead the firm. TruArc says its senior investment team has led approximately $3.4B of private equity investments across 24 platform companies. Lazard served as exclusive global placement agent, and Davis Polk & Wardwell LLP acted as legal counsel.

The next phase will determine whether a larger fund strengthens the strategy or stretches it. More capital creates more opportunities, but it also increases the pressure to maintain underwriting discipline, resist style drift, and continue finding businesses where operational improvement matters more than financial engineering.

Fund V now has to translate specialization into differentiated sourcing, operational judgment into measurable value creation, and a larger capital base into stronger businesses rather than simply more transactions. That is where the real work begins.