The Weight of Two Snaps

CVC Secondary Partners

Private equity asks institutions to commit capital for a decade. The industry is now raising $10B vehicles to make that promise more liquid.

CVC Secondary Partners has closed Secondary Opportunities Fund VI at $10B in aggregate commitments. The final close beat the $7B target cited in CVC's 2025 annual report and nearly doubled the $5.8B raised for SOF V in 2023. SOF IV closed at $2.7B in 2019. That is not a gentle growth curve. It is a market installing a larger pressure-release system.

More than 200 returning and new LPs committed to SOF VI, with CVC saying roughly 50% of the capital came from investors new to the franchise. The important signal is the behavior: a broad institutional base wants dedicated exposure to the machinery that lets private-market owners adjust portfolios before every underlying fund reaches its natural exit.

SOF VI will buy LP fund portfolios and back GP-led transactions, focusing on mid-market buyout funds managed by established sponsors. Those are two different forms of the same negotiation. An LP may need liquidity, pacing relief, or a cleaner portfolio. A GP may need more time and a new ownership vehicle for an asset it is not ready to sell. The secondary buyer has to decide whether the gap is temporary, structural, or merely well-presented.

Rob Lucas, CEO of CVC, described secondaries as an important and growing part of the firm's platform. Carlo Pirzio-Biroli, Managing Partner and Head of CVC Secondary Partners, framed the close around becoming the partner of choice for LPs and GPs. Managing Partner and CIO Charles Smith brings operating continuity from the predecessor SOF program built at Deutsche Asset Management before the Glendower spinout and CVC acquisition.

The timing is earned. Lazard estimates the secondary market reached $233B in 2025, almost evenly split between GP-led and LP-led volume. Evercore estimates $226B; Jefferies estimates $240B. Secondaries have moved from the edge of private equity into its operating core.

That scale changes the job. CVC now has more capacity to solve liquidity problems, but every additional billion raises the cost of mediocre selection. A mid-market strategy cannot let fund size quietly turn it into an index of whatever needs a buyer.

SOF VI will be measured in the portfolios it chooses, the prices it refuses, and the continuation vehicles where patience becomes worth more than a clean exit. The market has supplied the capital. Discipline now has to survive the appetite that created it.

Wafer

The expensive part of AI inference is easy to misread. The deeper constraint is the small number of engineers who know how to make each model behave on each accelerator without turning production into an experiment.

Wafer is building around that constraint. Co-founders Emilio Andere, CEO, and Steven Arellano have raised a $40M Series A co-led by Marathon Management Partners and Chemistry. Wing Venture Capital, AMD Ventures, and Outset Capital participated, while Fifty Years and Y Combinator returned. The September 1 financing follows a $4M Seed announced in April, bringing disclosed funding to $44M.

Wafer’s premise is that inference should keep improving after deployment. Its software learns a workload’s traffic patterns, then searches across the model, serving engine, kernels, and hardware. Batching, caching, quantization, scheduling, and chip selection become a continuous optimization loop.

The accelerator market is widening faster than most software stacks can absorb. NVIDIA’s CUDA ecosystem remains formidable, but AMD, Google TPUs, AWS Trainium, Cerebras, and custom silicon have made the menu longer. Choice matters only when a team can translate it into throughput, latency, reliability, and cost per token for its own workload.

Investor Wing reports that Wafer reached approximately $8M in annualized revenue in roughly 3 months. Wing also says its agents tune more than 100 serving parameters and compress work that can take days or weeks into hours. Those are investor-reported figures, not independent audits, but they explain why a small San Francisco team attracted both financial investors and AMD Ventures.

The Series A reveals the commercial job hiding inside the technical one. Wafer must stay credible across silicon ecosystems while proving its optimizations survive real traffic, changing models, and production reliability requirements. The customer is not buying a prettier benchmark. The customer is buying access to more of the hardware market without hiring a rare performance team for every deployment.

Andere and Arellano met at the University of Chicago, then brought experience from Argonne, Google, and Two Sigma into a company focused on intelligence per watt. The capital will help automate more of the loop and expand the team.

As inference demand grows, the winning layer may be the one that keeps asking a question after the endpoint goes live: can this workload run better somewhere else, on different silicon, with the economics recalculated again?

Félix Pago

Financial inclusion often loses its nerve at the interface.

Banks ask the customer to learn a new system. Félix built around a conversation millions of Latino immigrants already use.

The company just secured $200 million in combined financing: an $87 million Series C led by Andreessen Horowitz and a $113 million credit facility from General Catalyst's Customer Value Fund.

That split tells the real story.

The equity can fund AI, engineering, and financial infrastructure. The credit facility can support lending as Félix expands Send Now Pay Later.

One pool builds the relationship. The other carries the obligation created when that relationship becomes credit.

Félix started with a beautifully specific job: help people in the United States send money to family across Latin America through WhatsApp.

No separate app to learn. No unfamiliar flow between the family message and the financial action.

The company says it has processed more than $8 billion for more than 6 million people across 11 Latin American markets. That is not only transaction volume. It is evidence of a repeated behavior built around urgency, family, and trust.

Now Félix wants to extend that behavior into mobile top-ups, lending, savings, and AI-guided financial services.

This is where the opportunity gets more interesting, and where the responsibility gets heavier.

Remittance moves money that already exists. Credit creates a promise about money that arrives later. A conversational interface can reduce friction, but it cannot make underwriting, servicing, or collection simple.

The strongest version of this company will not be the one that turns WhatsApp into a menu of financial products. It will be the one that understands why the customer opened the conversation in the first place.

Félix has an unusually strong distribution wedge because the interface is already woven into daily life. The next question is whether it can preserve clarity as the financial relationship deepens.

Fees must remain legible. Credit eligibility cannot feel implied. Repayment terms have to survive the friendliness of chat. Savings needs a reason to exist beyond adjacency.

That is the standard when trust becomes infrastructure.

The headline is $200 million. The more important signal is that a remittance company has earned the right to be asked the next financial question.

What Félix builds in response may show whether the next financial platform is downloaded, or simply messaged.

Fambot

A family calendar only works after somebody has found the field trip form, decoded the 12-page newsletter, noticed the pickup change, and remembered which WhatsApp group mentioned soccer cleats. The event is the last step. The invisible work arrives first.

Fambot is building for that translation layer. Co-founders David Reich, CEO, Greg Karlin, CTO, and Jason Morrow launched an AI chief of staff for families with a $3.5M pre-seed co-led by NextView Ventures and Baukunst. Correlation Ventures, Karman Ventures, and Founders Network joined.

The company connects to email, calendars, and WhatsApp, then turns family information into a daily plan delivered by text, email, app, or web. Parents can ask questions, assign tasks, add events, and coordinate through a Shared Family Account instead of moving every obligation between inboxes and calendars.

Most family organizers begin after a parent has done the administrative labor. A calendar stores the appointment. It does not read the school email, extract 10 actions, find the permission slip, or decide when each detail matters. Fambot is trying to own the work between incoming information and a usable plan.

David Reich built the company from familiar pressure. The father of 3 and former UnitedMasters president and Uber product leader told TechCrunch that he was catching up on roughly 40 kid-related emails while trying to protect time with his children. Greg Karlin worked at Instagram, while Jason Morrow brought experience from Google and LinkedIn.

More than 1,000 families used Fambot before the September 1 launch. It is free during beta on iOS, Android, and web. Future pricing may land around a Netflix subscription, though that can change.

Usefulness brings a trust burden. Fambot may encounter children's names, schools, activities, schedules, and appointments. The company says passwords stay with account providers, connections use OAuth, data is encrypted, and contracted models may not train on family data. Those promises will need to remain clear as integrations expand.

NextView's Stephanie Palmeri describes the trade-off: human help can be useful but expensive, while affordable tools often create another system for parents to manage. AI changes the math only if it removes administrative work without making a family the unpaid integrator.

The $3.5M backs an unusually intimate operating role. Fambot is asking families to trust software with small obligations that become large when missed, then make itself quieter as the household runs better.

Playfly Sports

A fan watches one game. The revenue around that game can be split across a team, a league, a broadcaster, a streaming service, a sponsor, a ticketing system, and several databases that disagree about who the fan is.

That fragmentation is the business Playfly Sports is trying to connect.

The sports media and marketing company secured a $250M senior credit facility from Bain Capital's Private Credit Group. Bain served as lead lender and administrative agent. Playfly said the capital will support continued growth; the announcement did not disclose the facility's interest rate, maturity, lender syndicate, or a more detailed use-of-proceeds plan.

Playfly sits in Access Holdings' portfolio and operates across three connected lines: media, sponsorship, and advisory services. Its current website says the platform works with more than 2,200 brands and 250 rightsholders, including 95+ professional teams and 75+ colleges. The transaction announcement used a different snapshot—2,000+ brands, 100+ professional teams, and 65+ colleges—so the numbers should not be blended as if they were measured at the same moment.

The scale matters because the product is coordination.

A sponsor rarely wants one disconnected asset. It wants an audience, distribution, activation, measurement, and a way to understand whether attention became commercial value. A rightsholder needs media reach and revenue without stitching together every relationship by hand. Playfly's wager is that one operating platform can connect more of that chain.

Craig Sloan became CEO in 2024, while founder Michael Schreiber moved into the Executive Chairman role. Chris Marinak joined as president in March 2026 to oversee the company's media, sponsorship, and advisory divisions. Playfly has also elevated Nicolina O'Rorke as COO and CFO as it builds a more integrated operating structure.

Private credit is an interesting instrument for this stage of the story. Unlike a priced equity round, the facility gives Playfly growth capital without announcing a new valuation. It also creates a repayment obligation. The bet is therefore not only that sports attention is valuable, but that Playfly can convert its position between rights owners, brands, distributors, and fans into durable cash flow.

The sports market often gets described through the price of teams and media rights. This deal points to another layer: the infrastructure that helps all of those assets transact with each other.

Playfly is financing the connective tissue.

Lyte

Robotics teams can spend years teaching a machine to move, then lose the production deployment to sensors that disagree about when the world changed. A camera sees an object. An inertial sensor records motion. A depth system calculates distance. The robot still has to reconcile the timestamps before it can act safely.

Lyte is building at that unglamorous, expensive handoff. Co-founders Alexander Shpunt, CEO, Arman Hajati, CTO, and Yuval Gerson, VP Engineering, have raised a $165M Series C led by Maverick Silicon. Fidelity Management & Research Company, Atreides Management, Key1 Capital, and Ora Global also participated. The round values Lyte at $1.6B post-money and brings total disclosed funding to $272M.

This team has already watched perception leave the lab. Shpunt co-founded PrimeSense, whose 3D sensing powered Microsoft Kinect before Apple acquired it. Shpunt, Hajati, and Gerson later worked on Apple sensing technology foundational to Face ID.

LyteVision takes that experience into robotics. It combines 4D coherent vision, imaging, inertial sensing, custom silicon, and spatial software on one synchronized timeline. Lyte says it measures position and motion natively, reducing latency, calibration drift, and brittle data alignment across separate components.

The customer is buying more than a sensor catalog. It needs trustworthy perception that lets engineers focus on robot behavior, deployment, and the ugly edge cases missing in a controlled demo.

Lyte says it has entered production and is shipping to customers across inspection, logistics, and manufacturing. The customers remain unnamed, so the evidence is company-reported. The Series C funds production, deeper AI and perception work, deployments, and hiring across technical, manufacturing, operations, and go-to-market roles.

Andrew Homan, managing partner at Maverick Silicon, is joining the board. Chairman Avigdor Willenz has spent a career around foundational semiconductor companies. Their wager is that perception becomes a shared infrastructure layer for physical AI instead of a custom engineering tax paid inside every robot program.

The next factory deployment will care less about how intelligent a robot looked on stage than whether every sensor agreed quickly enough when a person stepped into the work cell. Lyte is carrying that argument from the transistor into production, one synchronized view of a moving room at a time.

Empirik

An infrastructure diagram is usually accurate right up until somebody needs it.

By then, a permission changed, a Terraform provider completed, or a forgotten dependency became the reason production stopped behaving. The environment remembers the change badly, if it remembers it at all.

Empirik is building for that gap.

The San Francisco company emerged from stealth with a $21M Seed backed by Sequoia Capital, S32, Canapi Ventures, and Alumni Ventures. Sequoia incubated the business in 2023 before it spun out as an independent company.

Co-founders Avon Puri and Sudheer Dhurjati came to the problem after operating infrastructure across VMware, Rubrik, and Sequoia. Kartik Chandrayana joined as CEO after leading more than 30 infrastructure products at Salesforce and serving as CPO at Quantum Metric.

That operating history matters because Empirik is making a different bet on observability. Most tools tell a team that latency rose, errors multiplied, or a service failed. Empirik treats the change itself as the primary signal.

Its platform captures intent from a pull request, ticket, or pipeline, then compiles cloud, on-prem, Kubernetes, IAM, CI/CD, and SaaS metadata into a live graph. The goal is to calculate blast radius before execution and decide whether to block, govern, or eventually automate the change. That context is the product, not a decorative data layer.

That is a bigger job than incident response. Coding agents can generate and ship software at machine speed. Infrastructure teams still rely on review boards, ticket queues, runbooks, and institutional memory built for human-speed change. More automation upstream increases the value of context downstream.

Empirik says it is already in production at Guardant Health, Avahi Systems, TCBPay, and two unnamed Fortune-ranked enterprises. Sequoia reports that TCBPay cut detection of sensitive configuration changes from 30 minutes to 10 seconds. Those are company and investor-reported claims, not independently audited performance.

The $21M gives Empirik room to expand the product and the enterprise footprint. The harder commercial work sits inside the promise: customers must trust one system to understand the environment broadly enough to govern the agents changing it.

Every enterprise wants faster software. The authority given to that software will depend on whether the infrastructure can explain itself before the change lands.

iPRONICS

An AI cluster can contain thousands of expensive accelerators and still lose time to the paths between them. Training changes shape. Inference arrives with a different traffic pattern. The network has to catch up before the hardware can stay useful.

iPRONICS is building for that moment. The Valencia-born silicon photonics company raised a $125M Series B co-led by Maverick Silicon and Light Street Capital, with NVIDIA participating. The round brings total disclosed funding to $177M.

The financing is not just a bet on faster components. It is a bet that the topology of an AI data center should be programmable.

The iPRONICS Optical Networking Engine, or ONE, puts an optical circuit switch, control software, telemetry, and APIs into a rack-mounted system. The company says operators can reconfigure optical paths in real time as training and inference workloads change, without treating the physical network as fixed.

That distinction matters because accelerator utilization is an economic question. The GPU can be powerful. The model can be capable. But if the fabric cannot connect the right resources for the current job, the capital already sitting in the rack waits.

The technical thesis began at Universitat Politècnica de València, where iPRONICS was founded as a spinout in 2019. Founder and CTO Daniel Pérez-López has spent years turning programmable photonics into a system data-center teams can deploy. CEO Christian Dupont is now leading the commercial scale-up.

Maverick Silicon senior managing director Manish Muthal and Geoffrey Tate, an adviser to Light Street Capital, are joining the board. Young Sohn, a founding member of Catalight Capital and former Samsung corporate president and chief strategy officer, becomes an adviser.

The money will scale operations and commercial deployment, expand product and customer programs, deepen AI-infrastructure partnerships, and grow the Santa Clara office. iPRONICS reports strong demand, but it has not disclosed customer names, revenue, deployment volume, or independent production benchmarks.

Optical switching has to prove more than speed in a lab. It has to survive operations, integrate with the software control plane, and make accelerator economics better under changing workloads.

The next contest in AI infrastructure may be decided by who gives the cluster a network that can change its mind as quickly as the workload does.

Descartes / Tai Software

A shipper buys one completed move. A freight broker has to survive every handoff required to make it real. The quote becomes carrier sourcing, the carrier becomes execution, execution becomes billing, and the customer still expects one clean answer when the freight refuses to behave.

That daily chain is why Descartes Systems Group paid approximately $100M in cash for Tai Software. The acquisition was announced as completed on August 24, bringing Tai's broker-specific transportation management platform into the Descartes Global Logistics Network.

Tai puts quoting, carrier sourcing, load execution, billing, and customer engagement into one workflow. The company says it connects directly with more than 100 carriers and technology tools. Walter “Mitch” Mitchell is Tai's founder and CEO.

Descartes is buying more than another TMS logo for the portfolio. Tai sits where shipment intent becomes transaction data: which carrier gets selected, how the load moves, where communication breaks, what gets billed, and how quickly the customer learns that reality changed.

That data becomes more useful when it meets Descartes' strengths in carrier onboarding, compliance, fraud prevention, real-time visibility, rating, payment, customs, and trade intelligence. CEO Edward J. Ryan framed the opportunity around better execution, margins, and customer and carrier relationships. Andrew Wimer emphasized the transaction and shipment data Tai adds to the network.

Accel-KKR invested in Tai in 2020. The exit now hands the platform to a public logistics-software buyer with a long acquisition record and a much wider distribution surface. For Tai customers, the promise is fewer seams between the broker's system of action and the network carrying compliance, visibility, and carrier context around it.

The press release does not disclose Tai's revenue, a stock component, an earnout, a separate regulatory timeline, or an integration schedule. That leaves the important operating work in front of both teams. A broader product catalog will not impress a broker handling a late truck, a suspicious carrier, an angry customer, and an invoice that still has to close before Friday.

The broker will measure this integration in saved minutes, safer choices, and fewer calls. The $100M gets Descartes into the middle of that day. The value will keep moving with every handoff the combined platform can make disappear while the freight keeps moving.

Foundation Software

A contractor can approve a $2M backhoe faster than a $12,000 software purchase. Mike Ode used that contrast while explaining the market Foundation Software serves, and it gets to the strange economics underneath this reported sale. Heavy equipment is visible. Back-office software looks smaller until it becomes responsible for job cost, payroll, billing, field time, safety, and the records that decide whether a project made money.

Reuters reports that Thoma Bravo is exploring a sale of Foundation Software at a potential valuation above $2B. William Blair is said to be running the process, with interest from private-equity firms and corporate buyers. The same report puts Foundation above $200M in annual revenue and $100M in EBITDA. No buyer, signed agreement, purchase price, structure, or close date has been announced.

That distinction matters because this is an auction story, not an acquisition announcement. The more useful question is why an Ohio construction-software company founded in 1985 can now command this kind of attention.

Thoma Bravo invested in Foundation in 2020, when the business had about $50M in revenue, less than $10M in EBITDA, roughly 5,000 customers, and two primary products, according to the firm's later case study. By 2024, Thoma Bravo said Foundation was near $150M in revenue, operating at a 45% EBITDA margin, serving more than 12,000 customers, and selling eight products after four debt-financed add-on acquisitions.

Mike Ode and the Foundation team expanded around the contractor's back office without wandering away from it. Accounting led to payroll, estimating, project management, field operations, safety, HR, expenses, and payments.

That is the quiet power of vertical software. The interface can look ordinary while the data model becomes institutional memory. Replacing it means more than migrating records. A contractor has to trust that job costs, union rules, certified payroll, change orders, field hours, and financial reporting still agree on Monday morning.

Thoma Bravo may now be testing how buyers value that trust after six years of product expansion and cross-sell. The reported numbers suggest Foundation has already crossed from niche application into a profitable construction operating layer. The next owner, if there is one, will inherit the awkward promise that makes this category durable: keep modernizing the suite without making the contractor relearn the business side of every job.

Superluminal Medicines

AI drug discovery gets to enjoy a long honeymoon with speed. Clinical development eventually asks the uncomfortable question: what did all that speed produce?

Superluminal Medicines has closed an oversubscribed $60M Series B led by BVF Partners, with Deep Track Capital and Perceptive Advisors joining existing investors RA Capital Management, Insight Partners, NVentures, Catalio Capital Management, Eli Lilly and Company, Cooley, and Gaingels.

The money is aimed at a precise handoff. Superluminal expects to begin a Phase 1 study by the end of 2026 for an oral, selective, biased MC4R agonist being developed for rare genetic forms of obesity and hypothalamic obesity. Additional pipeline programs and the company's GPCR discovery platform are also getting funded.

Co-founder and CEO Cony D'Cruz has described the moment as a move from discovery platform to clinical biotechnology. Co-founder and CSO Ajay Yekkirala built the lead program around signaling bias: activate the MC4R pathways associated with the intended effect while trying to avoid activity tied to unwanted effects. Preclinical studies look encouraging according to the company. Human data does not exist yet.

That distinction is where this round earns attention. A model can rank molecules, explore protein conformations, and shorten a design cycle. A patient cannot be simulated into safety.

Superluminal's Hyperloop platform combines protein dynamics, cryo-EM, machine learning, molecule design, predictive ADME/toxicology, and experimental validation. It is designed to turn moving GPCR structures into better small-molecule starting points.

There is serious capital behind the thesis. Superluminal launched with $33M in 2023 and added a $120M Series A in 2024. Including this round, disclosed financings total $213M. Its separate Eli Lilly collaboration carries up to $1.3B in potential payments, equity, milestones, and royalties. That is contingent deal value, not cash already banked and not venture funding.

This syndicate is underwriting more than another AI screen. It is backing whether one discovery system can repeatedly produce compounds worth carrying through the expensive human part of drug development.

The near-term record will be built in quieter rooms: protocol execution, dose selection, safety review, and the first data that tells the team whether its chosen signaling profile survives contact with biology. Every platform eventually needs a molecule willing to carry the whole argument forward.

PatientIQ

Three months after surgery, the clinical record can be full and the patient is still the missing data source.

A surgeon can see the procedure, an EHR can record the encounter, and an insurer can price the claim. Recovery is harder. Pain, function, mobility, and quality of life live with the patient, often somewhere beyond the tidy edge of a clinical workflow.

That missing handoff is the business PatientIQ has spent a decade building around. The Chicago company raised a $30M Series C led by Hughes & Company. Current investors Health Enterprise Partners and August Capital were named in the announcement, and Travis Hughes will join the board.

Matthew Gitelis founded PatientIQ in 2016 after moving from investment banking into medical school and outcomes research. That origin sits inside the product: medicine often has more important data than finance and a harder time getting it out of the workflow intact.

PatientIQ collects outcomes through EHR-integrated programs, then turns responses into clinical, research, regulatory, and benchmarking data. The company reports a network of 870+ healthcare organizations, 17M+ patients, 70M+ outcomes, 20+ medical specialties, and 50+ EHR integrations including Epic, Oracle Cerner, and athenahealth.

Those are company-reported scale figures, not an audited claim that every measurement improved care. Since PatientIQ's $20M Series B in 2022, its reported provider network grew from 200 organizations to more than 870. The platform is now carrying evidence across health systems, registries, medical-device programs, and life-sciences research.

That gives John Gerhardt, PatientIQ's CTO, a harder problem than "healthcare data" suggests. The platform must collect the right answer from the right patient, connect it to the right clinical record, and make it useful without adding another task to a crowded care team.

The new capital will deepen PatientIQ's presence across specialties and health systems, expand its real-world-evidence work for medical-device and life-sciences companies, and support a direct-to-patient commerce experience. That last move is the sharpest extension of the model. A dataset that helps explain recovery can also influence which products and services patients encounter.

The trust boundary is real. PatientIQ has built its value by making the patient's voice measurable inside care. The next chapter asks whether that relationship can support research and commerce while the answer still reaches the clinician who needed it first.

Resect AI

The AI answered in half a second. The team checking whether it was right may need half a quarter.

That gap is where enterprise adoption gets expensive. A model can produce a confident sentence before anyone knows what produced it, which evidence supports it, or who owns the damage if it reaches a loan decision, clinical workflow, or customer account.

Resect AI is putting $25M behind the idea that accountability has to live closer to the model itself.

The company launched from stealth on September 3 with funding from unnamed private equity investors. Resect AI did not disclose a round label, investor names, or valuation. The capital will support R&D, go-to-market work, and Seattle/Portland hiring.

Founder, CEO, and chairman Kevin Owens is building with co-founders Tim Walton, Chief AI Officer; Tyler Gerber, COO; and Tommy Lofgren, Chief Product and Marketing Officer.

Resect AI says its forthcoming NeuroWave suite will observe, detect, interpret, audit, and modify model behavior. The thesis reaches beneath output monitoring toward understanding why a model fails and intervening when it does.

There is public technical work to inspect. Resect Research Labs has released 0.6B and 8B Veritas fact-checking models on Hugging Face. Its model cards report 72.30% and 75.47% average balanced accuracy on LLM-AggreFact. Those are company-published benchmarks, not independent production proof.

The accountability market is filling with evaluation tools, guardrails, monitors, and policy engines. Most judge what comes out. Resect AI is betting there is value closer to the mechanisms that made the answer possible.

The timing has teeth. NIST treats confabulation as a distinct generative-AI risk. The European Commission began enforcing AI Act transparency requirements in August. Enterprises must explain systems just as those systems become more autonomous.

The $25M gives Resect AI room to turn a research position into an operating product. Named customers, production deployments, independent evaluations, investor identities, and general availability remain outside the public record.

Enterprise buyers will purchase accountability when it can trace a failure, show what changed, satisfy an auditor, and keep a persuasive machine from moving the wrong decision through the business.

The answer is getting cheap. Resect AI is building for the people who still have to explain it when the room gets serious, and that explanation now has to survive contact with production.

System One

There is a point where critical infrastructure stops being an engineering problem and becomes a people problem.

Money funds the assets. People make them work.

That is the layer Oaktree Capital Management and StepStone Group just backed.

System One announced a completed strategic investment from funds managed by Oaktree and StepStone. Financial terms were not disclosed. Oaktree is renewing an existing relationship with the company, while StepStone joins as a new investor.

This is not a software startup raising a round to chase a category. System One is an established provider of workforce solutions and outsourced services across energy, engineering, technology, life sciences, and government. The company reports more than 50 locations and a network of over 10,000 professionals.

Those numbers matter because infrastructure ambition is cheap in a press release and brutally expensive in execution.

Energy, data center, pharmaceutical, and public modernization programs all meet the same constraint: specialized work needs qualified people in the right place, with little room for error. System One sits inside that constraint.

Chairman and CEO Troy Gregory said the investment gives System One added resources to accelerate growth, expand capabilities, strengthen its platform, invest in employees, and serve clients. Oaktree Managing Director Jimmy Lee framed the renewed partnership around a model built for scale. StepStone Partner Geoff Dolan pointed to the company's ability to operate in complex and essential industries.

The language is polite. The signal underneath it is sharper.

Investors are not only funding the hard assets behind infrastructure. They are funding the human systems required to make those assets useful. The cranes, cables, reactors, laboratories, and networks get the glamour shot. The recruiting, training, deployment, compliance, and project execution determine whether anything actually moves.

System One's current organization grew out of a 2008 management buyout led by Troy Gregory and President and COO Greg Lignelli. This business was shaped by operating cycles, client demands, and long industry relationships.

The next proof will be operational. Can the new capital help System One widen its capabilities without thinning the quality that made the platform investable? Can it scale across essential industries while keeping specialized talent close to the work?

That is where this investment earns its name.

Diameter Pay

The border in cross-border payments is often a risk decision, not a line on a map.

A financial institution can have a legitimate customer, a real business need, and money ready to move. The transfer still waits on a harder question: can the bank understand the counterparty, corridor, and compliance burden well enough to say yes?

Diameter Pay just raised a $10M Series A to build more infrastructure inside that answer.

CMT Digital and Lightspeed Faction co-led the equity round. SixThirty Ventures, Stellar Development Foundation, Tech Council Ventures, Onigiri Capital, and BitRock Capital participated. Founder and CEO David Lighton said fundraising began in April and closed in July. The valuation was not disclosed.

Diameter Pay gives banks, fintechs, and digital-asset exchanges access to U.S. dollar virtual accounts, domestic and international payment rails, stablecoin on- and off-ramps, and embedded compliance controls through one API.

Stablecoins can make dollar value programmable and available around the clock. Regulated institutions still decide who receives an account, which transactions clear, how risk is monitored, and whether a corridor remains open.

Diameter Pay is selling that connective layer.

The company reports more than $10B in payment volume year-to-date in 2026 and more than 10,000 live end users. Named clients and revenue remain private. The Block identified Portage Bank and SSB Bank as sponsor partners, with a third public-company bank undisclosed.

The $10M will expand banking and payment capabilities, deepen stablecoin and FX infrastructure, and fund more technology and compliance work. Diameter Pay is also hiring a CTO and chief commercial officer. CMT Digital received a board seat, while Lightspeed Faction and SixThirty Ventures took observer seats.

That governance footprint says something about the round. The investors are backing a company whose software cannot outrun its banking relationships, licensing, transaction controls, or the institutions carrying settlement risk.

The market loves to describe stablecoins as a cleaner rail. Diameter Pay is building around the institutional seams the rail still has to cross. If it can widen dollar access while keeping sponsor banks comfortable with the risk, more global finance can move without asking regulated institutions to look away.

The next proof will be the institutions Diameter Pay can name, the corridors it can keep open, and whether $10B of reported movement becomes durable trust.

Thinking Machines Lab

The most expensive way to argue for distributed AI may be to build a frontier lab. Thinking Machines Lab wants organizations to shape models around their own knowledge, but that freedom still depends on a concentrated world of chips, power, talent, and capital.

The Information reports that Thinking Machines is in talks to raise at least $1B at a pre-money valuation of at least $40B. Existing investor Accel may lead, while NVIDIA has discussed participating. Those are proposed terms, not a closed financing, and the distinction matters when the numbers arrive this large.

Thinking Machines raised $2B in 2025 at a reported $12B post-money valuation before it released a model. Andreessen Horowitz led, with NVIDIA and Accel participating.

The company now has more than pedigree to sell. Inkling, released in July, is a 975B-parameter open-weight model built to reason across text, images, and audio, with 41B active parameters and a context window up to 1M tokens. Thinking Machines does not call it the strongest model available. The bet is more useful: give organizations a capable base they can own and customize through Tinker around the knowledge that makes their work different.

That changes where the business has to earn its keep. Open weights can leave the lab. Revenue has to come from training, fine-tuning, serving, and the tools that turn a general model into something specific enough to trust. The Information reports at least a few hundred million dollars in annualized revenue, but that figure comes from an unnamed source and describes a run rate, not audited annual revenue.

The infrastructure refuses to become philosophical. Thinking Machines and NVIDIA announced a multiyear partnership in March to deploy at least 1 gigawatt of Vera Rubin systems beginning in 2027. NVIDIA also made a significant investment of undisclosed size. A company selling customization still needs one of the most centralized industrial stacks on earth to train and serve the models being customized.

Investors would be pricing the possibility that enterprise AI moves away from renting one frozen intelligence and toward continuously shaping models around private knowledge. Thinking Machines now has to turn that possibility into durable customer economics while paying for frontier-scale infrastructure. The weights can travel farther than the power bill, and the company is trying to make a market in the distance between them.

StepStone Group

The seller in an infrastructure secondary usually wants liquidity. The sponsor still wants a partner.

Those 2 needs can share a transaction, but they do not begin from the same clock. An LP may be rebalancing a portfolio or making room for new commitments. A GP may be trying to hold a good asset beyond the life of the original fund.

StepStone Group has raised $1.7B across StepStone Secondaries Infrastructure Fund and related separate accounts for that market. The accounting matters: SSIF itself closed at its $1.5B hard cap, while separate accounts supplied the remaining $200M.

This is StepStone's first closed-ended commingled fund dedicated to infrastructure secondaries. It will acquire LP interests in infrastructure funds and invest in GP-led secondary funds managed by third-party infrastructure GPs.

James O'Leary, Partner and Head of Infrastructure & Real Assets, described secondaries as a relationship business. The LP wants a credible route to liquidity. The GP wants capital that understands the fund, the asset, and the years of work still attached to both.

StepStone is trying to turn its position across primary funds, co-investments, and secondaries into an underwriting advantage. The infrastructure team says it deployed an average of $13B annually over the past 3 years. SPI by StepStone gives the firm another view into funds and assets before they reach a sale process.

SSIF was already about 50% deployed across 26 closed LP-interest and GP-led deals as of August. Many were in the middle market, where fewer bidders and thinner information can make an existing relationship more valuable than another polished auction deck.

The timing is difficult to ignore. Jefferies expects infrastructure secondaries volume to reach $30B in 2026 after roughly $20B in 2025. Lazard estimates the wider secondary market processed $124B in the first half of 2026, split almost evenly between GP-led and LP-led transactions.

StepStone did not name the fund's LPs, disclose return targets, or identify the 26 deals. The 50% deployment figure and platform advantages remain company-reported, not performance results. What the close establishes is capacity in a market that increasingly treats liquidity as part of portfolio construction instead of an emergency exit.

Infrastructure assets were built to outlive the vehicles that first owned them. The next relationship often starts before anyone admits the clock has become a problem.