Where the Market Has Stopped Looking

DigitalBridge

SoftBank did not buy $108B of data centers. It bought the institution that decides which data centers, towers, fiber systems, small cells, and edge assets get financed.

SoftBank Group Corp. completed its acquisition of DigitalBridge on September 30. The closing announcement puts outstanding common stock at approximately $3.1B. The original agreement used approximately $4.0B of enterprise value. Those numbers describe different pieces of the deal.

DigitalBridge shareholders received $16.00 per common share in cash. Required regulatory approvals, including FERC and CFIUS clearance, arrived before closing. DigitalBridge common stock stopped trading on the NYSE, and the company became a controlled SoftBank subsidiary.

The operating platform stays separate under CEO Marc Ganzi. That continuity matters because DigitalBridge is not one infrastructure project with a logo on the gate. It is an alternative asset manager built around institutional capital, specialist teams, and relationships across data centers, towers, fiber, small cells, and edge infrastructure.

DigitalBridge reported $40.2B in fee-earning equity under management at June 30. The $108B figure used when the deal was announced referred to infrastructure assets managed on behalf of limited partners and shareholders as of September 30, 2025. SoftBank did not acquire every underlying asset for $3.1B.

It acquired the manager, the operating knowledge, and the machinery that turns capital into physical capacity.

That fills a different gap in Masayoshi Son's AI strategy. Models need chips. Chips need power, cooling, land, fiber, permits, financing, and operators who can keep a 20-year asset honest after the keynote ends.

SoftBank already knew how to place enormous technology bets. DigitalBridge gives it a platform built to originate, finance, and manage the infrastructure those bets eventually require.

The acquisition does not guarantee faster construction, cheaper power, or a smooth integration. No primary source promises any of that. It moves another control point in the AI economy inside SoftBank's orbit, while leaving Marc Ganzi and the DigitalBridge team responsible for making institutional continuity survive a change in ownership.

DigitalBridge has to convert SoftBank's scale into projects that survive utility negotiations, regulatory approvals, construction schedules, customer demand, and the patience of limited partners. Those obligations will keep moving after the acquisition disappears from the ticker.

Bright Machines

By the time an AI server fails electrical test, its evidence may be scattered across design files, assembly stations, robot logs, inspection images, and material handoffs.

Reconstructing that chain is the useful story inside Teradyne's strategic investment in Bright Machines.

The amount and ownership terms were not disclosed. The operating wager was. Teradyne plans to explore combining its robotics and board-test technologies with Bright Machines' software-defined manufacturing platform for AI and data-center infrastructure.

The proposed environment brings Bright Machines systems, Universal Robots cobots, and Teradyne test equipment onto one floor. Assembly, inspection, movement, and electrical results could share one product genealogy instead of becoming separate witnesses to the same build.

That matters because AI hardware refuses to stay still long enough for rigid automation to feel comfortable. New accelerators change power, cooling, networking, and physical configurations. Customers want different systems. Quality standards remain unforgiving while the line has to learn another version.

James Davidson, Teradyne's chief AI officer, put the constraint in engineering terms: the question is not only what a robot can physically do, but how much engineering it takes to tell the robot what to do. The ambition is to teach a new task in hours instead of rebuilding the workflow over weeks.

Bright Machines reports 130+ microfactory deployments across 10+ countries and 60+ customers. Those figures are company-reported, and the collaboration has not published customer deployments or measured production gains.

CEO Sviat Dulianinov is taking Bright Machines into that proof phase with co-founder and chairman Lior Susan. Bright Machines last disclosed a $126M Series C in 2024. Teradyne's new investment is additional, but its amount remains unknown.

Strategic capital carries a different obligation from a logo on a cap table. Teradyne owns technologies that touch the build and the verdict. Bright Machines is building the layer meant to connect them. The relationship gets valuable when a design change moves through assembly, inspection, and test without losing the reason each decision was made.

AI infrastructure will keep demanding more factories, but the sharper contest may be over how quickly each factory can understand what it just built. Bright Machines and Teradyne now have to turn a shared data thread into production evidence that follows every unit off the line.

SafeWorld

A worker carrying boxes around a blind warehouse corner should never become the crash-test dummy for a robot's latest software update.

That sounds obvious until the robot is powered by probabilistic AI. Lighting, sensor placement, floor traffic, human posture, and a routine model update keep rewriting the edge cases. Physical testing still matters, but nobody can responsibly stage every fall, occlusion, or bad turn around an actual person.

SafeWorld is building the evidence layer between what a robot can do in a demo and what an enterprise can defend in deployment. The Palo Alto company emerged from stealth with a $12.2M seed round co-led by Shine Capital and a16z Speedrun.

Co-founders Kyle Wong, Ding Zhao, and Simo Rachidi arrived through 3 different doors. Kyle Wong built and sold Pixlee, then led StartX. Ding Zhao directs Carnegie Mellon University's Safe AI Lab. Simo Rachidi brings production ML, cybersecurity, and large-scale data infrastructure experience.

The product turns robot safety questions into repeatable simulation tests. Teams build scenarios from requirements, incidents, and robot logs, vary human movement and operating conditions, run the robot's control software through those cases, and preserve the results for review.

That last part matters. A convincing animation is not safety evidence. An engineer needs to know which requirement was tested, what changed, where the model stopped being representative, and whether a later software version fixed one failure while creating another.

SafeWorld says it is already working with robotics companies and global enterprises. TechCrunch identified Gritt Robotics as a development partner, and company-distributed announcement material says pilots include automotive, medical-device, and warehouse-automation enterprises. Those are early operating signals, not certification or audited proof that the platform has solved robot safety.

The investor logic is easy to see. Every new robot capability expands the set of environments where autonomy can create value. It also expands the number of people, insurers, safety teams, and enterprise buyers who need evidence before permission follows capability.

$12.2M gives SafeWorld room to turn rare human behavior into an engineering workflow. The company still has to prove that simulated evidence transfers into the messy physical environments it models, but that is exactly where the commercial pressure lives: the robot may learn quickly; the organization standing beside it still needs a record it can trust.

Vocca

A medical practice can have an open appointment, a patient trying to book it, and a receptionist doing everything right, yet still lose the visit because the phone rang at the wrong minute.

That is the administrative gap Vocca is building against. The Paris- and New York-based company raised a $20M Series A led by Norrsken VC, with Heal Capital, Speedinvest, firstminute capital, and angels including the founders of Alan and Datadog participating.

Eliott Hoffenberg, co-founder and CEO, and Hugo Danet, co-founder and CTO, started Vocca in July 2024 around a plain healthcare frustration: access often fails before care begins. Staff repeat scheduling work while patients abandon calls or wait for someone to navigate the rules for an exam, specialist, preparation, and calendar.

Vocca's answer is more specific than a voice bot that sounds pleasant. Its AI handles calls around the clock, but the scheduling engine behind the conversation applies specialty rules, books or moves appointments inside existing software, sends reminders, and hands urgent or complex cases back to staff.

Vocca and Norrsken report that the platform now serves 15,000 practitioners across 1,500 practices and 20+ specialties in the US and Europe. They say revenue grew 7x and monthly conversation volume grew 10x in a year to more than one million. Vocca reports 70% average end-to-end resolution without human involvement, rising above 85% at its most mature customers.

Those are company-reported figures, not an audited scorecard. They still show why this round is larger than the phone demo. Healthcare organizations are buying completed administrative work: the right patient placed into the right slot, under the right rules, without asking a receptionist to rescue every edge case.

The $20M will fund deeper scheduling capabilities, more specialties, additional communication channels, and further US expansion. Vocca says it has about 40 people across Paris and New York and plans to hire 40 more by year-end.

The pressure now moves into the details that healthcare buyers do not forgive. Every new specialty adds rules. Every integration adds exceptions. Every market adds a different mix of privacy, workflow, and patient expectation. A voice can recover from an awkward sentence. A scheduling system has to recover the appointment.

Vocca has already moved from 2,000 to 15,000 practitioners in a year. The next phase is about preserving that precision while the calls, clinics, channels, and edge cases multiply.

Limited

A parent company can own the sender, the recipient, and every dollar in between. The banking system still sees 2 countries, 2 institutions, an FX spread, a cutoff, and a reconciliation problem.

That is the operating reality behind Limited's Seed round, which Third Prime preempted less than 10 months after launch. The financing expands the Seed to $18.5M, with North Island Ventures, The House Fund, and Collab+Currency returning. ParaFi Capital, Pharsalus Capital, Digital Currency Group, Onigiri Capital, and 1982 Ventures joined the round.

Founder and CEO Hussein Ahmed is building Limited for companies whose legal entities and cash flows crossed borders before their finance stack did. Instead of opening another bank account, managing another login, waiting through another cutoff, and rebuilding controls in every country, Limited puts local accounts, intercompany transfers, cards, spend controls, accounting, and treasury in one operating layer.

The stablecoin infrastructure sits underneath the workflow. A CFO funding a subsidiary in Mexico, paying a supplier in Brazil, or reconciling a team in the U.K. does not need a crypto lesson. The CFO needs money to arrive, books to stay legible, permissions to hold, and somebody to understand what happens when a payment crosses jurisdictions.

Limited's current product site says local payouts reach 140+ countries over 300+ rails. It offers company-named accounts in supported markets plus physical and virtual Visa cards. Limited also says the part fintech copy often buries: it is not a bank. USD balances sit with partner banks under pass-through insurance requirements. Stablecoin balances and local-currency accounts outside the U.S. are not FDIC insured.

The new capital is aimed directly at the hard parts. Limited plans senior hires across go-to-market, operations, and compliance, more local corridors across LATAM, APAC, and MENA, and deeper treasury features for larger multi-entity companies. Hussein Ahmed said customer demand pulled the round forward. Limited did not disclose a valuation or the incremental amount added by the extension.

The useful signal is not speed by itself. Multinational finance teams are starting to expect new rails without giving up the controls, records, and local access that make ordinary business possible. Limited now has more capital to make those pieces behave like one system while its customers keep adding countries, entities, and obligations.

Collibra / trail ML

Enterprise AI governance has spent years perfecting the document. trail ML built for the moment the document has to say no.

Collibra acquired the Munich AI governance company on October 5, bringing agent-powered assessment and runtime enforcement into its enterprise AI control plane. Financial terms and transaction details were not disclosed.

An enterprise can write the right policy, assign an owner, collect evidence, and approve a control. Then an AI agent reaches GitHub, Slack, Jira, an internal database, or another live tool and acts inside a context that changed after the assessment was signed.

That is where trail's software operates. Its agents analyze evidence around an AI system, determine which requirements and controls apply, assess whether those controls are effective, identify gaps, and trigger reassessment when the evidence changes. Its runtime layer can enforce Collibra policies where agents run and block actions that violate them before they complete.

Collibra brings enterprise context across data, models, applications, and agents. The company says more than 700 customers use its platform and that it governs more than 2 billion assets.

trail gives that context somewhere to act. The acquisition turns that connection into an operating bet, with controls expected to travel from the policy record into an agent's path.

Anna Spitznagel, Nikolaus Pinger, and Sven Hölzel founded trail in Munich in 2023 around the handoff between technical and compliance teams. Anna Spitznagel serves as CEO, Nikolaus Pinger as CPO, and Sven Hölzel as CTO. trail previously raised a €1.45M pre-seed led by CapitalT, with NP-Hard Ventures, Mozilla Ventures, Common Magic, and business angels participating.

That funding history explains how trail reached the deal. It does not reveal the acquisition value, which remains undisclosed.

Governance platforms were built to help enterprises know what data, models, owners, risks, and policies they had. Autonomous agents force a harder question while the system is moving: is this action allowed right now, under the evidence and controls that apply here?

Collibra is betting that governance belongs closer to execution. trail now has to make continuous assessment and enforcement reliable across the kinds of enterprise environments where a false approval can create an incident and a false block can quietly kill adoption.

The policy can stay on the page. The consequence has already reached the tool call.

Spiko

Payroll has a deadline. Supplier invoices have due dates. The cash waiting for both can spend the rest of the month sitting in an operating account that serves the institution holding it better than the business that earned it.

Spiko has raised a $90M Series B led by New Enterprise Associates. Index Ventures, Bpifrance, Speedinvest, Flourish Ventures, White Star Capital, and several other firms joined, alongside Axel Weber and the founders of Qonto. Spiko says the round brings its total funding to $120M.

Paul-Adrien Hyppolite and Antoine Michon built Spiko after working inside French public institutions, where financial infrastructure is less a pitch deck and more a pile of rules that still has to function on Monday morning. The company designs regulated cash funds for its own apps and embeds them in financial platforms through an API, across EUR, USD, GBP, and CHF.

That combination matters because tokenization is the visible part of the product, not the entire job. A finance team needs cash ready for payroll and suppliers, excess in an appropriate fund, credible custody, controls, and reconciliation. Spiko is trying to turn those decisions into standing software rules instead of a recurring spreadsheet ritual.

The scale has moved quickly. Spiko reports $2.7B in assets under management, more than 5x AUM growth in 12 months, and 10,000+ businesses and individuals across more than 25 jurisdictions. The company also says its range now leads the tokenized cash-fund category, citing RWA.xyz. That is a category claim, not a suggestion that a 3-year-old company has outrun BlackRock or Franklin Templeton across global asset management.

NEA's investment reads like a bet on the handoff between regulated finance and continuously operating software. Money-market funds were built around business days. Stablecoins and smart contracts do not clock out, and neither do global payment obligations. Spiko already offers instant withdrawals and says it is working toward interest that accrues continuously, with allocation rules adjustable through an API, a treasury-management system, or eventually an AI agent.

The $90M will fund new products, new markets, and local teams across Germany, Italy, Spain, the Netherlands, and the Nordics. The harder work is making that expansion feel boring in the best possible way: the right cash stays available, the rest keeps working, and the finance team can explain every movement after the software makes it.

Procuros

Before an AI agent can run a supply chain, it has to survive the invoice.

Then the purchase order, shipping notice, ERP mapping, product code, delivery date, payment term and exception hiding inside somebody's PDF.

Procuros has raised a €20M Series A, about $22.5M at the October 6 ECB reference rate, to build the infrastructure that lets those records agree across company boundaries. XAnge led the round. W23 Global joined as a new investor, while Point Nine, Creandum, b2venture, SquareOne Venture Capital and angel investors participated.

Patrick Thelen, CEO and co-founder, met this problem while sales orders moved through fax machines, PDFs and customer tools. Benjamin Wulff, CTO and co-founder, saw interfaces at Airbus take 1 to 2 years to build. They founded Procuros in Hamburg in 2021 because “digital trade” was still asking people to carry information between systems that could not understand one another.

The company connects a retailer or supplier once, then translates EDI, APIs, spreadsheets, PDFs and portal data into the customer's ERP. The harder job comes after translation. Procuros checks product numbers, identities, quantities, prices, tax, payment terms, addresses and delivery details against the commercial context behind the transaction.

That context is where the AI story becomes useful. An agent can process an order quickly. It still needs to know whether the quantity matches the agreement, whether the shipping notice changed the delivery, whether the invoice belongs to the order and when an exception deserves a person.

Procuros and XAnge report more than 300 customers across DACH and the UK and more than €3B in annualized order volume. Procuros customer stories say KoRo runs the platform across 300+ trading partners and that Niche Beauty brought 100 suppliers live in one week. Those are company-reported results, but they show a network doing real operational work before the word “autonomous” gets the microphone.

The new capital will support product development, go-to-market expansion and growth across Europe and the US. The difficult part will be carrying the same trust into new supplier bases, ERP environments and commercial rules without turning every connection back into a custom project.

AI agents may become the visible operators of B2B trade. Procuros is betting that the shared record underneath them, where every order, invoice and exception has to describe the same reality, will decide how much authority those agents can safely earn.

Cleavr

A finance team can close the sale twice: once when the contract is signed, and again when the money finally arrives.

The space between those moments is where Cleavr is building a business. The French company has raised an €8M Seed round led by Varsity, with Kima Ventures, Better Angle, 100IN, Portfolio Ventures, Kerala Ventures, Financière Saint James, Station F, Clover, and business angels joining.

Six months ago, Cleavr raised a €1M pre-seed. Now it has more than 100 customers across six European markets and says its system processes several billion euros of invoices each year. That speed explains the capital.

Accounts receivable is usually described like administration. A due date passes, somebody sends a reminder, and the dashboard turns red. Inside the business, the chain is less polite. The right contact changed jobs. A purchase order is missing. A customer disputes one line. Sales wants patience while finance needs cash.

Baptiste Nassoy, Arthur Guerin, and Antoine Grenard built Cleavr to handle that operating mess, not merely display it. The platform connects to financial systems, identifies the right contact, sends multichannel reminders, interprets replies, tracks promises, surfaces disputes, and reconciles payment. Sensitive cases still return to the human team.

Cleavr reports that customers reduce DSO by 37%, collect 40% more cash, and recover 80% of the time spent on reminders. Those are company-reported averages, not independently audited results. They are the right metrics to watch because collections software gets judged by cash movement, not demo polish.

The round funds European expansion, product development, and 20 hires across sales, product, and technology. Language, payment behavior, ERP stacks, customer relationships, and escalation rules change across borders. Cleavr has to earn trust inside each one.

The European Commission has linked late payment to one in four bankruptcies in the EU. A company can report growth and still run short of operating room while customers finance themselves with somebody else's receivables.

Cleavr's seven-month sprint shows that finance teams are willing to delegate repetitive collection work. The €8M now has to turn that willingness into durable behavior across more markets, harder disputes, and longer customer histories. If Cleavr succeeds, the finance team stops spending its week asking everyone to pay and starts managing the exceptions that determine when revenue becomes usable cash.

Vinci

A hardware design can look finished on a screen and still be waiting for physics to ruin the meeting.

The chip package bends under heat. A cooling assumption fails. A material expands differently than expected. By the time a traditional simulation workflow returns the answer, other choices may already depend on the mistake.

That delay is the business Vinci is trying to remove.

The Palo Alto company has raised a $250M Series B at a $1.5B valuation. Advent, Temasek and Xora Innovation co-led the round, with AMD Ventures, Madrona, Eclipse, Khosla Ventures and others participating. Vinci had disclosed $46M across its Seed and Series A less than a year ago.

Hardik Kabaria, founder and CEO, built his Stanford doctoral work around computational geometry and high-fidelity meshing. Sarah Osentoski, co-founder and CTO, came from large-scale machine learning and autonomous systems. They are not asking hardware engineers to tolerate another cheerful AI guess. They are building for the part of engineering where an answer has to survive contact with matter.

Vinci calls the platform Continuous Physics Reasoning. It combines automated design preparation, a Foundation Model for Physics, agentic orchestration and GPU-native physics kernels. The company says it can deliver deterministic, solver-accurate analysis at manufacturing resolution without training on each customer's proprietary design data.

The first proving ground is semiconductors, where heat and warpage have become design constraints rather than cleanup work. Vinci has commercialized thermal, thermo-mechanical and convective-fluid capabilities. The company says its software can analyze designs ranging from hundreds of millions to more than 15B degrees of freedom in minutes instead of hours or days.

Those are Vinci's claims. Customer names and revenue remain undisclosed. Reuters reported that the roughly 70-person company plans to use the round for compute, hiring and new products while growing pilot deployments from 2 to 20.

Cadence and Synopsys have spent decades becoming trusted parts of engineering workflows. Vinci does not win because AI sounds newer. It wins only if teams can ask more physical questions earlier without trading speed for confidence.

$250M buys Vinci the compute and talent to make that case across more programs. It also puts a $1.5B price on the company's ability to turn physical truth into something engineers can use while the design is still a choice.

Type One Energy

Type One Energy's most consequential design choice may sit outside the reactor: who gets to build the thing.

The Knoxville fusion company completed a $200M Series B co-led by Breakthrough Energy Ventures and Clutterbuck Capital, with Lowercarbon Capital, Siemens Energy Ventures, SiteGround Capital, and other new investors participating. Reuters reports that total financing now exceeds $400M. The company did not disclose a valuation.

CEO Chris Mowry is using that capital to make a commercial bet. Type One Energy wants to operate as a fusion technology provider and OEM, combining its stellarator design with the manufacturing, engineering, construction, and operating capacity of established partners.

Fusion's difficult question no longer ends with holding plasma in place. It extends to whether utilities, engineers, manufacturers, regulators, laboratories, and investors can work as one delivery system without forcing Type One Energy to recreate every capability on its own balance sheet.

Co-founders David Anderson, Chris Hegna, and John Canik bring work reaching back to the HSX stellarator at the University of Wisconsin-Madison. CTO Thomas Sunn Pedersen adds experience from Wendelstein 7-X. Their design uses computational optimization and high-temperature superconducting magnets to pursue stable, steady-state fusion.

Project Infinity gives the strategy a physical address. At Tennessee Valley Authority's retired Bull Run coal site in East Tennessee, Infinity One is planned as an engineering verification and workforce-training platform. Infinity Two is designed as a 400 MWe commercial plant, with Type One Energy targeting grid operation by 2034.

The coalition is deliberate. Tennessee issued Project Infinity a fusion-specific byproduct material license in August. TVA is evaluating potential deployment. AECOM is supporting engineering, while Siemens Energy's venture arm joined the cap table alongside Breakthrough Energy and Lowercarbon Capital.

The partnership model can reduce the amount of infrastructure Type One Energy has to own. It also moves the commercial risk into the handoffs. Supplier quality, component integration, plant financing, licensing, construction discipline, and TVA's final approvals all have to arrive in sequence.

This round buys more than reactor work. It buys Type One Energy time to prove that fusion can be organized like an energy industry before it has earned the right to operate like one.

Multiply Labs

A therapy can survive discovery, clinical testing, and regulatory review, then meet a factory built for a different kind of medicine.

Multiply Labs just raised a $75M Series B to rebuild that handoff with robots. Dr. Patrick Soon-Shiong with NantWorks led. AstraZeneca, Lingotto, Teradyne, and Strange Ventures joined new; Casdin Capital, Lux Capital, Fifty Years, Ora Global, and Founders Fund returned.

AI can propose more therapies. A living medicine still has to move through instruments, sterile transfers, quality records, trained hands, and a manufacturing process that cannot behave like a beta test.

CEO Fred Parietti came from MIT robotics. CSO Alice Melocchi came from pharmaceutical science and the MIT-Novartis Center for Continuous Manufacturing. The co-founders built Multiply Labs where those disciplines stop being useful separately.

Its clusters connect equipment pharmaceutical companies already use. Robotic arms and custom modules handle unit operations while software coordinates instruments, captures sensor data, and sends records into a customer's quality system.

Imitation learning trains the robots on workflows demonstrated by scientists. Drugmakers have spent years validating their processes. Multiply Labs is trying to automate the work without making the customer abandon what it already trusts.

The pharmaceutical or biologics company owns and operates each cluster. Multiply Labs has to deliver dependable capacity inside somebody else's regulated operation, not merely produce a good robotics demo.

The company reports a 74% reduction in cost per dose and up to 100x more throughput per square foot. A 2024 peer-reviewed study supports the concept, but its disclosures include company employment, founder equity, and company-supported research. The figures show promise, not a universal commercial scorecard.

The new capital will expand capacity, speed the roadmap, and grow engineering, regulatory, and commercial teams. Multiply Labs is moving toward commercial-scale production while expanding beyond cell and gene therapy into antibodies, viral vectors, and mRNA.

The next phase lives in ordinary factory moments: whether a robot handles an exception, whether a record survives review, and whether a customer can add capacity without rebuilding the process around it.

Multiply Labs has spent years teaching robots how scientists manufacture advanced therapies. This round gives the company room to teach an organization how to trust those robots with the work.