When the Weather Turns

SiMa.ai

A robot does not get to pause the factory floor while the cloud thinks. The machine still has to see, decide, and move under a tight power budget when the network cannot help.

That control loop is the market SiMa.ai is financing. The San Jose company raised a $150M Series C at a company-stated $1.45B valuation, with Fidelity Investments and Amplify Partners co-leading. Alter Venture Partners, Dell Technologies Capital, Maverick Capital, +ND Capital, Point72, and StepStone Group participated, while AllianceBernstein, Baron Capital, J.P. Morgan, and the State of Michigan joined as new investors.

Krishna Rangasayee founded SiMa.ai around a stubborn semiconductor truth: the customer experiences the software, but the physical machine lives with every decision made in silicon. A cloud benchmark still has to fit inside a drone, robot, vehicle, camera, or industrial system where power, heat, latency, safety, and integration collect their bill.

SiMa.ai is trying to make that bill smaller. Its Palette Neat environment gives developers an agentic software layer for building Physical AI applications, while Modalix provides the purpose-built MLSoC hardware underneath. SiMa.ai says it can move development from months toward days or hours while preserving much of a customer's existing code. That claim identifies the commercial problem clearly: changing the compute platform cannot require rebuilding the product around it.

The new capital is intended to scale Palette Neat and fund SiMa.ai's next generation of machine-learning IP, chiplets, and systems-on-chip. The roadmap targets 1,000 dense TOPS and first-half 2028 availability, with medium- and high-end drones, humanoid robotics, ADAS, and AI-powered cockpits in view.

The traction is promising and still needs context. SiMa.ai says revenue quadrupled between 2024 and 2025 and names Bosch, Emerson, Intrinsic, Micron, Synopsys, TRUMPF SE, and others in a combined customer-and-partner network. The company has not disclosed audited revenue, unit shipments, margins, or a customer count.

That leaves SiMa.ai with a more interesting burden than proving an edge accelerator can run a model. It has to become the platform engineering teams trust across machines that fail differently, ship on different schedules, and leave very little room for power or software mistakes.

The $150M buys time, talent, and silicon. Krishna Rangasayee's harder job is making the migration feel ordinary before the hardware carrying it becomes anything close to ordinary.

Deeper Analysis 👇 https://devcuration.com/articles/sima-ai-raises-150m-series-c-physical-ai

HiringCafe

Job search looks like a matching problem until the listings go stale. Then it becomes a trust problem with a search bar attached.

HiringCafe built around that distinction. Instead of waiting for employers or recruiters to submit openings, the company crawls career pages directly, cleans the data, refreshes listings several times a day, and gives job seekers filters designed around the work they are actually trying to find.

That simple experience hides a loud infrastructure bill.

Vasona Technologies, Inc., the company behind HiringCafe, filed a Form D for a $6.695M equity offering. The filing says $6.194995M had been sold to 25 investors as of January 14, 2026, with $500,005 remaining. HiringCafe separately confirmed that it had accepted funding to extend its runway and hire.

The intake headline called this a $6.8M Pre-Seed round. The primary record is more precise and less convenient: it does not identify the stage, investors, lead, or valuation. An offering amount is not automatically cash received, and a familiar round label does not become true because it fits neatly in a database.

Ali Mir and Hamed Nilforoshan started with a job seeker complaint the incumbents had learned to monetize without fully solving. Open roles live across thousands of employer systems. Sponsored inventory, duplicate postings, stale listings, agency feeds, and search filters built for volume can leave the applicant doing the quality control.

HiringCafe moved that work into the product. In September 2025, Ali Mir reported 1.3M+ monthly active users, a 68K+ Reddit community, and $0 spent on marketing. By January 2026, the company said its systems were crawling 130,000+ employer websites roughly 3 times a day. Those are company-reported figures, not audited market share, but they show where the expense lives.

Freshness is a recurring obligation across crawlers, models, filters, and infrastructure.

The funding came with a public promise. HiringCafe said it rejected most investor interest and chose backers who would not force a monetization path that damaged search. A free job-search product needs enough capital to improve the data without turning the user's urgency into the inventory.

Capital gives HiringCafe more time to hire, widen coverage, and make the source data cleaner. It also makes the investor-alignment claim measurable. Every sponsored shortcut, stale result, and forced compromise will be judged against the reason users arrived in the first place: they wanted the job, not the job-board business model.

Deeper Analysis 👇 https://devcuration.com/articles/hiringcafe-files-6-7m-equity-offering-for-job-search

Mona

Small-business owners are expected to make capital decisions like finance teams while running payroll, serving customers, chasing invoices, and doing whatever broke before lunch.

The money may exist. The time, map, and trusted person to help decide whether it fits usually do not.

Mona has raised an oversubscribed $3.5M round led by Sandberg Bernthal Venture Partners, with Alumni Ventures, Wisdom Ventures, Eric Schmidt, and other angels participating. The company is also backed by the Mastercard Strive USA Innovation Fund.

Co-founders Andrew Leon Hanna, CEO, and Anny Dow, CPO, are building Mona around the part of small-business finance that rarely gets its own department. An owner may need to compare a loan with a grant, judge repayment pressure against cash flow, and submit the same business story through another application. Larger companies hire people for that work. Small companies assign it to the person already doing everything else.

Mona starts with a common application, uses AI to identify relevant loans, grants, and public programs, and helps entrepreneurs prepare applications. Then the model gets more interesting: software does not get the last word. Human coaches stay involved as the business changes and new decisions arrive.

Capital discovery and capital judgment are different jobs. Search can surface 20 possibilities. It cannot make the consequences disappear when the owner chooses one. The Federal Reserve's 2026 Small Business Credit Survey found that 60% of employer firms sought financing in the prior year. Only 42% received all they wanted; 36% received some or most, and 22% received none. The money often covers operating expenses or expansion: inventory, hiring, equipment, rent, and the next opening a business may not get twice.

Mona says it has supported thousands of small businesses and helped secure millions in affordable capital across several U.S. cities. Those figures are company-reported, but the operating thesis is clear: the smallest companies face consequential financial choices with the thinnest advisory bench.

The new capital will expand Mona's AI platform and financial coaching. The company did not disclose a round label, valuation, or detailed hiring plan. The useful question is what Mona can make repeatable without making the guidance impersonal.

Every funding product has terms. Every business has a different week. Mona is building for the space where those two realities meet, with software carrying the search and people still accountable for the judgment.

Deeper Analysis 👇 https://devcuration.com/articles/mona-raises-3-5m-in-funding

Presto Phoenix

“Actually” separates a voice demo from restaurant infrastructure.

Actually, make that a large. Actually, no pickles. Actually, the app shows a promotion. One correction can touch speech recognition, menu logic, pricing, the POS, and the kitchen screen.

That is the operating problem behind Presto Phoenix’s $10M financing, led by Metropolitan Partners Group with participation from Remus Capital, Link Ventures, and angel investors including the CEO of ElevenLabs.

Presto offered $15M of equity, sold $9,999,877 to 16 accredited investors, and recorded the first sale on December 23, 2025. The company announced the $10M on January 8, without naming a round series or valuation.

Presto Phoenix is also a reset, not a clean continuation of the old public-company capitalization story. The private entity was formed after acquiring Presto Automation’s Voice AI platform, technology, and team through an Article 9 sale in December 2024. That transaction brought $18M of new equity capital. On a filed basis, the Phoenix entity has now disclosed $28M of equity sold across its first 2 offerings.

Co-founder and CEO Krishna Gupta is building around a narrower premise: restaurants do not need another impressive voice demo. They need software that can survive hundreds of stores, regional menus, changing promotions, accents, background noise, deep modifiers, and the integrations holding the order together.

CTO Bill Healey and the Presto team have built multiple operating modes into Presto Voice, from pure AI to supervised and agent-led ordering. Menu Unification centralizes restaurant menu data so brands can update products and promotions across locations without rebuilding the operating logic store by store.

Presto says the platform served millions of drive-thru interactions in 2025 and is live across hundreds of locations for 12 enterprise brands. Those are company-reported metrics, but they describe the standard this capital now has to support: production behavior, not a controlled demonstration.

Restaurant voice AI has already shown the market how public its failures can become. An error in a dashboard waits for a ticket. An error at the speaker gets repeated by a hungry customer while a line of cars watches the clock.

The $10M gives Presto more room for deployments, product development, and the broader restaurant automation platform it wants to build. The value will be decided in the ordinary corrections where the customer changes one thing and every system behind the speaker still has to understand what “actually” means.

Deeper Analysis 👇 https://devcuration.com/articles/presto-phoenix-raises-10m-in-funding

F2

David Golub is putting Golub Capital's investment judgment into the system the firm just backed with $5M. At more than $90B of capital under management, that choice carries weight.

Golub Capital is making a strategic investment in F2 and plans to deploy the platform across its investment workflows. The partnership is designed to connect Golub's proprietary data, accumulated experience, and underwriting process to AI systems while keeping that institutional knowledge under the firm's control.

Foundation models keep getting faster, cheaper, and more capable. A credit manager's durable advantage lives in the deals it declined, the structures it accepted, the risks its investment committee challenged, and the people who remember why one borrower survived a cycle while another did not.

F2 was built around that gap. Don Muir, CEO and Co-Founder, and Emre Kazdagli, CTO and Co-Founder, have built an operating layer connecting data rooms, models, credit files, memos, market data, and prior deals. Its agents support screening, diligence, underwriting, and portfolio monitoring, with Excel-native workflows and an audit trail to source material.

F2 reported in June that more than 100 private-credit funds and commercial banks were using the platform, representing more than $400B in client-managed assets. Those are company-reported operating metrics, not proof of investment returns, but they explain why deployment depth now matters more than demo quality.

Golub brings the pressure test. At its scale, useful AI has to respect permissions, preserve lineage, fit established risk processes, and survive the scrutiny of people paid to find what everybody else missed. The decision reaches past the smartest model this quarter to whether the firm's knowledge stays usable and defensible when that model changes.

The partners are also forming the Global Credit AI Council, with David Golub as inaugural Chair. Its members are still to come, but the forum will put senior credit executives around the adoption and governance questions carried by the people holding the risk.

F2 disclosed $24M in total equity funding in June, including a $14M seed round led by HighlandX. This $5M strategic investment arrives with a customer, an implementation environment, and proximity to the operating questions shaping AI in credit. Valuation and detailed terms were not disclosed.

The deeper asset may be Golub's willingness to make its history part of the build, then keep it under institutional control as the models continue to move.

Deeper Analysis 👇 https://devcuration.com/articles/f2-lands-5m-strategic-investment-from-golub-capital

Feather Robotics

Every physical-AI startup has a second job hiding inside the first one.

The team may be building a model for restaurant work, laboratory cleanup, manufacturing, or another physical workflow. Before that application can earn a dollar, somebody still has to source motors, design arms, choose cameras, integrate compute, manage batteries, support failures, and turn a prototype into hardware a customer can actually use.

Feather Robotics raised $7.6M in Pre-Seed financing to take more of that repeated job off the developer's desk.

Gradient Ventures led the SAFE round, with Builder Capital, Geometry, SEED Innovations, and Virgo VC participating. SEED's $1M commitment is included in the round and carries a $60M valuation cap. That cap describes its conversion terms, not a priced Feather valuation.

CEO and co-founder Hoa Mai and CPO and co-founder Parsa Bakhtiari built Feather around a specific handoff: provide the robot body, modular hardware, support, and open SDK so other teams can spend their scarce engineering time on the intelligence and application.

The current system starts at $29,990. It uses two 7-degree-of-freedom arms, a vertically adjustable torso, a holonomic base, configurable Nvidia Jetson compute, dual hot-swappable batteries, and an open Python and ROS2 SDK.

Feather says it built its first prototype in 2 months, fulfilled its first order in 9 months, and has crossed $1M in revenue. The company reports robots working in manufacturing, food service, and laboratories, including for an unnamed manufacturer with $4B in revenue. Those are company-reported metrics, but they show where the story has moved: beyond a staged demo and into the less photogenic work of deployment.

That is also where Feather's platform thesis gets tested.

The physical-AI market can produce better models faster than most application teams can reproduce hardware, field service, replacement parts, and support logistics. If each new developer has to become a robot manufacturer before becoming an application company, a lot of promising intelligence will stay in simulation.

Feather is betting the body can become shared infrastructure. The $7.6M gives the company room to scale production, hiring, support, and a larger product launch while developers decide whether a configurable American platform can carry their models into paid work.

The next proof will come from the applications that no longer have to spend their first year rebuilding the machine underneath them.

Deeper Analysis 👇 https://devcuration.com/articles/feather-raises-7-6m-in-pre-seed-funding

Outmarket AI

Insurance still sells through people. The work around those people still moves through PDFs, agency-management systems, carrier portals, email threads, and policy language that has to be correct before a client ever sees the result.

That is the wager behind Outmarket AI's $34.5M Series B. SignalFire led the round, with Fika Ventures, Permanent Capital Ventures, TTV Capital, and Dash Fund participating. It arrives roughly 4 months after Outmarket announced a $17M Series A.

Vishal Sankhla and Anshu Jain built Outmarket around a useful distinction: an insurance professional's judgment is valuable; rekeying the same client and policy context across 5 systems is expensive theater.

Outmarket says more than 10,000 active users across 300-plus agencies now use the platform, including more than 25% of the Top 100 insurance agencies. Those are company-reported numbers, but the speed matters. Insurance agencies are buying AI where the output has to survive a real workflow, a real client, and real E&O exposure.

The new Certificates of Insurance workflow makes the point. It reads a contract or lease, extracts the insurance requirements, checks them against the policies inside the agency-management system, flags gaps, and prepares the ACORD certificate with the right holders and endorsements.

A certificate is one document to the client. To the agency, it is a chain of obligations that has to connect the contract, policy, coverage, endorsements, account record, and deadline without dropping the part that becomes somebody else's problem later.

That is why Outmarket's ambition is larger than document automation. The platform is trying to become an intelligence layer across commercial, employee benefits, personal lines, and specialty insurance, with the same underlying context available to policy checking, proposals, loss runs, forms, market search, and custom workflows.

SignalFire moving from participant in the Series A to lead investor in the Series B is its own signal. Fast adoption earned more capital. More capital now raises the standard from useful software to dependable operating infrastructure.

Outmarket plans to extend its capabilities to carriers later this year. If that works, the company moves closer to the handoff where agencies assemble risk, carriers evaluate it, and both sides spend time translating the same account through different systems.

The insurance relationship stays human. The machinery around it is finally being asked to remember what the people already know.

Deeper Analysis 👇 https://devcuration.com/articles/outmarket-ai-raises-34-5m-to-scale-insurance-ai

Quartermaster

A ship's lookout sees a vessel before a spreadsheet, satellite pass, or self-reported signal can explain what it is doing.

That difference between physical presence and remote inference is the commercial opening behind Quartermaster. Neil Sobin built the company around a simple piece of maritime reality: millions of working vessels are already moving through the places where governments, fleets, insurers, energy operators, and autonomy companies need better information.

Quartermaster has announced $140M in new financing, and the accounting matters. Insight Partners led a $100M Series B, with new participation from Overmatch Ventures and returning investors First Round Capital, Quiet Capital, Steel Atlas, and TMV. Stifel provided a separate $40M venture-debt facility. It follows a $43M Series A closed in May.

SmartMast is the hardware underneath the thesis. Quartermaster mounts cameras, radios, satellite connectivity, geolocation, and edge processing on commercial and working vessels. Each installation becomes another observing node, feeding real-time data into a network that can identify nearby traffic, flag unusual activity, and preserve verified records.

AIS tells an operator what a vessel chooses to broadcast. Satellites add valuable coverage from above. A ship already on the water can supply a different class of evidence when signals disappear, conditions change, or an event develops between remote observations.

The network is moving beyond a clever diagram. Neil Sobin told TechCrunch that more than 650 vessels in 25 countries are equipped with SmartMast, while more than 800 units have shipped. Each unit records tens of gigabytes per day, and Quartermaster has doubled manufacturing capacity as it prepares for whole-fleet deployments.

Those are company-reported metrics, not audited proof that Quartermaster owns the ocean's intelligence layer. They do show why this round is about more than training another model. Someone has to manufacture the sensors, install them, keep them working in hostile conditions, recruit the vessels, and turn unruly observations into information customers will pay to use.

The financing will grow engineering headcount and recruit more vessels across active fishing grounds and commercial shipping lanes. Every new mast adds coverage, but the consequential handoff happens when somebody responsible for a ship, cargo, policy, or rescue can act before the old picture catches up.

Deeper Analysis 👇 https://devcuration.com/articles/quartermaster-raises-140m-maritime-intelligence

Modulate

A transcript can remember every word and still forget what happened.

The caller hesitated before authorizing the transfer. The customer stopped sounding curious and started sounding frustrated. The voice belonged to software, not the executive it imitated. That evidence gets lost when a conversation is flattened into text.

Modulate has raised $25M in new funding led by Future Ventures, with returning investors Hyperplane and Lakestar participating. The company says the round brings total funding to $60M.

That money is financing a claim the voice AI market now has to take seriously: teaching machines to speak solved only half the interface.

Co-founders Carter Huffman and Mike Pappas started Modulate in 2017 around audio itself. Carter Huffman is now CEO, and Mike Pappas is chairman. Their proving ground was live game chat, where people interrupt, joke, threaten, whisper, switch languages, use bad microphones, and expect software to know the difference.

That work became ToxMod and then Velma, Modulate's broader voice-intelligence platform. Its Ensemble Listening Model architecture combines signals from more than 100 specialized models.

Words still matter. So do tone, emotion, intent, emphasis, speaker behavior, background audio, and whether a voice is synthetic. Modulate packages those signals into APIs for transcription, deepfake detection, fraud prevention, voice-agent supervision, customer experience, and trust and safety.

The company reports more than 10M hours of audio analyzed each month and more than 600M hours processed in total. It also points to leading positions on public Hugging Face transcription and deepfake benchmarks. Those are useful signals, not audited performance inside every bank, hospital, contact center, or AI-agent workflow. Deployment will decide the rest.

The $25M will expand AI/ML research, engineering, developer relations, APIs, SDKs, deployment options, and partnerships. The wager: every company building with voice should not need its own audio research lab.

The harder work begins at the handoff. Modulate has to make acoustic intelligence accurate, affordable, explainable, and simple enough for live decisions where a false alarm wastes time and a missed signal can move money, expose a customer, or let an agent repeat the same mistake.

Voice became an interface because machines learned to talk. The next contracts will belong to the systems that can hear what the transcript left behind.

Deeper Analysis 👇 https://devcuration.com/articles/modulate-raises-25m-for-audio-native-voice-ai