Physical Infrastructure Capital Shift
Seven of the ten largest venture rounds disclosed on September 21, 2026, went to companies that build or operate physical assets: water infrastructure, semiconductor capacity, nuclear reactors, residential batteries, satellites, robotics hardware, and power electronics. A founder closing a software only round this quarter is competing for capital against a market that has started pricing physical assets alongside computation, and the financing structures, diligence questions, and term sheet provisions that come with physical infrastructure capital are now showing up in software negotiations as well.
A. Executive Summary
On September 21, 2026, seven of the ten largest venture rounds, worth roughly $442 million combined, funded physical infrastructure companies such as Rainmaker and Kairos Power, not AI application software. Rainmaker closed a $100 million Series B for drone based cloud seeding, Xingyun Integrated Circuit raised roughly $110 million for domestic graphics processor capacity in China, and Kairos Power took $70 million in equity from Samsung C&T, which separately committed up to $100 million in engineering services toward a reactor being built for Google in Oak Ridge, Tennessee. Amber Electric closed a $56.2 million Series E for residential battery automation and HEO closed a $25 million Series B for satellite imaging, and two smaller rounds, Zhongyi Embodied Intelligence for humanoid robotics hardware and Jierendi Electronic Technology for power electronics, complete the physical infrastructure group.
Only three rounds funded companies whose product is primarily software: Corridor, an artificial intelligence native health benefits brokerage; Kheops, a French retail assortment intelligence platform; and Sprive, a United Kingdom mortgage overpayment application. Each was smaller than every physical infrastructure round except HEO’s, and each owns a narrow, licensed, or contractual workflow instead of a general interface layered on a foundation model, which is the trait that separates a funded software round in September 2026 from an unfunded one.
For a founder raising a software only round this quarter, the commercial lesson is not that software funding has stopped. Capital now flowing to physical infrastructure is priced, structured, and diligenced using strategic corporate investors, government contracting relationships, and non-dilutive engineering support as substitutes for pure equity risk, and generalist investors are importing those same diligence habits into software term sheets this quarter.
Detailed Findings
How strategic investors are pricing physical infrastructure differently
Samsung C&T’s investment in Kairos Power shows a financing structure that a pure software round almost never uses: cash equity paired with a separate, non-cash commitment of in kind engineering labor. Samsung C&T put $70 million of equity into Kairos and, apart from that equity check, agreed to provide engineering services worth up to $100 million toward construction of the Hermes 1 and Hermes 2 reactors in Oak Ridge, drawing on the roughly dozen nuclear plants Samsung C&T has built or helped build worldwide. The reactor output counts toward Kairos’s separate commitment to supply Google with approximately half a gigawatt of power by 2035, a commitment that predates this specific financing and gives the equity check a revenue anchor a typical seed or Series A software round does not have.
The pattern is not new to nuclear. Samsung C&T made a comparable investment in NuScale Power in 2021, pairing capital with construction expertise for small modular reactor deployment. What is new in September 2026 is the frequency with which this equity-plus-services structure is showing up across a single day’s largest rounds, which tells a founder evaluating a strategic term sheet that the corporate investor’s engineering team, procurement relationships, and information rights are now a negotiated line item alongside price per share, not an afterthought.
A founder who signs a term sheet with a strategic corporate investor, even in software, should expect the same instinct: a request for board observer rights, a right of first negotiation on a commercial agreement, or a services or reseller arrangement bundled into the financing. None of those provisions are inherently hostile to the company, but each one narrows a future acquirer’s field of view and belongs in the same negotiation as valuation, not as a side letter signed after the term sheet is final.
Where government contracting money is substituting for pure equity risk
HEO’s $25 million Series B drew backing connected to Australia’s National Reconstruction Fund after the company had already secured a contract reported at more than $100 million, a sequence in which a government contract came first and equity capital followed to expand delivery against work already under contract. That ordering matters commercially: a company that can show a signed government contract before its next equity round is negotiating from a position where the investor is financing known revenue, not projected revenue.
The federal government’s own machinery for this substitution in the United States runs through the Small Business Innovation Research and Small Business Technology Transfer programs administered under SBIR.gov policy directives, through Other Transaction Authority agreements available through the Defense Innovation Unit and AFWERX, and through procurement postings on SAM.gov. A software company selling into a regulated industry, critical infrastructure operator, or public sector buyer has the same option HEO used: a Phase II SBIR award or an Other Transaction Authority prototype agreement converted into a production contract before the next priced round, which lets the founder walk into a Series A with contracted revenue instead of a pipeline.
Why export control exposure separates hardware rounds from software rounds
Xingyun Integrated Circuit’s roughly $110 million raise, backed by CATL linked Puquan Capital, was explicitly framed around building domestic graphics processor capacity inside China, a framing that only makes commercial sense against the Bureau of Industry and Security’s October 2022 controls on advanced computing chips and semiconductor manufacturing equipment, later tightened through successive rule packages restricting the export of advanced logic chips, chip design tools, and fabrication equipment to China. Capital raised for domestic chip capacity in that environment is priced against a regulatory wall, not against a competitive market alone.
A software only company rarely faces an Export Administration Regulations classification question, a Committee on Foreign Investment in the United States filing, or an International Traffic in Arms Regulations determination unless its product touches encryption controlled beyond the mass market exemption, trains or deploys models for a foreign government end user, or sells into a regulated defense or intelligence customer. The commercial takeaway for a founder raising this quarter is that a hardware or deep tech company’s cap table already carries diligence costs a software company does not, and investors who have spent September pricing those risks into infrastructure term sheets will ask export control and foreign ownership questions of software companies too, even where none of those regimes actually apply to the company at all.
What the valuation data actually says about the AI application layer
Artificial intelligence companies captured approximately 86 percent of the $412.7 billion deployed in United States venture capital during the first half of 2026, or roughly $355.9 billion, based on figures published on July 9, 2026. That figure is frequently read as proof that software is thriving, but the same publication notes a large share of the dollar total sat with two frontier model labs instead of spreading across application layer companies, which is consistent with a single day, September 21, producing only three application software rounds against seven physical infrastructure rounds.
Jasper’s trajectory is the clearest completed example of what happens to a thin application layer once a foundation model vendor closes the gap. Jasper’s internal valuation fell from $1.5 billion in October 2022 to roughly $1.2 billion by 2024, a $300 million markdown, after OpenAI’s November 2022 release of ChatGPT eliminated the writing assistant’s core differentiation and Jasper’s revenue fell from about $120 million in 2023 to about $55 million in 2024. A founder pitching a general purpose interface over someone else’s model should expect a diligence question built directly on that history: what happens to this company’s revenue the week the underlying model vendor ships the same feature natively.
What still gets funded in software and why
The three software rounds disclosed on September 21 share one trait the unfunded thin wrapper category lacks: each company owns a narrow, defensible workflow instead of a general chat interface. Corridor raised $25 million from Bain Capital Ventures to build an artificial intelligence native health benefits brokerage for small businesses, a licensed insurance workflow with compliance obligations a wrapper cannot replicate overnight. Kheops raised roughly $15 million to move French grocery retailers from manual assortment planning to data driven local assortment decisions, a workflow embedded in a retailer’s purchasing system. Sprive raised $10 million for a mortgage overpayment application that sits inside a regulated United Kingdom lending relationship.
In three decades of reviewing venture financings for companies moving technology from a laboratory or a prototype into a paying customer’s hands, the clause that stops a term sheet negotiation is rarely the price. It is the moment an investor asks what happens to the company’s revenue if the underlying model vendor adds the feature for free, and a founder who cannot answer that question with a specific workflow, dataset, or regulatory relationship the model vendor cannot replicate is negotiating from a weaker position than the valuation on the term sheet suggests.
Risk factors for a software only raise closing this quarter
Google vice president Darren Mowry said in February 2026 that wrapping thin intellectual property around a foundation model leaves a company exposed to margin compression and commoditization, warning that large language model wrappers and artificial intelligence aggregators face existential threats from native feature releases that can erase a wrapper’s value proposition overnight. The economics behind that warning: a wrapper business pays token costs on every query, which compresses gross margin to roughly 40 percent, against 70 to 80 percent for a typical software as a service business that does not carry that per query cost.
The Series A bar has moved against every software founder regardless of whether artificial intelligence touches the product. The median Series A company in 2026 shows approximately $2.8 million in annual recurring revenue, against roughly $1.2 million in 2021, and institutional investors are asking for trailing twelve month growth above 200 percent where 150 percent satisfied the same investors five years earlier. Median pre-money valuations have compressed to a range of $20 million to $40 million, down from a 2021 median near $55 million, which means a founder raising $10 million today accepts roughly 25 percent dilution where a comparable round in 2021 cost 20 percent.
The specific risk for a founder closing a software only round this quarter is that diligence questions built for physical infrastructure and thin wrapper scrutiny are now applied to every software pitch in the same meeting: a defensibility question borrowed from the wrapper collapse, a revenue contract question borrowed from HEO and government contracting rounds, and a foreign ownership or export control question borrowed from Xingyun and CFIUS practice, regardless of whether the company’s product touches any of those regimes.
Want more industry insights? I publish daily analysis on the intersection of deep tech and venture law.
Open Questions
- Whether strategic corporate investors such as Samsung C&T, Morgan Stanley Investment Management’s 1GT, and CATL linked Puquan Capital will begin asking for the same board observer rights, information rights, and right of first negotiation provisions in software rounds that they already ask for in physical infrastructure rounds, and what that does to a founder’s ability to run a competitive process among multiple acquirers later.
- The exact division between Samsung C&T’s $70 million equity check and its separate up to $100 million engineering services commitment to Kairos Power was reported inconsistently across contemporaneous coverage on September 21 and 22. A founder evaluating a comparable strategic package should insist the equity, the in kind services, and any commercial offtake commitment be valued and documented separately, since only the equity component affects the capitalization table and any future 409A valuation.
- Whether the 86 percent artificial intelligence share of United States venture deployment in the first half of 2026 will hold through the fourth quarter once the two frontier labs that drove most of that figure complete their current financing cycles, or whether the physical infrastructure concentration seen on September 21 continues into the next several quarters of application layer fundraising.
- Whether Xingyun Integrated Circuit’s roughly $110 million raise for domestic chip capacity draws incremental scrutiny under the Bureau of Industry and Security’s evolving Export Administration Regulations for advanced computing, and what that means for United States investors weighing exposure to Chinese semiconductor financings through the Committee on Foreign Investment in the United States review process.
- Whether a software company selling into a regulated industry or public sector buyer can replicate HEO’s sequence of a signed government contract preceding its priced equity round using a Small Business Innovation Research Phase II award or an Other Transaction Authority prototype agreement, and how much dilution that non-dilutive contract revenue actually saves a founder at the next priced round.
Interested in analysis about the intersection of venture capital, tech, artificial intelligence, public policy and the law? Check out my Substack channel. https://theinnovationattorney.substack.com/