Dear Friends and Colleagues,
With the third quarter now behind us, I want to share what we have been building at Z2Sixty Ventures, how our thinking has evolved, and what we are seeing in the markets that shape our decisions. On the firm side, Mike Mehawich joined us as Venture Partner, and we relaunched our website around Throughput Technologies. Our portfolio companies also had a notable quarter: DevRev became a launch partner on the OpenAI Marketplace; ReJoule was named a winner in a US Department of Energy breakthrough contest; Dr. Greg Hitz presented Ion Storage Systems’ path to launch at the Solid State Battery Summit; and Gravitrex was featured in the Japan Times.
The market backdrop confirmed what we described in August. According to PitchBook, US venture investment reached $515.8 billion through September, but 82.7% of it went to AI. Deal value fell roughly 40% in the third quarter compared to second quarter as the largest frontier-lab rounds did not repeat, and only 18 US venture-backed companies went public (58 VC-backed for the year vs. 50 for all of 2025). Two shifts matter most for us: the binding constraint on AI has moved from chips to power, and the gap between AI adoption and AI profit remains wide. Both favor the coordination layer we invest in. We cover the details in our Markets Review below.
Portfolio Update
As in our first-half letter, this update draws only on developments our companies have announced publicly.
DevRev. In our first-half letter we argued that as capital concentrates at the model layer, value migrates to the coordination layer that makes those models usable inside real institutions. DevRev’s September 29 partnership with OpenAI shows that argument playing out. DevRev joined the OpenAI Marketplace as a launch partner, and the announcement makes the logic explicit: a frontier model on its own does not know who is asking, which information applies, or what action is allowed. DevRev’s Computer supplies that layer, drawing on the institution’s own knowledge graph, permissions and approval workflows, so the customer’s data and rules remain the context while the model supplies the reasoning. The partnership also opens a new enterprise channel, since eligible OpenAI customers can contract with DevRev directly and may apply part of an existing OpenAI commitment toward it. When one of the most heavily funded companies in AI partners for the coordination layer rather than building it, the buy-over-build pattern we see in M&A extends to distribution as well.
Ion Storage Systems. In August, CTO and Founder Dr. Greg Hitz presented “Launching Solid-State Batteries” at the Solid State Battery Summit in Chicago. He described how ION’s customer engagement has shifted conversations from the lab to the field: rather than asking “What can the battery do?”, customers now ask “What device designs are now possible with ION’s battery in it?”
ReJoule. ReJoule was selected as a winner in the US Department of Energy Office of Critical Materials and Energy Innovation’s breakthrough contest. The recognition underscores that building a circular economy also requires rapid, reliable diagnostic testing to safely collect, sort, transport, and repurpose batteries at scale. Co-Founder and CEO Steven Chung was also in Frankfurt for the Automechanika conference, where he spoke on the panel “Hidden Business Opportunities in the EV Battery Aftermarket.”
Gravitrex. In July, Gravitrex was featured in the Japan Times, where CEO Kira Burns was interviewed about how the company can help serve an aging population. “We have always seen Japan as having the elements in place because of the demographics,” she said. Its walking-assistant system for gait rehabilitation in post-acute care is fully mechanical, with no battery or motors. Burns said this makes it “much simpler and feasible for us to create cost-effective devices with a smaller footprint” than larger overhead harness systems. Earlier this year, Gravitrex won the Japan Prize at Venture Cafe Global Institute’s Pitch2Tokyo global finals, where judges saw strong potential to sell in Japan, along with a year of workspace at Cambridge Innovation Center’s Japan hubs.
Building the Firm
A refreshed website. We have relaunched z2sixtyventures.com. It reflects how we now describe our work in Throughput Technologies, and its News section will feature both firm updates and our portfolio companies’ public announcements. We hope you will take a look, and we welcome your feedback.
Mike Mehawich joins as Venture Partner. Mike is a fellow Chicago Booth classmate and co-founder of Nextpower, a NASDAQ-listed energy technology company. He brings operator depth to our Energy Orchestration underwriting, helping us judge which companies can actually deploy into utilities, developers and large power users.
Time in the ecosystem. In September, we were back at Chicago Booth, speaking with first-year MBA students and meeting with the Polsky Center for Entrepreneurship and Innovation. Congratulations to our former classmate Jason Starr on his new role at the Polsky Center, helping commercialize breakthrough technologies at our alma mater. We look forward to partnering with him. Since quarter-end, we attended Venture Atlanta in October, where we saw the region’s startups and caught up with ecosystem colleagues at the Cox Cleantech Accelerator and ATDC. In the weeks ahead, we will be at Tough Tech Week and traveling to Asia. These are the rooms where the coordination layer is being built, and where we continue to meet the founders and co-investors building it.
Earned Conviction: What We Updated This Quarter
On exits, the dual track is the point. We expected the reopened IPO window to begin pulling mid-cap infrastructure names behind the frontier listings. It has not yet. But the quarter’s largest acquisitions were IPO-scale outcomes: Hugging Face, World Labs and OpenRouter together represent about $28.6 billion of announced or proposed deal value, and any of them could plausibly have been a listing candidate. Strategic buyers are paying IPO-sized prices for coordination-layer capability rather than waiting for it to reach the public markets. That is the outcome our Endgame Framework is built to underwrite: companies valuable enough to go public, and strategic enough that a buyer will not wait for them to.
On energy, conviction has strengthened. In August we called the energy constraint a tailwind. This quarter it became the bottleneck. When grid connections take years and load arrives now, software that coordinates flexible demand, storage and on-site generation is the fastest new capacity available, and incumbents are already paying for it.
On capex-heavy AI infrastructure, we are more careful. Free cash flow compression at the largest cloud providers, component price inflation and the depreciation debate all raise the bar for venture-stage bets that depend on the buildout continuing at today’s pace. We prefer layers that win whether that spending accelerates or digests.
Markets Review
We read these markets through the lens of our experience advising on M&A and capital raises and operating inside a NASDAQ-listed energy technology company: who will buy our companies, at what price, and through which window. We start with the questions we said we would watch in August.
What We Said We Would Watch
The third quarter answered two of the three.
| What we said we would watch | Q3 evidence | Our read |
|---|---|---|
| Whether the reopened exit window pulls mid-cap infrastructure names behind the frontier IPOs | Only 18 US venture-backed IPOs in Q3, most in healthcare; 992 unicorns worth $5.7 trillion still waiting; one acquisition (SpaceX–Anysphere) was 53% of US exit value | Not yet. Anthropic expected in November, OpenAI in 2027 |
| Whether AI’s energy constraint forces orchestration toward the center of the market | Interconnection queues of about seven years in Northern Virginia; the largest data center developers slowing capex on grid limits; grid technology’s best first half on record | Yes |
| Whether buy-over-build keeps bringing acquirers to earlier-stage capability | Nvidia’s $12.9 billion bid for Hugging Face (late stage in 2025); AMD for World Labs ($8.2 billion vs Series C in 2/26); Stripe for OpenRouter ($7.5 billion vs Series B in 5/26) | Yes, though most targets were late-stage |
Sources: PitchBook, Relve, Avanza Energy, Crunchbase.
The two dynamics we flagged also held up: adoption has outrun profit for a second straight year, and power has replaced chips as the constraint every buyer of compute plans around.
Venture: Record Totals, a Narrowing Base
Global venture funding totaled $159 billion in the third quarter, according to Crunchbase, down 25% from the second quarter but up 53% year over year. AI’s share fell to 64% from 80% in the first quarter, not because AI cooled but because no single lab raised tens of billions; instead, a record 27 companies raised $1 billion or more.
Seed is a barbell. Of $13 billion in global seed funding, $2.6 billion went to “seed” rounds of $100 million or more. On Carta, the average US pre-seed round rose 27% to about $276,000, across fewer deals. Multistage funds now take a record 17.3% of Series A deals and 37.8% of Series A dollars, yet remain largely absent at pre-seed and seed. For a first-check investor, that means thinner competition at entry and more buyers for the next round.
Capital is moving toward our markets. Data centers, semiconductors, energy and robotics, the physical layer our pillars sit on, each raised at least $10 billion in the quarter, according to Crunchbase. Climate tech, a proxy for Energy Orchestration, raised $26.1 billion in the first half, up 55%, with low-carbon data center technology rising from 3% to 34% of the total and grid technology posting its best first half on record. On the data side, Databricks raised $5 billion, and regional model and compute players such as Mistral AI and Nscale each raised more than $3 billion.
Beyond the Bay Area. AI pulled pre-seed capital toward California, which took 52% of US pre-seed dollars on Carta in the second quarter. But the South kept gaining, from 16% to 18%, while the Northeast fell from 27.4% to 18%. Our presence in Atlanta, Chicago and San Francisco lets us meet founders in markets with fewer seed funds and less AI-driven price inflation.
Firepower and Exits
Venture capital is plentiful but concentrated, and the concentration is thinning the seed layer. Firms on their fourth fund or later took a record 88.2% of the $108.5 billion raised by US venture funds through September, and PitchBook data show just 211 emerging firms closing a fund this year, down from 927 in 2022. Company formation is moving the other way: an estimated 8,625 US startups raised their first venture financing through the third quarter, a record. The largest firms do write seed checks, but mostly large ones into consensus AI rounds. Outside that lane, more founders are seeking first checks from fewer seed specialists, which is the gap we are built to fill.
Liquidity remains the industry’s problem: 992 unicorns worth a combined $5.7 trillion await a path to liquidity. OpenAI and Anthropic have filed confidentially to go public, though OpenAI has reportedly pushed its timing to 2027, so for now strategic M&A is doing the work. The quarter’s largest deals fit the buy-over-build pattern we described in August: Nvidia’s $12.9 billion bid for Hugging Face, AMD’s planned acquisition of World Labs, and Stripe’s purchase of OpenRouter. What these buyers paid is telling. In each case, it was the layer between a platform and the developers or customers who use it, not the underlying hardware or model. That is why our Endgame Framework asks, at seed, which buyers would need a company and what they could afford to pay, before we ask whether it could go public.
Secondaries are becoming a third path. ElevenLabs ran a $300 million employee tender at double its February valuation; G Squared closed a $2.3 billion fund largely focused on secondaries and tenders; TrueBridge more than doubled its venture secondaries fund to $508 million; and Goldman Sachs filed to launch a fund that buys venture secondaries. Pricing is two-speed: sought-after AI names trade at or above their last-round valuations, while most others trade at a discount. For us, secondaries are a way to return capital from breakout positions ahead of an M&A exit, not a substitute for one.
Public Markets: The Payback Question
The S&P 500 gained 2.3% in the third quarter after its best quarter since 2020, while the Russell 2000 fell 7.2%; the S&P 500 and the Nasdaq then hit records in early October on expectations of roughly 29% earnings growth. Underneath, investors are separating AI spending that pays back from spending that does not. Software has been repriced on fears that AI agents undercut per-seat pricing, and Oracle’s shares were down 36% for the year as its free cash flow turned negative.
Adoption without profit, yet. In McKinsey’s 2026 survey, nearly nine in ten organizations use AI, and 80% of respondents say it makes them more productive. Yet only 37% attribute any EBIT impact to it, unchanged from last year. Index-level margins are near records, so the gains are real, but they are concentrated among AI suppliers and early adopters. Most companies have the tools but have not rebuilt their workflows around them. That gap is the Human-System Bridging opportunity.
Pipes first, rails later. The AI product-and-adoption story resembles early programmatic advertising more than most people want to admit: enormous capability, enormous hype, and almost no infrastructure to support either responsibly at scale. The ability to identify and target audiences arrived well before anyone solved brand safety, measurement, or fraud. The pipes got built fast. The rails came later.
We are in the pipes-first moment with AI. The models are extraordinary. The surrounding layer — identity governance, monitoring, measurement, pricing accountability, and safety at the system level — is still being invented. Gartner expects AI agents to outnumber sellers ten to one by 2028, yet predicts fewer than 40% of sellers will say agents have improved their productivity. In McKinsey’s 2026 survey, the share of organizations scaling agents tops out at 12% in any single business function and is below 10% in most. The pipes are built. The rails are not. That layer is where our Data Sovereignty Tools and Human-System Bridging pillars sit. We have written about both the system-level governance dimensions and the product-experience gap in more depth on our site: AI Governance and AI Disclaimers.
Data centers: spending ahead of returns. Global data center capex is forecast to top $1 trillion this year, though much of the 92% second-quarter increase reflects memory prices more than doubling rather than added capacity. Alphabet posted its first negative free cash flow quarter as a public company. The largest cloud providers are expected to add about $1.57 of capex for every $1 of new operating cash flow through 2027, and investors are debating whether AI server depreciation schedules overstate profits.
The physical limit is now power. Interconnection queues in Northern Virginia run about seven years, and Texas’s grid operator was tracking roughly 474 gigawatts of large-load requests in June. Wood Mackenzie expects the largest developers to slow capex this year for the first time since 2023 because of grid limits.
Since quarter-end, the financing has started to crack. CoreWeave’s bonds, issued in June at 9.675%, traded out to roughly 15% in mid-October as investors singled out leveraged borrowers. Nvidia-backed Firmus pulled its planned A$7 billion Australian listing amid concerns over its price and about US$30 billion of expected debt. And Anthropic moved its IPO to November after third-quarter revenue reportedly grew 45%: strong, but well short of the prior quarter’s doubling. Our read is that the repricing is arriving through credit before equity, as lenders separate borrowers with contracted power and customers from those that have borrowed ahead of both.
Echoes of 2000, with one difference. In 1998–2001, carriers spent more than $100 billion laying US fiber, financed by debt and optimistic forecasts, while hosting companies built data centers ahead of tenants who later vanished; Global Crossing, WorldCom, and Exodus Communications, once valued at about $32 billion, all failed. Today’s leveraged, single-purpose builders carry similar risk, and GPUs lose value far faster than fiber or buildings. The difference is the grid. In 2000, a data center could be built on a forecast and then sit empty; today, facilities cannot be energized faster than utilities can connect them or than turbines and transformers can be delivered to power them on-site. That caps how much powered capacity can be overbuilt and shifts the risk from too much capacity to too much debt. Either way, the scarce asset is getting more useful work out of constrained power, which is the coordination layer we back.
Looking Ahead
Into the fourth quarter we will be watching three things: whether the largest cloud providers hold their capex guidance through late-October earnings, whether the frontier-lab IPO filings finally widen the window for infrastructure companies behind them, and whether the gap between AI adoption and AI profit begins to narrow. We expect the answer to the last to depend less on better models than on better coordination, which is where we have chosen to build.
As always, we welcome your questions on our strategy, our portfolio or the broader venture landscape, and we are grateful for your continued interest in what we are building.
With appreciation,
| Hernando Bunuan | Zachary Debelak | Mike Mehawich |
|---|---|---|
| Founder & Managing Partner | General Partner | Venture Partner |