Liquid Compute Builds AI Compute Financial Infrastructure

Liquid Compute: The AI Buildout Needs a Financial System to Match When I started my career in the late 2000s, I worked with companies helping enterprises hedge their biggest costs and exposures. Fuel for an airline. Wheat for a food producer. FX for a global business. Different industries, same problem: you could execute well and still watch your margins get crushed by something you didn’t control. Today, compute is bringing that problem to AI. Which chips? Open-weight or managed models? Own the infrastructure or rent it? Which workloads need sovereign capacity? These are technology decisions with very real financing and margin consequences. And “compute will get cheaper” isn’t an answer for the operator that financed hardware against yesterday’s assumptions. Enter Liquid Compute, building the capital markets layer for AI compute: a physical marketplace, pricing benchmarks, and derivatives, while working toward a regulated exchange and clearinghouse. Buyers want visibility into costs. Operators want predictable revenue. Lenders want to manage the price exposure underneath their loans. Liquid Compute is connecting those needs. My close friend and former Index colleague Mark Goldberg at Chemistry introduced me to CEO Ronit Jain. An hour together one Friday evening left me blown away by his technical depth and commercial maturity. He was already proving people would transact, not just pitching why the market should exist. Ronit and co-founder Aarav Patel are Berkeley engineers (go Bears!) who combine technical credibility with real commercial hunger. Ronit brings finance training; Aarav brings the product ambition and ability to pull people into a bigger vision. They’ve surrounded themselves with talent from NVIDIA, Meta, Google Cloud, Jump, and Point72. That’s a big swing. Exactly the kind we want to take with this team. At K8, this sits at the center of our hybrid equity and credit model. AI infrastructure companies need equity to build something new AND financing to deliver against demand that already exists. “Raise another equity round” isn’t always the answer. We focus credit on assets and contract-backed cash flows that can actually be underwritten (think offtake agreements or B300 chips). Through the Koo family’s relationships, we bridge Taiwan’s AI hardware ecosystem and the companies building on top of it. Across multi-chip infrastructure, sovereign compute, robotics, and AI applications, having demand isn’t the same as having the hardware and financing to serve it. Liquid Compute adds a critical piece: better pricing and risk management could make more of that ecosystem financeable. We’re excited to back Liquid Compute's $15m seed round, co-led by Chemistry and FirstMark, alongside longtime friends Mark Goldberg, Bohan Lou, and Adam Nelson. The next phase of AI needs more than access to compute. It needs better ways to pay for it, price it, and manage the risk around it. Let’s go. - The Wall Street Journal article below

OpenAI and Anthropic are already pushing for bond market respectability. Having their cake and eating it too in a way. Barely have IPOs entered the picture than their bankers are seeking an investment-grade rating from S&P, Moody’s and Fitch. Morgan Stanley and Goldman Sachs are pushing the case in order to quickly open the door to abundant, cheaper debt. What for? A gigantic IPO. A mountain of cash raised from equity investors, a temporarily boosted balance sheet, a reassuring rating, bonds issued on the cheap. Then debt pays for GPUs, servers, data centres, electricity and multi-year compute contracts. It’s a party for everyone, at least for a while. Banks collect IPO, lending and bond-issuance fees. Private investors get a more liquid exit. Chipmakers and cloud providers lock in their orders. Partners already burdened with guarantees or capacity commitments get to shift some of the pressure elsewhere. Elsewhere, it’s called the bond market. Insurers, pension funds, credit ETFs, bond managers: the next candidates to finance the great AI build-out. They will not merely finance a technology, but a particularly voracious industrial machine whose costs still run far ahead of proven cash flows. Money raised through an IPO provides oxygen. It produces neither recurring margins, nor free cash flow, nor a sustainable model. A triple-digit private valuation does not pay interest. An annualized revenue run rate guarantees neither actual cash collection, nor profitability, nor refinancing capacity in a stress scenario. In the jargon, this will be called the maturing of the AI market. Basically preparations are under way to migrate risk from the balance sheets of Nvidia, Oracle, Google, Broadcom, banks and private investors to the bond portfolios of millions of savers. Manufacturers sell the chips, hyperscalers sell compute, banks arrange financing, builders put up the data centres, models absorb the infrastructure. Then bonds keep the wheel turning. Ultimately, the achievement may lie less in AI than in the art of turning gigantic industrial outlays, complex guarantees and growth promises into bond paper rated highly enough to become somebody else’s problem at some point.

OpenAI expects to burn $278 billion in cash through 2030 That isn't just an eye-popping headline it’s the core financial thesis driving the entire AI ecosystem. According to leaked pitch documents reviewed by the Financial Times, OpenAI projects a cumulative negative free cash flow of $278B between 2026 and 2030, despite forecasting a massive $840B in cumulative revenue over the same period. Why such a massive deficit? Because OpenAI plans to spend $856 billion on compute and data center infrastructure alone. Building frontier AI is no longer a high-margin software play; it’s heavy infrastructure development. The Numbers Behind the Bet To justify this infrastructure, OpenAI's internal roadmap calls for an unprecedented growth curve: 📈 Revenue trajectory: $36B (2026) ➔ $350B (2030) 💸 Cash depletion timeline: The $122B raised in March 2026 is projected to run out by 2028, forcing continuous capital raises before a prospective IPO. 🎯 Current ambition: Seeking a fresh funding round at a $1.2 trillion valuation. 🎯 New revenue engines: OpenAI projects $100 billion in ad revenue by 2030 to help subsidize model inference. Anthropic & The Capital Race Anthropic isn't far behind. They raised $65B at a $965B valuation with run-rate revenues topping $47B. They've secured $100B+ across 10 years with AWS (targeting 5 Gigawatts of compute) while diversifying into Google TPUs and NVIDIA GPUs. The classic AI question "Who has the smarter algorithm?" is dead. The real game is a closed-loop capital flywheel: Raise mega-rounds ➔ Pre-fund gigawatt power & GPU clusters ➔ Expand model capacity ➔ Monetize enterprise & ads ➔ Service project debt. The Financial Engineering Behind the Boom How do you finance hundreds of billions in hardware without wrecking a balance sheet? Creative corporate finance. Off-Balance-Sheet SPVs: Special Purpose Vehicles backed by private equity (Apollo, Blackstone) to finance data centers and GPU leaseback deals Debt Guarantees & Off-Take Agreements: Securing long-term power purchase agreements (PPAs) and cloud capacity commitments that act as synthetic debt. The entire supply chain AI Labs ➔ Cloud Hyperscalers ➔ Chipmakers ➔ Data Center REITs ➔ Energy Utilities ➔ PE & Debt Markets is now financially bound together. The Ultimate Investor Risk If OpenAI reaches $350B/year by 2030, this capex cycle will look like the buildout of the modern internet grid. The core risks to watch: Model Commoditization: What happens if open-weights or competing models drive inference prices to near zero? Power Grid Bottlenecks: Can utilities supply the gigawatts needed before cash reserves deplete? Fixed Commitment Drag: Data centers and long-term contracts can't be downsized as fast as revenue can fluctuate. The winners of the AI era won't just be the best software engineers they will be the best capital allocators and financial architects. #AI #OpenAI #Anthropic #Finance #VentureCapital #PrivateMarkets #Technology #Investing

The Capital Velocity Trap: Hardware Dependency Escalates The Wall Street Journal confirms Chinese artificial intelligence developer Z.ai just initiated a $5.0 billion US dollar capital raise. They need servers. This exposes a brutal mathematical reality about continuous technical iteration. Z. AI plans to secure $2.0 billion US dollars through a share placement and another $3.01 billion US dollars via zero-coupon convertible bonds. They will issue $21.97 million new shares priced at 714 Hong Kong dollars each. This represents a 9.96% discount against their closing price of 793 Hong Kong dollars on Friday. The zero-coupon bonds mature in 2027. These instruments convert into 26.37 million shares at an initial conversion price of 892.50 Hong Kong dollars. That price carries a 12.6% premium over the Friday close. This package arrives soon after Z.AI, formerly Zhipu AI, absorbed a $4.0 billion US dollar placement in July. The equity tranche secures 15.68 billion Hong Kong dollars. Domestic rivals like DeepSeek, MiniMax, and Kimi (Moonshot AI) are deploying heavy capital into competing systems. The Macroeconomic Shift This fundraising succession reveals extreme dependency on computing capacity. Continuous development of next-generation language models completely escalated baseline hardware requirements. Competitors like OpenAI, Anthropic, and Google set a relentless benchmark. Infrastructure spend dictates operational survival. From an executive vantage point, this transaction highlights a volatile market dynamic. Zero-coupon structures suggest institutional investors are betting heavily on future equity upside rather than yield. Burning through $9.0 billion US dollars in under ninety days for compute infrastructure is unsustainable for standard operators. Hardware dependency is a structural vulnerability. As enterprise clients integrate generative systems to automate manual tasks, foundational builders are trapped. They must continuously purchase rapidly depreciating silicon assets. Broadening the shareholder base diffuses risk, but signals founders need vast liquidity to survive. This trajectory points toward global consolidation. Will escalating compute costs force artificial intelligence developers into heavy sovereign subsidies or complete monopolistic consolidation? I track and break down breaking technology research and market shifts daily. Hit follow so you do not miss the next executive analysis. I am taking 1 new client this month for our Executive Media Strategy Engine. Send me a message to secure the last spot. ✨🚀

Starting my second company. Today I'm officially establishing QAI Labs in Singapore. Quantum technology, AI-native operations, applied to financial markets. My first was Digifinance(formerly TrustVerse), which I started after my MBA at Chicago Booth. Over five years we built across digital wallet private key security, digital wealth management, VC investment and advisory, and quantitative trading. We had real wins, and more importantly, the whole experience rewrote how I think about what business success actually means and how financial and corporate management really work up close. QAI is an AI-native company. We build solutions and services that monetize in financial trading, and over time in quantum compute applied to algorithm enhancement and financial data research. On quantum, I'll be direct: it's still too early for most real-world application. But the technology needs use cases and it needs a path to monetization, and those two things won't appear on their own. My view is that quantitative financial data research is one of the better fits available today. The problems are the right shape — large search spaces, heavy optimization, portfolio and signal work where even a partial edge compounds — and the industry has both the data and the willingness to pay for an edge. That's where I want to be positioned when the hardware catches up. As per quantum, problem formulation, classical algorithm, benchmark line, and simulator access are big opportunities for financial data research. Technology is moving faster than at any point in my career, and I want to be inside that. I spend about ten hours a day building with AI-orchestrated systems, and I haven't been this energized by work in a long time. Two things are already live(Focused on signal accuracy and results from trades): 1> QAI Algorithmic Trading Desk(Prop) — my own quant logic and algorithms for perp long-short positions, traded and backtested over five years. Current focus is digital assets, oil, and securities. 2> Market Structure Desk — liquidation mapping, order book depth, signal win-rate tracking, and confluence scoring. Where the finance use cases are least hand-wavy Portfolio optimization under real constraints, because it's a combinatorial problem classical solvers approximate. Monte Carlo for derivatives pricing and risk, where amplitude estimation has a theoretical speedup. Feature and signal selection across large candidate sets. Scenario generation for stress testing. These are the four that survive scrutiny from someone who knows the field. The goal is fast growth through net-new customers and B2B partnerships. We have four global partners across quantum computing, exchange infrastructure, and data services and more in discussion. More to share as we build. If you're working in this space, I'd like to talk. #QAILabs #FoundersJourney #StartupLife #NewBeginnings #SingaporeStartups #AIEngineering #DeepTech #QuantumFinance

A GPU-hour has an expiration date. Miss its delivery window and the inventory disappears without ever sitting on a shelf. That small fact changes what a compute market has to do. Liquid Compute emerged from stealth with a $15M Seed round co-led by Chemistry and Firstmark. K8 Capital, Night Capital, TrueBridge, Brainchild Holdings, UFO Holdings, Y Combinator, and Dmitry Balyasny also participated. Led by Ronit Jain and Aarav Patel, the New York company is building a regulated exchange for cash and physically settled contracts on AI infrastructure, starting with compute. The harder work is verification, price discovery, clearing, and settlement around capacity that varies by chip, cluster, geography, network, delivery period, contract length, and counterparty. Two agreements can both say “H100” while delivering very different economic exposure. An AI builder may want reliable computation, not a bet on hardware prices. A capacity owner may want utilization and financeable cash flow, not an open position on future demand. Both still inherit price and delivery risk. Buy too early and a company may overpay if supply loosens. Wait too long and the capacity needed for a launch or training run may be gone. Reserve too much inventory and an operator can watch it expire. Lock in too soon and stronger demand can pass by. Liquid Compute wants to make some of that exposure visible and transferable. The regulatory work is equally important. Liquid Compute has filed through PMEX Markets and PMEX Clearing for Designated Contract Market and Derivatives Clearing Organization status. The CFTC lists both applications as pending. Pending is the operative word. The filings show intent and process. They do not make the venue approved, and they do not remove the adoption work ahead. Liquid Compute still has to standardize contracts without flattening the technical details that determine value. It has to verify real delivery, establish benchmarks participants trust, attract enough buyers and sellers for useful price discovery, and manage counterparty risk. It also has to persuade participants that transferring exposure is better than negotiating every contract privately. The round matters because AI infrastructure is becoming too large for procurement teams to carry every pricing and delivery problem alone. Liquid Compute is betting that the next layer of AI infrastructure is not another cluster. It is a market where the risk can move before the inventory disappears. Deeper Analysis 👇 https://lnkd.in/gSwFdCXq #AIInfrastructure #Fintech #MarketInfrastructure #VentureCapital

The Quiet Rise of AI Technical Debt: Your Pilots Are Building Tomorrow's Mess Let me tell you about a company I'll call Northbridge. Its procurement team built a generative AI assistant to summarize supplier contracts, and it worked. Legal wanted a version. Finance wanted one tied to payment terms. Compliance wanted region-specific checks. Within a year, the company had 27 versions of the same AI pattern, each with its own prompts, data sources, access rules and failure modes. When a model upgrade shifted output quality, nobody knew which workflows were affected. When a hallucination showed up, nobody could trace whether it came from the model, the prompt, the retrieval layer or the source document. The original win had become a maintenance liability. I've watched enterprises repeat this pattern for 35 years. Client-server. Web applications. SOA. Cloud. Now AI, only faster. Scary fast. I wrote about this in my latest article: https://lnkd.in/eYUVUvB3 AI technical debt behaves differently from the old kind. It doesn't live in code and infrastructure. It lives in prompts, embeddings, vector databases, fine-tuning decisions, access policies, evaluation sets and shadow workflows that nobody manages like software. A prompt driving a customer support workflow is not just a prompt; it is business logic. A retrieval pipeline feeding a compliance assistant is not just an experiment; it is an information supply chain. A model selection is an operational dependency. Too many enterprises haven't internalized that yet. The most expensive debt is created in the pilot phase, before anyone thinks the system is real. The workaround becomes the architecture. The sample data path becomes the data strategy. The prototype becomes production because nobody wants to slow down a visible win. Then the enterprise congratulates itself on speed and spends the next three years paying interest on decisions made in three weeks. Believe me, I know. I built some of those systems in the '90s, and some are still in production. Here's the uncomfortable part: the worst AI debt is operational, not technical. Who owns the prompt when regulations change? Who validates retrieval sources when a process changes? Who decides whether a hallucination is a defect, a data issue or an acceptable risk? In many enterprises, the answer is nobody. That's how the AI junk drawer forms. Speed without ownership is not agility; it is deferred cleanup. What should you do now? Treat prompts, retrieval pipelines, evaluation sets and model configurations as enterprise assets with versioning, ownership, review and retirement rules. Require every AI pilot to declare its production path before funding beyond experimentation. Build an AI debt register and catalog every model, prompt, data source and agent, then find out which ones are duplicated, unmanaged or unsafe.

Over the weekend I shared a piece about Oracle's "Project Jupiter" financing struggles being potential canary in the coal mine of the AI bubble. This inquiry looks at the weakest link in the chain, the Neocloud's, with Coreweave being the most unstable of them all. With AI investments competing for global cash flow at a staggering rate, the 'cost of money' is most likely to increase, potentially unraveling a very large house of financial cards that will affect anyone connected to the global financial system in ways that are both predictable and unpredictable. https://lnkd.in/e6P34ung

Alphaville: "The share of queries that are truly closed-weight frontier-model-worthy is declining. Back at the start of the year, 3/5s of its queries were routed through to closed-weight proprietary models. The latest share is just 1/4...OpenAI and Anthropic are private companies yet to list or issue rated debt. While totemically and technologically significant, companies at this stage of their lifecycle are really not supposed to be financially important...But their centrality to the industry that they dominate and the degree of circular financing in place means that this doesn’t look nailed on...The leading AI labs are cash-burning monsters, kept alive by the faith that their hyperscaler investors have in them. This is faith not only in the evident brilliance of their inventions, but also in the promise that their genius can be monetised at a pace that eclipses payments due to service debt, lease commitments, forward power agreements, the works." In my new report on "The AI Trade" (https://lnkd.in/ggsktfYN), I dive deep on the open-weight threat and what it could mean for investor expectations about the sustainability of Anthropic's and OpenAI's business models. To quote the report: "Since ChatGPT’s debut, AI has been a speculative bet on an unprecedented collective scale. There’s never been a clear path to profitability. There’s never been guarantees about the technology’s rate of improvement and timeline to game-changing enterprise transformation. From tech giants to financial institutions to retail investors, AI enthusiasm has rested on faith-driven patience—belief that the promise so many see in the technology’s current capabilities will at some point lead to future capabilities that generate skyrocketing profits. The question today: how much longer will that patience last?... It’s hard to overstate how much hinges on Anthropic's and OpenAI's IPOs. They will account for 13% of AWS revenue and 27% of Google Cloud revenue this year. As for Microsoft, OpenAI alone accounts for roughly 70% of its AI-specific revenue. OpenAI has committed to tens of billions of spending on chips. Deepening circularity concerns, tech giant earnings growth has been increasingly driven by paper gains in the private-market valuations of Anthropic and OpenAI. In Q2, Amazon, Alphabet, Nvidia, Meta, and Microsoft reported $160b in cumulative “other income”, trouncing the $69b in “other income” reported in Q1. To quote the FT: “These one-off valuation boosts, derived in large part from enthusiasm around AI, risk distorting the financial picture at a time when investors are closely scrutinizing tech earnings.” If either Anthropic or OpenAI stumble, it’ll hit tech giants on multiple balance-sheet fronts. A pin-prick popping of the AI bubble may not be our base case, but if a pin is out there, it’s likely the revenue versus spending trajectories of Anthropic and OpenAI." FT link: https://lnkd.in/grhQySdM