The mechanics of capital are shifting. Lightspeed Venture Partners is raising $600 million for a secondary fund, codenamed Project Mercury, to acquire shares in OpenAI and Anthropic. The move is not a venture round. It is a liquidity event disguised as an investment thesis.
This is not about funding AI research. It is about capturing equity in companies that are structurally closed to new capital. OpenAI and Anthropic have not gone public. Their valuation churn is driven by private rounds and secondary trades. Lightspeed, a latecomer to the AI frontier, is using a financial instrument to buy its way into the table.
I have seen this pattern before. In 2017, I audited ICO smart contracts for a Shanghai fintech firm. The compliance protocols I wrote were designed to catch arbitrage structures hiding behind token allocation. Back then, capital was chasing whitepapers. Today, capital is chasing illiquid equity. The structural problem is the same: the price of access is decoupled from the price of fundamentals.
Context: The Geometry of AI Illiquidity
The AI market in mid-2024 is a two-tier system. Tier one: OpenAI and Anthropic, with valuations above $100 billion and $20 billion respectively. Tier two: everyone else, fighting for scraps. The number of shares available for purchase in tier one is finite. Early investors—Sequoia, Thrive Capital, Khosla, Spark Capital—hold the bulk. Employees hold options. No one wants to sell at a discount, and the IPO window is sealed shut for high-loss, high-CAPEX companies.
Enter the secondary fund. A GP-led continuation vehicle allows existing LPs to cash out partially, new LPs to enter at a premium, and the GP to keep collecting management fees. Lightspeed is doing exactly this. It is raising $600 million from a combination of its Select V Fund, Opportunity II Fund, and a separate managed account (SMA) for a single large institutional client. The SMA is the tell. A sovereign wealth fund or university endowment is demanding bespoke AI exposure, unwilling to pool with retail-level capital.
Core: The Double Bet on Two Technical Routes
Lightspeed is not betting on a single AI company. It is betting on the entire frontier. OpenAI represents the general AGI route—capability-first, scaling-aware, API-ecosystem-dominant. Anthropic represents the safety-aligned route—constitutional AI, long-context, enterprise-grade compliance. By buying both, Lightspeed is hedging the philosophical divergence of AGI development.
This is a classic portfolio construction move. The question is whether the underlying assets justify the premium. OpenAI’s annualized revenue by mid-2024 was estimated at $30–35 billion, driven by ChatGPT subscriptions, API tokens, and enterprise contracts. Anthropic’s revenue was in the low single-digit billions but growing at 50%+ quarter-over-quarter. The secondary market is likely pricing these companies at a 20–50% premium over the last primary round. That implies a 3–5 year revenue CAGR of 30–60%.
From my 2020 DeFi liquidity stress test, I learned that growth projections are often built on fragile assumptions. Back then, I modeled how M2 expansion correlated with on-chain volume spikes. The same principle applies here: AI revenue growth depends on the durability of the model capability curve. If the rate of improvement slows—if open-source models like Llama 3, Qwen, or DeepSeek catch up—the premium evaporates.
Contrarian: The Decoupling Myth
Conventional wisdom says Lightspeed is buying the only two companies that matter. I see a different risk. The secondary fund provides no operating capital to OpenAI or Anthropic. The $600 million goes to selling shareholders—early employees, seed investors, or strategic partners like Microsoft and Amazon. This is a liquidity exit for them, not a growth injection for the companies.
Furthermore, the investment is built on an assumption that the technological moat is wide enough to withstand open-source erosion. I am skeptical. Open-source models have been closing the gap at every release. The cost of training frontier models is dropping. The API price war is compressing margins. If the advantage of OpenAI and Anthropic becomes a matter of brand and distribution rather than raw capability, the premium paid today will look like a mistake.
There is also the regulatory tail risk. The EU AI Act is entering enforcement. The U.S. is still debating model registration requirements. Copyright lawsuits against OpenAI (New York Times, Getty Images) could force changes to training data pipelines. Lightspeed’s dual bet on both companies does hedge the “safety regulation” risk—if one company is penalized, the other may benefit—but it does not hedge systemic regulatory closure.
Takeaway: The VC-to-Asset-Manager Transition
This fund marks a structural shift. Venture capital is no longer just about funding early-stage innovation. It is becoming a permanent asset class for institutional capital. Lightspeed is transforming from a risk-taking investor into a fee-collecting inventory manager of AI equity. The exit strategy is written in ice, not in hope. They are not waiting for an IPO. They are building the secondary market themselves.
The deeper question is whether this model is sustainable. If every top-tier VC firm launches a secondary fund, the market becomes a closed loop of capital recycling among the same set of assets. Innovation gets funded, but the upside is captured by secondary buyers who added no operational value. The AI industry becomes a financialized asset class rather than a technology revolution.
I have seen this before. In 2022, when Terra collapsed, everyone who had a liquidity exit protocol survived. The ones who used hope as a strategy lost everything. Lightspeed is building an exit protocol for the AI era. But the protocol only works if the underlying assets keep growing. If the AI bubble bursts, the secondary fund will be the first to lock its doors.
Exit strategies are written in ice, not in hope. Lightspeed’s ice is cold, but it is still a bet on the same fragile curve.