a16z Jen Kha: Writing Bad Checks is a Professional Cost, Missing SpaceX, OpenAI, or Stripe is the Fatal Mistake

Jen Kha from Andreessen Horowitz states in a lengthy article that the mindset of venture capital can be summarized by an Eisenhower matrix: writing bad checks is a professional cost, while missing out on SpaceX, OpenAI, or Stripe is a fatal mistake; currently, limited partners may be making another active mistake by justifying their actions based on whether the portfolio is asset-balanced.

She provides internal insights: for every dollar invested by the managed fund, there is a loss of about 15 cents; a single Databricks position accounts for about 20% of managed assets. Weekly, there are still limited partners with nearly zero exposure to leading models like SpaceX, Anthropic, and OpenAI, and no positions in leading open-source models. Historically, donation funds could allocate 25% to 40% of their total portfolio to venture capital, while ordinary institutions often limit venture capital to 5% to 10% of alternative assets. After its IPO, SpaceX's market value is about $2 trillion, with the combined equity value of Anthropic and OpenAI estimated at approximately $3.8 trillion to $5 trillion, primarily growing in private markets. Google’s market value on its IPO day in 2004 was about $27 billion, which, adjusted for inflation, is still only a fraction of the median post-money valuation in large venture financing since 2025.

CalPERS reduced its new commitments in controlling acquisitions from about 91% to 58% after 2022, while growth stocks and venture capital combined rose from 9% to 43%. The ranking of private equity projects in large pension funds has risen from 30th to 1st in three years. MassPRIM and the North Carolina Treasury are also following suit. According to PitchBook, among 2,143 global venture funds with DPI disclosures from 2000 to 2018, only 17% returned at least double the principal, 6.7% achieved three times, and 2.4% reached five times. Accolade statistics indicate that among approximately 3,000 venture capital firms in the U.S., about 20 have consistently achieved three times net returns over twenty years. The DPI of decile funds can reach 2 to 3 times that of quartile funds and 4 to 10 times the median.

a16z once offered liquidity solutions for Stripe and Databricks positions to old fund LPs, but no one chose to sell.

The secondary market shows almost no discounts for truly desired assets and discounts for unwanted assets. Vista's Robert Smith once said that software contracts are superior to first-lien debt; Kha notes that a new wave of disruption may make recurring revenue less durable, with acquisition fund returns nearing a low point over the past fifteen years. Private credit is shifting towards AI infrastructure, but early loans made to software acquisition targets are facing multiple compressions and refinancing challenges.

In market mechanisms, this represents a dual auction for allocation weights and fund access. Buyers must explain zero exposure to pension and donation funds to investment committees; sellers are the few managers who can repeatedly enter mega rounds. Beneficiaries are old funds that already hold platforms and whose LPs are unwilling to cash out; those under pressure are balanced models capped at "cottage industry" proportions. Funds are stuck in compounding positions that they are unwilling to sell, while new commitments are squeezed towards a narrower list of managers.

Supplementary structure: The article compares the largest public listings and acquisition exits of private equity with SpaceX and Cursor to illustrate exit value rather than the number of exits rewriting the alternative asset ledger.

Source: Public Information