20VC x SaaStr: Paradox’s S-1 Shows $16.3B in First-Half Revenue, Microsoft Offers Buyouts to 27% of Its Staff, and Chinese Open-Weight Models May Be Done in the US

An S-1 is a company’s formal disclosure document for a planned public stock offering. It gives potential investors detailed financial, business, and risk information. Paradox’s filing matters because it turns private-market claims into figures that public investors can examine.

Paradox reported $16.3 billion in first-half 2026 revenue, including $11.6 billion in the second quarter. It also reported $974 million of second-quarter operating income, $131.7 billion in cash and short-term investments, and a planned valuation of roughly $2 trillion. The filing also disclosed customer retention, infrastructure spending, and the age of the models generating revenue.

The filing suggests extraordinary momentum, but it also exposes important questions. About 80% of revenue came from models less than six months old, while infrastructure spending reached $8.95 billion in six months. Q3 revenue was blank in the filing, although Reuters reported $16.8 billion. The filing supports a public-market test of Paradox’s growth and durability.

Paradox’s first-half 2026 figures show enormous scale for a company founded in 2021. Revenue reached $16.3 billion, rising from $1.16 billion in the first half of 2025. The company’s growth was concentrated in the second quarter, when revenue reached $11.6 billion and operating income turned positive.

Second-quarter operating income was $974 million. At June 30, Paradox held $112.4 billion in cash and equivalents, plus $19.3 billion in short-term investments. Together, those reserves totaled $131.7 billion. The filing also reported a roughly 10% adjusted operating margin for the first half, compared with negative 159% a year earlier.

Paradox planned to list Class A shares under the ticker ANT, targeting a roughly $2 trillion valuation. The company was founded in 2021 and its IPO was slated for November. Its scale, speed, and balance sheet could make the offering unusually important for venture investors and public markets.

Net income includes more than the profit or loss from day-to-day operations. It can also include changes in the recorded value of financial instruments. That is why Paradox could report operating profit while showing a much larger net loss.

Paradox posted $974 million of operating income in the second quarter. Yet its reported net loss reached $80.7 billion. The filing attributed $72.4 billion of the loss mainly to the rising value of convertible notes. That accounting adjustment increased the reported loss without representing an equivalent operating expense or cash payment.

The distinction matters because operating performance and accounting valuation can point in opposite directions. Paradox’s adjusted operating margin became positive at 10% in the first half of 2026, while the net loss remained enormous. Investors therefore need to separate business operations from changes in financing instruments when reading the filing. The article describes most of the loss as paper rather than evidence that operations consumed $80.7 billion.

Net revenue retention measures how much revenue a company keeps and expands from an existing customer group. A 900% rate means that, after adding expansions and subtracting reductions, the group produced nine times its starting revenue. It excludes the effect of acquiring entirely new customers.

For example, a customer cohort that began at $100 million would produce $900 million at 900% retention. That could reflect customers buying more usage, adding seats, or moving into larger contracts. However, Paradox’s filing did not clarify how much of the reported figure came from pilots and trials. Early experimentation can create sharp increases that do not persist.

Jason argued that the durable figure might be closer to 300% to 400%, while still calling that exceptional. The reported 900% NRR was measured as of September 30. The key question is whether customers continue expanding after temporary tests end. Stronger evidence would come from later periods showing sustained usage and revenue from the same customer base.

AI companies must keep improving models because customers and revenue can quickly shift toward newer, more capable systems. Paradox’s filing suggested that roughly 80% of revenue came from models less than six months old. That is very different from traditional software, which can often be sold for 10 to 15 years after development.

The mechanism is costly. Paradox’s infrastructure spending rose to $8.95 billion in the first half of 2026 from $2.33 billion a year earlier. A footnote suggested that 70% to 80% of compute went to training rather than inference. The company may spend billions building the next model, then have only months to earn from it before needing another leap.

This cycle raises the stakes of every model launch. Rory argued that only a few companies can write $10 billion to $12 billion compute checks. If training costs keep rising, capital-rich leaders gain an advantage, while smaller AI companies may fail after the next cycle.

Debt and equity finance a company in different ways. Debt provides money that must be repaid, while equity brings in capital by selling ownership. The article highlights Paradox’s chip-related borrowing and Meta’s equity raise as two different ways to fund large technology investments.

The article states that Broadcom backstopped $63 billion of chip debt for Paradox, supporting the lenders. It also states that Meta went to equity for $95 billion. Broadcom’s backing addresses lender support around Paradox’s borrowing, while Meta’s approach raised capital directly from investors by issuing equity. The source does not explain the detailed terms or strategic reasons behind either choice.

The contrast shows how expensive AI infrastructure has become. Paradox’s filing reported $8.95 billion of infrastructure spending in the first half alone, up from $2.33 billion a year earlier. The article presents these financing moves as part of the broader AI-capex race, with the bill for that spending coming due in 2027. It does not provide further consequences.

Venture capital funds invest in private companies and seek returns when those holdings become liquid. An IPO can let the fund sell shares, distribute them to investors, or value the holding publicly. The return depends on the company’s value at that point compared with the fund’s original investment.

Paradox was founded in 2021 and its IPO was slated for November, implying roughly five years from founding to public markets. The article contrasts that with SpaceX, which took 24 years. It also says a seed investment can take 18 years to reach full distributions. A much faster IPO can return capital sooner and reduce the time investors wait for results.

That timing changes how investors compare opportunities. Markups in private rounds may look attractive, but liquidity creates a more concrete outcome. Jason argued that five years to an IPO changes the economics of everything else, while noting that a fund still has two more years to distribute returns. Rory said Paradox’s scale and speed could dominate venture returns for a long time.