Lightspeed's Project Mercury puts the firm on both sides of the OpenAI price

Lightspeed is moving its OpenAI stake into a continuation fund that Lightspeed will also manage, in a deal the firm calls Project Mercury, Bloomberg reported, citing people familiar with it ^[1]. That matters because the position has no share price to consult, so a manager selling on behalf of one set of investors also gets to help set the number its own new vehicle pays ^[7].

The assets come out of two Lightspeed funds, Select V and Opportunity II, plus a separately managed account, with OpenAI as the anchor ^[2]. Coller Capital is the lead buyer and UBS is advising ^[3]. Lightspeed did not reply to Bloomberg's messages and Coller declined to comment ^[11].

The structure is not exotic. A venture fund has a defined life, usually about ten years, at the end of which it must sell what it holds and return cash ^[4]. A continuation fund breaks that clock: chosen holdings move into a new vehicle, an outside buyer funds it, the original investors either take the cash or roll into the new fund, and the firm keeps the position ^[5]. The problem it solves is real, because companies now stay private far longer than a decade ^[6].

Pricing is where it gets awkward. The available reference point is the last primary round, which valued OpenAI at $852bn post-money in March after a $122bn raise, according to Tech Funding News ^[8]. That figure was set by buyers putting new money in, not by sellers exiting ^[9]. Coller is the only discipline in the room: an arm's-length buyer with its own investors and no reason to overpay ^[10].

One detail separates this from a plain extension. Bloomberg describes the Anthropic exposure as a new commitment rather than an existing holding ^[12], which means investors asked to roll are also funding a fresh position ^[13]. Anthropic's last round valued it at $965bn, three months after a round at $380bn ^[14], roughly 2.5 times higher in a quarter ^[15]. Lightspeed has backed the company across three consecutive rounds including a $3.5bn Series E in 2025 ^[16].

The rest of the box spreads risk away from the frontier labs, though not far. Verkada was lifted to $5.8bn by a CapitalG investment in December ^[17]. Rippling raised at $16.8bn in May 2025 ^[18] and Glean at $7.2bn a month later ^[19]. Reflection AI raised at $8bn in October 2025 and in March reportedly sought fresh money at $25bn ^[20], about three times the earlier mark ^[21], with a $6.3bn compute deal with SpaceX for Nvidia chips running to 2029 behind it ^[22].

The arithmetic explains the spread of the structure better than any thesis. Venture-led secondary deals reached $35bn in 2025, roughly double the 2023 figure, according to PJT Partners ^[23], implying about $17.5bn two years earlier ^[24]. The wider secondaries market ran to $121bn of transactions in the first half of this year alone ^[25]. Megadeals took 87.5% of the dollars deployed in US venture in the second quarter ^[26], and a vehicle designed to hold winners longer concentrates that further rather than loosening it. Accel raised $3.5bn across four funds in a single process the same week ^[27]. Thrive Capital's first investor letter showed how completely OpenAI has come to define that firm ^[28]. Coller itself is a London secondaries specialist recently acquired by EQT, the Swedish investment group ^[29].

Watch the transfer price against the $852bn primary mark, and whether it is disclosed at all ^[8]. Watch how many original investors take cash rather than roll, since that choice is the product ^[5]. And watch whether other holders of late-stage AI paper copy the mechanism.