Start with the bridge, because that is the part a prospectus has to survive. Subtract the $41.55bn conversion charge [8] from the $39bn net loss attributable to OpenAI [6] and the result is a $2.55bn profit [6]. That figure never appears anywhere; the number reported after the charge and other non-cash items is a loss of about $8bn [9]. Getting from the first to the second takes roughly $10.6bn of items pushing the other way [7]. The word doing quiet work is attributable: a loss attributable to OpenAI [6] can sit inside a larger group loss, and a charge of that size need not sit wholly within it. The material does not say which, and the person familiar with the matter says only that the large majority of the increase was non-cash and unrelated to underlying operations [7], which is an assertion about causation rather than a reconciliation.
Calling the year a $34bn burn describes the headline; it skips the mechanism. Against about $13bn of revenue [4], spending of about $34bn [1] is $2.62 of outlay per dollar booked [2] and a $21bn shortfall [1], and part of that outlay never left as cash, since staff stock and Microsoft computing credits are among the non-cash items being stripped out [9]. That relocates the question rather than shrinking it: a supplier taking credits instead of invoices is a funder, and the runway then turns on that appetite as much as on the next equity round.
Research and development runs at $1.46 for every revenue dollar [3], sales and marketing at another 46 cents [4], and the residual, about $9bn, is everything else [5]. The sales line is where I would stare longest, because monthly revenue climbed from roughly $333m to $2bn inside a year [9], hardly a product short of pull, yet paying 46 cents of every dollar to push it anyway is a decision to buy distribution rather than bank it.
Provenance matters more for the weaker numbers. The audited figures reached the FT from independent journalist Ed Zitron [10]. The secondhand set, relayed from the Wall Street Journal via Gary Marcus, has OpenAI adding $1bn of revenue and $3bn of losses in the quarter before its hoped-for listing, a loss of about $1.84 per dollar of revenue against the $1.22 Zitron had been citing [12]. Apply $1.84 to a quarter at $2bn a month and the implied loss is about $11bn [12], not $3bn, so the ratio and the increment are not measuring the same base, and neither is audited.
This is probably wrong, but on these figures the near-term issue reads as funding rather than solvency: an exit run rate of $24bn [8] sits $11bn above booked revenue, roughly half the gap [14], and an $8bn ex-charge loss closes inside a year if research spending stops compounding. The other two paths are that the listing gets its price, which at the trillion-dollar ask [11] is 77 times last year's revenue [10] and 42 times the exit run rate [11], or that spending keeps pace with revenue and outside capital becomes a permanent annual line. Only a vendor refusing paper turns any of this into insolvency. What would prove this desk wrong is checkable and dull: a 2026 audit showing revenue compounding off that run rate while research sits still would make this an ordinary scaling story.
