The Drawing Room

This wing is new. It was built because visitors kept asking what happens next, and the answer kept being drawn in the air with a hand, which is not a standard the rest of the building holds itself to.

So: a frame. It is chalk on slate, it is propped on scaffolding, and you will notice it does not touch the floor. None of that is styling. A pole is bolted down because somebody measured something. This is not bolted down because nobody has.

The hinge is brass, and it is the only brass on the object. Below it, the last figure anyone actually agreed. Above it, arithmetic. The building will vouch for the hinge.

Every frame in here carries a card with four lines: what is assumed, who assumed it, when, and what would prove it wrong. A projection that cannot fill in the fourth line does not get a frame. It gets a conversation, in an office, elsewhere.

Anthropic, from the nursery. $183 billion in September 2025, the second clamp you saw there. $965 billion in May 2026, the latest price anyone has agreed. Both agreed by people who signed something, and both on the pole, where agreed prices go.

The frame beside it takes those two points, finds the second is 5.27 times the first over 268 days, and assumes that happens once more: $5.09 trillion, projected for February 20, 2027. That is the entire method. It is on the card, because a projection whose arithmetic is hidden is just a confident voice.

There was a round between those two, in February 2026, at $380 billion. A line through that one and May would be steeper and would get there sooner. Which two points is a choice, and the card says which were chosen.

OpenAI next door: $300 billion in March 2025 and $852 billion in March 2026, both rounds in which the company sold new shares and received the money. 2.84 times in a year, assumed to happen again: $2.42 trillion, projected for March 31, 2027.

Between them sits October 2025’s $500 billion, when employees sold their own shares and the company received none of it. The frame would draw a line through that just as happily. It cannot ask what kind of price it was given. That question belongs to whoever holds the chalk, and the card names them.

The same rate as next door, not applied once but left running: 5.27 times every 268 days, on the assumption that it simply carries on.

Projected, if it holds: $5.09 trillion by February 20, 2027; $26.8 trillion by November 15, 2027; $141 trillion by August 9, 2028, at which point the chalk leaves the room through the clerestory. By January 27, 2030 it is $3.93 quadrillion, projected, which this building cannot draw at any scale it owns.

The conservatory’s response has been practical. At one centimetre to the billion, the last stop needs 39.3 kilometres of frame. A works order has been raised. It specifies a hinged roof, 9,837 additional trestles at four-metre centres, and consultation with the neighbours. It is with planning, and the caretaker has been advised that the timescale is eighteen months.

Nothing in that arithmetic knows about customers, electricity, chips, or how many dollars exist. It was told a rate and it continued the rate. That is the entire competence of the instrument, and the reason it is chalk: a pole that did this would be broken, and a frame that does it is only being obedient.

The card’s fourth line reads: falsified by the first agreed price that comes in below the line.

Wiping a frame is a maintenance task, and the conservatory files the card rather than the chalk. The cards are what is worth keeping. A projection that came true and one that did not look identical in hindsight unless somebody wrote down what was assumed at the time.

The lit one is ours. When this route was planned, the room about the launch company was written on an assumption: that it was private, filed nothing, and could only be priced by a tender offer it organised itself. That assumption was the whole point of the room.

It listed on Nasdaq on the twelfth of June, 2026. The building’s correction is dated the eighteenth of September: three months in which a room was being built on something that had stopped being true.

Beside it hangs the first card drafted for this wing. It was drafted in September for a date in March, after March had happened, and the February round had already come in below it. It never went up. It is filed anyway: a card that was wrong before it was hung is exactly the kind worth keeping.

The launch room was rebuilt and it is better now; it is about the reading changing kind, which is more interesting than what it was about before. That is luck, and the card stays up anyway. The rest of these hooks are empty. They are not decorative.