Ottawa's photonics loan prices each promised Xanadu job at C$709,000

A repayable contribution is debt wearing a policy name ^[2], and the term worth reading twice here is the counterparty: Xanadu's own filing lists net losses, a limited operating history, and substantial doubt about its ability to continue as a going concern ^[11]. Ottawa is lending C$195 million over five years, against a required co-investment the announcement calls significant without sizing ^[3], to a Nasdaq- and TSX-listed company founded in 2016 ^[12] whose first quantum data center is not targeted until 2029 ^[21]. Service on that loan, if it is serviced at all before 2029, comes out of capital raised elsewhere.

The coverage arithmetic is the part that should govern expectations. C$195 million against a C$893 million project is 21.8 percent ^[13], which leaves C$698 million outside the federal cheque ^[18]; if Ontario writes the balance of the up to C$390 million Xanadu has asked the two governments for ^[4], the public share reaches 43.7 percent and C$503 million still has to come from somewhere else ^[14]. Against up to 275 highly skilled jobs ^[5], the federal money works out to roughly C$709,000 per job ^[15] and the full project to about C$3.25 million per job ^[16]. It is also 2.5 times the C$78 million of government support Xanadu has received to date ^[17], which is why Xanadu's press release can call it the largest government investment in quantum manufacturing in Canadian history ^[22].

Now the sovereignty question. Most chip packaging and assembly still goes to contractors in other countries ^[6], and Inception is explicitly built to take that in house, with cleanrooms, round-the-clock test facilities and what Xanadu describes as first-of-its-kind heterogeneous integration ^[8]. Weedbrook's figure is that an outside partner takes as long as three months to house and connect a photonic chip and that the plant will do it in eight hours ^[19]; at 90 days, that is 2,160 hours compressed to eight, a 270-fold improvement ^[23]. This is probably wrong, but I read 270-fold as a design target rather than a measured one, because nothing has run yet. Meanwhile the tooling comes from ASMPT, Bluefors, DISCO, EVG, FiconTEC and MPI ^[9], and the announcement does not say where any of those machines are built. The loan domesticates a process step, while the equipment that performs it still comes from other people.

The counter-thesis is decent: this is unusually patient capital for capex that a loss-making issuer could not raise cheaply on public markets, and if heterogeneous integration works at volume, owning the packaging line is the durable asset rather than the qubit count. The version I find more interesting is the allocation cost. Every dollar of that co-investment is a dollar not going to software, physics or the contractors Xanadu currently pays, and taking a 320-person team toward as many as 595 people ^[20][24] means hiring manufacturing capability three years before the data center that is supposed to consume the output.

What would prove the thesis wrong: Ontario funding the rest on comparable terms, and Inception actually in use early next year ^[20] with the offshore packaging contracts wound down rather than merely supplemented.