I wonder how the "It's all heavily subsidized!" people are going to spin this.

In page F-7 they write:

Audited, revenue in 2025 is $4.6 with a computing cost of $7.3 (i.e. heavily subsidizing tokens). Unaudited, revenue in 2026 (first six months) is $16.3, computing cost $8.9 (1.83:1). But after adding personnel, sales and administrative costs, they are still spending $17.0 to get that $16.3 revenue (still subsidizing tokens, but less).

my thesis is that it will go down and not up over time. The moat they had in coding which is by far the highest token consumption is not that huge compared to Nov-Feb period where token maxxing was prevalent and no other vendor's model was that good.

Assuming computing costs do not go up and revenue per customer does not go down.

I think the 500 billion $ of compute commitments are at higher prices than what they've been paying so far. Would be a shame if a price war between the AI labs intensifies at the same time.

Hmmm

Less than a year would also match what DeepSeek was reporting, where compute pays off in just 10 months. I think they also said they had an 80% margin on inference.

Does that include R&D?

Over time that margin persists only if they maintain their competitive edge, which is not free.

Impressive, but probably based on depreciation servers over 5 years. But is that realistic? I wonder how many people are still using 5 year old servers in their AI training or inference.

When will generic AI compute capability trickle down to us plebeians?

Seems like that would imply you should be getting people to adopt now now still even if it costs a stupid amount, if you think it's valuable, rather than limiting token budgets. It'll only get cheaper over time.

ANT stock valuations might be high at listing, but it will be a good stock to add on declines. Great numbers!

Just renting them out as hardware is close to 12 months.

Including training?

That's their API cost price. No one will pay this on that scale? People are using tokens via their Max Plans?

Why are you ignoring the $80.7 Billion loss?

$72 billion of this $80.7 billion loss consists of non-cash accounting charges— related to convertible debt—triggered by a surge in the company’s valuation.

The hardware related liabilities are staggering.

AI psychosis is real.