Investment analysis by Ruslan Averin — originally published at averin.com.
In one week, two Asian technology companies went to the market for roughly $16.5 billion, both explicitly to fund artificial intelligence. Both stocks fell on the news.
Alibaba placed 710 million new shares at HK$112.70 for HK$80 billion — about $10.2 billion, and the largest primary follow-on ever by a Hong Kong-listed company. SoftBank filed to issue ¥1 trillion of seven-year retail bonds, roughly $6.3 billion, at an indicative coupon of 4.3% to 4.9%.
Two ways to pay for the same thing
| Alibaba | SoftBank | |
|---|---|---|
| Instrument | New equity | Seven-year retail bonds |
| Size | HK$80bn (~$10.2bn) | ¥1tn (~$6.3bn) |
| Price / coupon | HK$112.70 a share | 4.3%-4.9% indicative |
| Use of proceeds | 100% into full-stack AI | AI investment and refinancing |
| Share reaction, 24 Aug | −8.5% to HK$112.50 | −5.3% to ¥4,975 |
One diluted its owners. One levered them. The market marked both down.
The Alibaba detail worth stopping on
The placement priced at HK$112.70. The stock closed Monday at HK$112.50 — beneath the price the new investors paid.
That does not happen in a well-absorbed deal. It means the buyers of $10.2 billion of new stock were, within a session, holding a loss, and it tells you the size cleared through price rather than through demand.
Alibaba was explicit that 100% of the proceeds go into AI infrastructure. So the market was asked to fund a capex programme with no revenue attached yet, and it agreed — at a discount, and grudgingly. The stock recovered to HK$115.50 by Thursday, which is a partial answer, not a verdict.
The SoftBank detail worth stopping on
A 4.3% to 4.9% coupon on seven-year paper, sold to Japanese retail investors, for the largest retail bond in the country's history.
Japanese households are being offered a yield that did not exist domestically for a generation, in exchange for financing an AI portfolio whose returns arrive well beyond the seven-year tenor. The bond matures long before most of what it funds is expected to pay.
That mismatch is the trade. SoftBank has run it before, successfully and unsuccessfully. What is different now is the price of the money: the previous record issue in April 2025 was ¥600 billion, and rates were a different animal.
Why the timing is the story
These two raises priced in the same week that Fed chairman Kevin Warsh moved September rate-hike odds to a coin flip, and the same week Nvidia guided to $108 billion of quarterly revenue.
Put those together and the picture is uncomfortable. Nvidia's revenue is somebody's capital budget. The capital budget is increasingly funded with issued paper — equity where the stock will absorb it, debt where it will not. And the cost of that paper is being repriced upward while the projects it funds are dated years out.
AI capex has quietly become a financing story rather than a demand story. Demand is not in question; Nvidia settled that on Wednesday. What is in question is who funds it and at what cost.
How I read it
Watching who has to issue and on what terms is a better read on this cycle than watching order books.
A hyperscaler funding data centres from operating cash flow is in a fundamentally different position from a company placing stock below the last traded price or selling seven-year retail paper at 4.9%. The first can slow down voluntarily. The second has a maturity schedule.
What I would watch: SoftBank's final coupon on 4 September. The indicative range was set before Warsh spoke. If it prices at the top of the range or above it, that is the first hard number on what a September repricing costs the AI build-out — and it will be a real number, not a survey.
More market analysis by Ruslan Averin at averin.com.
Top comments (0)