On Sunday evening, 23 August 2026, Alibaba priced the largest follow-on share sale a Hong Kong-listed company has ever done: 710 million newly issued ordinary shares at HK$112.70 each, HK$80 billion in total, roughly US$10.21 billion. One hundred per cent of the net proceeds go into what the company calls its full-stack artificial intelligence capability — its own silicon, its own data centres, its own models. On Monday the Hong Kong line opened badly, touched HK$110.10 intraday and closed at HK$111.00, down 9.76 per cent on turnover of HK$27.76 billion.
That is the headline, and it is entirely accurate. It is simply not the interesting number. The interesting number appears only when you set the issue price of this placement next to the price at which the very same company took the very same shares out of the market over the preceding two fiscal years. Because Alibaba was not just any issuer. Until recently, Alibaba ran one of the largest share repurchase programmes on earth.
What was sold — and who was not allowed to buy
The mechanics fit in three sentences. Alibaba used its general mandate, the standing shareholder authorisation that lets a Hong Kong issuer place new stock without going back for a separate vote. The shares went to professional and institutional investors outside the United States, expressly under Regulation S of the U.S. Securities Act. And they went at an 8.4 per cent discount to Friday’s close, which therefore sat around HK$123.
The second point deserves more attention than it has been getting. Regulation S means no U.S. persons. Yet Alibaba’s single largest holder base has sat in New York since the 2014 listing, holding American Depositary Shares that each represent eight ordinary shares. Those investors were not permitted to subscribe for the discounted stock. They carry the roughly 3.7 per cent post-issue dilution in full. An institution in Hong Kong got a discount; a fund in Boston got the dilution. That is not an irregularity — it is the plain consequence of a structure optimised for speed — but it explains part of the irritation.
And the discount did not hold. Anyone who subscribed on Sunday at HK$112.70 was already 1.5 per cent under water by Monday’s close at HK$111.00. On a deal this size that is not an accident of the tape. It is a finding: the market did not read the discount as an entry point, it read it as the price of news it did not want.
The reversal: from $24.4 billion of buybacks to 710 million new shares
Here is where the real story starts. In the fiscal year ended 31 March 2024, Alibaba repurchased 1,249 million ordinary shares — the equivalent of 156 million American Depositary Shares — for a total of US$12.5 billion. Net of stock issued under employee plans, the share count fell by 1,057 million, or 5.1 per cent. In the year to 31 March 2025 it did almost exactly the same thing again: 1,197 million ordinary shares for US$11.9 billion, a net reduction of 995 million shares, again 5.1 per cent.
Two years, US$24.4 billion, roughly 2.45 billion shares retired. That was not an investor-relations gesture. It was the central capital allocation decision of a company whose growth story had stalled and whose stock was on the floor.
Then it turned. In the fiscal year to 31 March 2026, repurchases came to US$1.046 billion — down more than 90 per cent. In the June 2026 quarter Alibaba bought back 13.4 million ordinary shares, about 1.7 million American Depositary Shares, for US$162 million, against US$815 million in the same quarter a year earlier. Two months later, the same company issued 710 million new shares in a single night.
The ratio is what matters: one night created 53 times more shares than the entire preceding quarter retired. And the 710 million equal 29 per cent of everything taken out of the market across fiscal 2024 and 2025 combined. Nearly three in every ten repurchased shares are back.
One detail sharpens the reversal further. The buyback authorisation is still live. As of 30 September 2025, US$19.1 billion of it remained unused, and the programme runs through March 2027. Alibaba therefore still holds formal permission to buy its own shares — and has just sold US$10.2 billion of them.
The arithmetic almost nobody is doing
Put the two prices side by side. Across fiscal 2024 and 2025, Alibaba paid US$24.4 billion for 2,446 million ordinary shares. That works out at roughly US$9.98 per ordinary share, or about US$79.80 per American Depositary Share.
The placement price is HK$112.70. At the exchange rate implied by the transaction itself — HK$80 billion equalling US$10.21 billion — that is about US$14.38 per ordinary share, or roughly US$115 per American Depositary Share.
Alibaba retired stock at an average of just under US$80 per depositary share and is reissuing it at about US$115. That is a premium of some 44 per cent over its own buyback average. Applied to the 710 million ordinary shares now being sold, the gap between the repurchase price and the issue price is worth roughly US$3.1 billion. That is what the sequence — buy cheap, sell dear — delivered to continuing holders relative to a world in which the buybacks had never happened.
This is the textbook definition of countercyclical capital allocation, and it is rare. The corporate norm is the exact inverse: repurchase at the highs, when cash is abundant and sentiment is warm; issue at the lows, when there is no choice. In this particular sequence Alibaba did it the right way round. You do not have to like the strategy to concede the point.
Two qualifications belong immediately alongside it, or the observation becomes a slogan. First, the gain is arithmetic and retrospective; it appears in no income statement, because transactions in a company’s own shares run through equity. Second, and more seriously, the calculation says nothing about whether the US$10.21 billion will earn more than its cost of capital at its new destination. The pricing advantage is a one-off. The investment it funds has to justify itself every year.
What the money buys: almost exactly one quarter
This is where it gets uncomfortable. Alibaba spent RMB 67.68 billion on capital expenditure in the June quarter, up 75 per cent year on year. The US$10.21 billion raised converts to a little over RMB 68 billion.
The largest follow-on offering in the history of the Hong Kong exchange therefore funds roughly one single quarter of capital spending — at the current run rate, not the planned one. Anyone hoping this placement settles the funding question has mis-sized the problem. It buys about fourteen weeks.
The frame around it: RMB 380 billion, around US$56 billion, for artificial intelligence and cloud between 2026 and 2029. Roughly half of that had already been spent by the time the June quarter was reported. Chief executive Eddie Wu has repeatedly signalled the total is more likely to be overshot than met. The network now spans 104 availability zones across 30 regions, most recently adding a third data centre in South Korea.
Part of the money goes into silicon Alibaba designs itself. Its in-house accelerator, the Zhenwu M890, is said to deliver three times the performance of the preceding 810E — which is widely placed around the level of Nvidia’s H20 — and is fabricated domestically. Together with the M530 it sits on China’s approved-processor list for state procurement. That matters economically: a company that cannot simply order compute, because access to the best Western chips is restricted, has to pay for the whole value chain itself — design, fabrication, software, buildings, power. It explains part of the capital intensity. It also explains why the capital requirement is so hard to forecast.
Why the stock fell anyway
On Monday the market was not pricing the arithmetic. It was pricing the signal — and the signal landed three days after a set of results that had exposed the core of the problem.
June-quarter group revenue rose 9 per cent to RMB 268.95 billion, essentially in line with the RMB 268.88 billion consensus. Cloud and artificial intelligence revenue grew 45 per cent to RMB 48.44 billion, the fastest in 22 quarters. That segment’s adjusted operating margin climbed from around 7 per cent to 11.6. Annual recurring revenue from the model business passed RMB 16 billion. Every one of those numbers supports the thesis.
And yet: net profit fell 75 per cent, and adjusted earnings came in at RMB 8.52 per share against RMB 10.53 expected — a 19 per cent miss. The cause is not weak demand; it is accounting with a lag. Depreciation on data centres hits the income statement immediately and in full, while the revenue from that same hardware arrives over years. Wu puts the break-even on the AI spend at two and a half to three years — explicitly at today’s average gross margins. That assumption is the most fragile part of the whole model, because gross margin is the first thing to give in a price war, and Chinese cloud is one.
In that context the market does not read an equity raise as a well-executed placement. It reads it as an admission that operating cash flow no longer covers the build. Being right on the arithmetic is little help in the short run.
Alphabet, Intel, Alibaba: the capital stack is walking downhill
This is not an isolated event, and that is the larger finding. Alibaba’s deal is the third-largest primary follow-on offering anywhere in the world this year. Only two were bigger, and both fund exactly the same thing.
In early June, Alphabet priced a package totalling US$84.75 billion across common stock, mandatory convertible preferred, a US$40 billion at-the-market programme and a US$10 billion private placement with Berkshire Hathaway — the largest equity capital markets transaction ever executed. On 10 August, Intel followed with US$20 billion at US$95 per share, upsized, with a greenshoe on top, against a backdrop that includes a 10 per cent U.S. government equity stake. And on 23 August, Alibaba with US$10.2 billion.
Roughly US$115 billion of fresh equity in twelve weeks, all for the same purpose. Equity is the most expensive form of funding a company can use: no maturity and no coupon, but a permanent claim on profits. Firms reach for it when the cheaper rungs are exhausted — or when they think their own stock is dear enough to be worth selling. The sequence of the past two years is now complete: free cash flow first, then bonds and off-balance-sheet vehicles, now shares. If you want to know where a capital cycle stands, read the liabilities side, not the guidance.
The comparison that shows what Alibaba’s holders did not get
For investors used to London, there is a precise counter-example. On 23 May 2024, National Grid announced a fully underwritten £7 billion rights issue — 1,085,448,980 new shares at 645 pence, seven for every twenty-four held — to help fund roughly £60 billion of network investment across the United Kingdom and United States between 2024 and 2029. The shares fell sharply on the day, and the discount was enormous: 645 pence against a theoretical ex-rights price of £9.88.
Same underlying story: an infrastructure operator raising equity because a capital programme has outgrown its cash flow. Completely different treatment of existing owners. A rights issue gives every holder a tradeable entitlement; those who cannot or will not take it up can sell the right and be compensated for the dilution. British institutional practice, codified by the Pre-Emption Group, keeps non-pre-emptive issuance to a limited slice of capital precisely for that reason. Hong Kong’s general mandate grants no such entitlement, and United States corporate law generally provides no statutory pre-emption right at all — which is why Alphabet could run a US$40 billion at-the-market programme without asking anybody.
For an American holder of Alibaba depositary shares, the practical takeaway is narrow but real: the dilution is unavoidable, the discount was unreachable, and there was no entitlement to sell. Anyone who holds non-domestic equities should know which of the three regimes their issuer sits in before, not after, the placement lands.
Risks, counterarguments, and the week ahead
The bear case is serious. First, the objection that a one-off pricing advantage is being netted against permanent dilution is fair: the US$3.1 billion happens once, the 710 million shares are forever. Second, Alibaba had already redirected capital by stopping the buyback before the placement existed; this is the second step of one movement, not the first. Third, the argument that a company issues stock because it considers it expensive sits uncomfortably close to saying management thinks the shares are overvalued — which is rarely supportive in the short term.
Fourth, and heaviest: the entire payback calculation rests on gross margins that need not hold in a market with state-backed competitors, aggressive pricing and an open-ended model race. If cloud margin in two years sits where it sat a year ago, break-even moves by years, not quarters.
The next few days supply the context. The placement is due to settle on Wednesday, 26 August — the same morning the July core personal consumption expenditures index, the gauge the Federal Reserve builds its forecasts on, is published at 8:30 a.m. Eastern, with consensus at 0.2 per cent month on month and 3.3 per cent year on year, unchanged. After the close that same day, Nvidia reports second-quarter results, the single most consequential date for the entire capital expenditure complex. Then, from Thursday to Saturday, the Jackson Hole symposium, where Chair Kevin Warsh delivers his first keynote in the role on Friday — three weeks before a September meeting that markets currently price with roughly a one-in-three chance of a rate hike.
Seen in that order, the Alibaba placement is less a story about one Chinese company than a data point about the funding side of the largest investment cycle since electrification. On Monday the stock paid for the dilution. Whether it also gets paid for the arithmetic will not be decided by the placement, but by the margin these data centres produce three years from now.
Try TradingView Free for 30 Days
Plus get a $15 discount on your first subscription through this link.
