Global liquidity peaks: the companies that spent most aggressively on AI in H1 are discounted first
A rising market needs new money to keep taking the other side. Global liquidity is already showing signs of peaking, and the companies that invested most aggressively in AI in the first half are shifting from being rewarded to being discounted.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
A low break-issue rate is not a good signal
Do not read Hong Kong's low IPO break-issue rate as a strengthening market. In the first half of this year the break-issue rate was only in the teens, far below the roughly fifty-fifty level normal under a market-based mechanism — meaning the main participants in new listings are short-term speculative arbitrage funds, not long-term investors. Recently the supply of listed companies has shrunk, and IPO-related funds have blown up, yet the two newly listed companies still broke issue, showing that temporary control on the supply side has not solved the fundamental problem: Hong Kong faces the same peak in liquidity.
The more aggressive on AI, the sooner you are punished
Market attitude reversed within a single quarter. Before Q1, the capital market rewarded aggression: the louder the AI CapEx, the higher the valuation. By the time half-year reports were disclosed in late July, any large internet company that had made aggressive capital expenditure in the first half — especially those that spent their free cash flow to zero — was discounted, and even Google was not exempt. Q2 was precisely the quarter of maximum FOMO across the whole supply chain, with everyone afraid of price increases and spending money ahead of time; inventories could still hold up gross margins. By Q3, when inventories are exhausted and the price increases come back, the negative factors will show up together.
Nvidia is closer to Cisco in 2000
Nvidia's earnings look good, but its cash flow is changing shape. Before the dot-com bubble burst in 2000, Cisco rolled out "lease instead of buy" to help customers finance purchases, recognising future revenue early as sales revenue; Nvidia is now pushing lease-instead-of-buy as well. Free cash flow in the report halved quarter on quarter, and receivables on the books are far above historical levels — speculated to be partly related to the credit enhancement and collection structure that lease-instead-of-buy brings. Nvidia may therefore have reached a stage where it needs more aggressive financial engineering to sustain capital-market expectations, similar to Cisco's position back then.
Primary-market money comes from bullets loaded three years ago
The answer lies in when the funds raised. The money concentrated into AI and robotics from H2 2025 to H1 2026 mainly comes from funds raised at the 2020-2022 market peak; from 2023 to 2025 they barely deployed, until they saw state capital betting during a period when no one dared to invest and earning the most — the stories of ChangXin and ChangCun, two large models and two large chips stimulated market-based capital, and the mentality was: copy the state capital's homework. So this is not a rational long-term increment, but more like a delayed emotional release.
The new property policy detaches households from the funding chain
The key to recent real estate policy is not saving developers, but pulling households out of the old financing loop. Buying a pre-sale home used to mean the household played the role of financing object in the investment-driven chain: pay first, receive the home later; after the policy pushes the chain toward marketisation, buyers get completed homes and bear only the consumer role, with risk reset back to developers and financing institutions. The policy was introduced when prices were near the bottom in order to first stabilise the denominator on the asset side, avoiding panic from dismantling the chain during a decline. Japan after the 1990s and the consolidation of China's coal and mining industries both followed similar paths.
A slow bull is not a slogan, it is a chip structure
A slow bull does not rely on slogans; it relies on layering market chips by duration: at the bottom, low-valuation high-dividend blue chips taken up by insurers as long money that rarely trades; above that, public funds and index funds, performing a quasi-stabilisation function; above that, China Securities Finance and Central Huijin adjusting the index; and at the top, retail, private funds and active managers. The new insurance rules require liabilities and assets to match in term, return and liquidity — institutionally ensuring long money invests long, medium money medium, and short money short — and do not allow ordinary people's short money to be put into ten-year low-liquidity assets.
Crises get bigger with each rescue, and bubbles at least one and a half times
After the dollar left gold in 1971, global central banks became increasingly accustomed to intervening directly in crises by printing money, each time faster than the last. The quantitative easing after 2008 should have been cleared in a rate-hiking cycle, but was interrupted by political intervention, and then came the pandemic in 2020, so a larger flood of money was layered on before the previous bubble had dissipated — today at least one and a half bubbles have accumulated. Central bank governors have learned to print more money immediately when a crisis appears, and the endogenous result is inflation, rising financial assets and widening wealth inequality all at once. As long as the bubble is not allowed to clear, the world may face long-term stagflation in some form.
In their own words · checked verbatim
The reason I say it feels a bit precarious is that today the whole world has a bit of a liquidity-peak problem.
我说有点风雨飘摇的原因 就是今天全球都有一点流动性见顶的问题了
In the capital market it was still whoever spends aggressively is the impressive one, and then when everyone published half-year reports it became whoever is aggressive gets punished.
资本市场还是谁花钱激进 我就觉得谁厉害 然后到大家公布半年报的时候 就变成了谁激进惩罚谁
Perhaps Nvidia has also reached a point where, in some sense, it needs lease-instead-of-buy to drive revenue growth and meet capital-market expectations.
也许英伟达也到了某种意义上的 需要借助以租代买来推动收入增长 达到资本市场预期
Everyone looks at it and says, if you can pull this story off this way, then I can certainly pull this story off this way too — psychologically it is roughly like that.
大家一看说 你都能把这个故事这么搞一遍 那我也肯定能把这个故事这么搞一遍 从心理上大概就是这样
The so-called slow bull we always talk about is not slow and bullish in itself; on the basis of a slow bull it first has to build a chip structure.
就所谓我们老讲的慢牛 不是由于他本身既慢且牛 他要在慢牛的基础上 先做一个筹码结构出来
When the bubble piled up last time had not yet dissipated, a bigger one was added on top — and added without hesitation.
上一次堆积起来的泡沫还没有消散的时候给他加上了一次更大的 而且是没有犹豫就加上去了
Figures
| Hong Kong H1 2025 IPO first-day break-issue rate | in the teens | 0:33 |
| Total US stock market capitalisation | over 70 trillion USD, about 2.3 times GDP | 19:10 |
| Bank loans as a share of social financing in the period | 60% | 30:19 |
| Direct financing as a share of social financing in the period | just over 30% | 30:19 |
| US direct financing as a share of total financing | over 80%, with indirect financing in the teens | 30:19 |
| Equity public fund scale (source refers only to equities) | from one to two trillion to about five trillion | 57:33 |
| Upper limit on insurers' equity asset allocation | can now allocate up to 30% | 1:03:34 |
| Household time deposits maturing H1 2025-2027 | 50 trillion | 1:07:36 |
| China household sector debt ratio | down nearly two percentage points | 1:11:38 |
| Time taken to complete the previous three-and-a-half-year QE total after March 2020 | seven months | 1:19:47 |
Glossary
- Capex
- Money a company invests in long-term assets; here mainly capital spending on data centres and compute.
- FOMO
- Fear of missing out on money others are making, so the whole supply chain spends ahead of price increases.
- Mega 7
- The seven largest US technology companies by market cap, used by the show to observe whether the capital cycle is still rising.
How to listen
For investors watching both US AI stocks and China/Hong Kong asset allocation, and for primary-market practitioners judging where 2025's hot money comes from and its rhythm.
The case-by-case discussion of Hong Kong IPOs can be skipped; enter the main thread at 8:11 with Nvidia's liquidity experiment.