Ai

Xiaomi’s current downturn cannot be attributed solely to an “AI spending spree” narrative. Over the past year, the stock price fell steadily from a high of around HKD 60 in July 2025 to HKD 25.84 on June 11, 2026. What is truly difficult is the convergence of three factors: smartphone profits are squeezed by rising storage costs; the auto sector has moved from rapid volume growth to a stage of subsidy tapering and model switching; and AI investments have pushed the market’s patience for free cash flow further into the future.

Looking back at this on Beijing Time June 9, 2026—the line from June 5 is no longer sufficient. On June 8, the A-share market failed to digest last Friday’s retracement of tech stocks; instead, the ChiNext board, STAR 50, CPO, and semiconductors continued to decline sharply. Simultaneously, US stocks also saw intraday pullbacks on June 8, with QQQ, SOXX, and a group of AI chip stocks exhibiting obvious rebounds. When looking at both markets together, the conclusion is even less straightforward: A-shares are continuing to dismantle crowded trades, while US stocks are performing high-elasticity rallies intraday. Neither of these events serves as concrete evidence that “clearing has been completed.”

After observing the capital movements of large model companies these past few days, it’s easy to confuse two issues.

Zhipu and MiniMax are both pursuing the A-share listing path. Anthropic has already made a confidential filing of draft S-1 to the SEC, and OpenAI was also reported by Axios to be preparing a confidential IPO prospectus. Having these pieces of news lined up suggests that this industry has finally reached its peak/harvesting period.

But merely opening up an IPO window does not mean that the profitable window is open. Large model companies do have revenue—some revenue streams are growing very fast. What has not generally turned around yet is the set of accounts related to net profit, operating cash flow, and sustained model investment.

/goal is easily misinterpreted as a command to “let the agent work for a bit longer.”

This, of course, is merely its surface manifestation. If you give Codex a goal, it can continuously progress around that objective, instead of stopping after a single round of answers. But what is truly noteworthy is not how long it “runs,” but rather that it converts “what constitutes completion” from a temporary reminder into an intrinsic part of the task itself.

A standard prompt describes what needs to happen next. A goal, however, is more like attaching a checklist/acceptance form to an agent: What is the objective? Where are the boundaries? Which validations must pass? What conditions must be met for it to be considered complete?

When the A-share semiconductor and AI hardware chains surge rapidly, two extreme reactions tend to emerge: one type of person feels that missing out on gains (FOMO) is more painful than actually incurring losses, while another believes that excessive rises necessarily signal a bubble.

Both of these are too fast. In areas like semiconductors, computing power, optical modules, and storage, the industrial logic might be true. AI training and inference will indeed boost hardware demand, and domestic substitution has certainly provided narrative space and order opportunities for local companies. The problem is that just because the industrial logic holds true does not mean that the probability of investing in it now is good.

Similar market trends have occurred repeatedly in history: the liquor sector (Baijiu), new energy, pharmaceuticals, core asset grouping, and TMT. Each time, there was real logic behind it. When these sectors decline, it doesn’t necessarily mean the logic has disappeared; rather, the timing/rhythm between valuation, positioning, earnings realization, and liquidity was off.

After Zhipu and MiniMax were included in the Hang Seng Tech Index, the most common question that arises is: Since they have not yet reached their first lock-up period, are index funds compelled to buy shares?

This question cannot be answered directly based on emotion.

Inclusion in the index must first adhere to publicly disclosed methodologies. Delisting will affect future supply and stock price pressure, but it is not a hard threshold within the Hang Seng Tech Index methodology. Passive funds buy before and after the index takes effect because their goal is to track the index, not because the index company is arranging exits for existing shareholders.

Let’s first establish the date. ChatGPT’s public research preview version was released on November 30, 2022, not 2023. [1]

After this point, NVIDIA’s data center GPU main roadmap is quite clear: The conclusion of Ampere, followed by Hopper taking over. Hopper focused on expanding VRAM capacity and refresh rate, while Blackwell will shift its focus from “single-card dense compute power” toward “inference throughput, power consumption, and system-level interconnection.” The China-specific versions represent a different story: A800, H800, and H20 are fundamentally compliance versions created under US export control constraints, and therefore cannot be viewed using the same metrics as the global flagship line.

In the previous article The end point of this semiconductor cycle is probably not in 2026, I presented my conclusion first, but deliberately did not dive too deep into the specific details of the financial reports.

What we are covering this time is the part that is most easily obscured by market sentiment: When semiconductors rise, everyone knows they are profitable; but what truly determines whether a cycle can be extended or which company can capitalize on high growth more thoroughly is often not the stock price, but rather the profit and loss statement, capital expenditure, and product investment direction during the trough.

If I must make a more specific judgment, as of May 13, 2026, I still do not pinpoint 2026 as the end of this upcycle. However, if I have to pick just one major player among the giants that is most worth watching, it would be SK hynix. Not because it hasn’t gone through a downturn—quite the opposite—but because it made the most representative strategic choices when things looked their worst in 2023.