On Tuesday (October 27), A-shares adjusted with reduced volume, and the three major indices fell rapidly in late trading, with the ChiNext Index dropping over 3% after previously rising more than 1%, and over 4,900 individual stocks declining across the market.
Computing power hardware such as optical modules came under renewed selling pressure, with Eoptolink Technology Inc (SHE: 300502) falling 5.6%, Suzhou TFC Optical Communication Co Ltd (SHE: 300394) dropping more than 4%, and Zhongji Innolight Co Ltd (SHE: 300308) declining nearly 3%.
Among popular ETFs, the underlying index of Huabao ChiNext Artificial Intelligence ETF (SHE: 159363), which focuses on optical module leaders, surged then retreated, closing down 4% at a new low for the year, with investor attention recently running high.
Where to start
At the current juncture, how should one view computing power and optical technology? Huatai Securities stated that the exhibitions and exchanges at this year's OCP (Open Compute Project) summit confirmed that the computing power industry trend remains intact, and this week's North American cloud厂商 earnings could become a key turning point for market expectations, with some AI commercialization increments potentially reflected in related cloud platform revenues. The optical module and optical fiber sectors have already undergone a round of bottoming, and with earnings catalysts approaching, it is recommended to watch for valuation repair opportunities among leading companies.
What signals does Anthropic's prospectus reveal?
Beyond the above industry trends and earnings catalyst observations, the release of Anthropic's prospectus has become a major event in recent AI investment focus. The order, funding, revenue structure, and cost information it disclosed provide another window for observing the ability of AI computing power demand to be realized.
Cao Xuchen, fund manager of Huabao ChiNext Artificial Intelligence ETF (159363), commented that the rising expectations of a CoreWeave bond default and the statements from Broadcom and Intel CEOs regarding capital constraints essentially both point to market concerns about the ability of major AI companies to fulfill orders. Although Anthropic's prospectus does not provide a new narrative for the AI computing power sector, it shows that AI LABs have not overdrawn future orders and hold substantial cash to pay for future orders.
Four signals from the prospectus
1. Computing power investment commitments are relatively clear: Broadcom TPU leasing approximately $161 billion; SpaceXAI (NVIDIA GPUs) up to $121 billion, about $3.1 billion per month starting from December; Google no less than $115 billion, plus $125-150 billion in TPU capacity for 2028-2031; Amazon no less than $110 billion; Microsoft no less than $45 billion; Nscale approximately $45 billion; Lambda approximately $41 billion; AMD over $20 billion. Total approximately $780-810 billion.
2. Capital burn rate is lower than expected: cumulative financing to date is $141.1 billion, current book cash plus short-term investments is $131.7 billion, with only about $10 billion spent so far, indicating a capital consumption pace below expectations. Cao Xuchen stated that if the potential $100 billion raised through an IPO and the $77 billion in borrowing facilities provided by Google, Amazon, and Broadcom are included, Anthropic's disposable funds after listing could be estimated at over $300 billion.
3. Demand authenticity is relatively high: Token consumption billing model accounts for 80% of total revenue in the first half of 2026, with various subscription services accounting for 20%, and To B customers dominating. The net revenue retention rate on a trailing three-month basis is approximately 900%; customers paying over $100,000, $1 million, and $10 million annually number approximately 11,900+, 1,900+, and about 250 respectively; no single customer accounts for more than 10% of revenue.
4. Gross margin performance is strong: gross margin reached 85% in the first half of 2026. Assuming operating costs are entirely inference computing power, the ratio of inference computing power to training computing power spending in 2025/first half of 2026 is approximately 0.61/0.36; training computing power spending growth this year was actually faster than inference. The typical task cost for Fable 5.1 in September dropped about 25%, and for Opus 5.5 dropped about 40%, which the company attributed to a combination of price reductions and efficiency improvements while maintaining profit margins.
Overall assessment
In summary, Cao Xuchen pointed out that the possibility of large-scale credit risk exposure in data centers in the first half of next year may be relatively limited, and computing power order payment capability may be strong; long-term profitability of large model companies may be better than the market's relatively conservative expectations. Zhipu can serve as a sample for observing broad A-share tech sentiment, and the recent recovery may reflect early signs of expectation correction, but whether it constitutes a starting point remains to be seen.
In terms of ETF direction, Cao Xuchen stated that at the current stage, more attention should be paid to the possibility that the industry cycle may be longer than expected and that quality companies will continue to deliver earnings. It is recommended to watch allocation opportunities in Huabao ChiNext Artificial Intelligence ETF (SHE: 159363), Huabao Hong Kong Stock Connect Information Technology ETF (SHE: 159131), and Huabao STAR Chip ETF (SHE: 589190).
For exposure to optical module leaders plus AI applications, it is recommended to watch Huabao ChiNext Artificial Intelligence ETF (SHE: 159363) and its off-exchange feeder funds (Class A 023407, Class C 023408), focusing on optical module CPO leaders while also covering AI applications. The underlying index has over 35% weight in "Zhongji Innolight + Eoptolink + TFC Optical Communication," making it a core flagship for AI computing power.
Data source: Shanghai and Shenzhen Stock Exchanges, Wind, etc. Note: As of September 30, 2026, according to the China Securities Index, the top three constituents of the ChiNext Artificial Intelligence Index are Eoptolink (weight 12.98%), Zhongji Innolight (weight 12.09%), and TFC Optical Communication (weight 10.8%).
Reminder: Recent market volatility may be significant, and short-term gains or losses do not predict future performance. Investors should invest rationally based on their own financial situation and risk tolerance, and pay close attention to position and risk management.
*Institutional views reference source: Huatai Securities "How to View Computing Power and Optical Technology at the Current Juncture."
ETF fund-related fee disclosure: When investors subscribe or redeem fund shares, the subscription and redemption agent may charge a commission of no more than 0.5%. On-exchange trading fees are subject to actual charges by the securities company, and no sales service fee is charged. Feeder fund-related fee disclosure: Huabao ChiNext Artificial Intelligence ETF Initiating Feeder C does not charge subscription fees; redemption fees are 1.5% within 7 days and 0% on or after 7 days; sales service fee is 0.3%. Huabao ChiNext Artificial Intelligence ETF Initiating Feeder A subscription fee is 1% for amounts under 1 million yuan, 0.6% for 1 million yuan (inclusive) to 2 million yuan, and 1,000 yuan per transaction for 2 million yuan (inclusive) and above; redemption fees are 1.5% within 7 days and 0% on or after 7 days; no sales service fee is charged. According to the fund manager's assessment, the risk levels of Huabao ChiNext Artificial Intelligence ETF, Huabao Hong Kong Stock Connect Information Technology ETF, and Huabao STAR Chip ETF are R4-Medium High Risk, suitable for investors with Aggressive (C4) rating and above. Please refer to the sales institution for suitability matching opinions.
Risk disclosure: Huabao ChiNext Artificial Intelligence ETF passively tracks the ChiNext Artificial Intelligence Index. The index base date is December 28, 2018, and the launch date is July 11, 2024. The annual returns of the index from 2021 to 2025 were: 17.57%, -34.52%, 47.83%, 38.44%, and 106.35%, respectively. The annualized volatility of the index over the same period was 23.73%, 27.34%, 38.02%, 45.42%, and 41.1%, respectively. The constituent stocks of the index are adjusted in due course according to the index compilation rules, and the backtested historical performance does not predict future index performance. The index constituent stocks mentioned in this article are for illustrative purposes only. Individual stock descriptions do not constitute investment advice in any form, nor do they represent the holdings or trading activities of any fund under the manager. Any information appearing in this article (including but not limited to individual stocks, comments, forecasts, charts, indicators, theories, and any form of expression) is for reference only, and investors are responsible for any investment decisions they make independently. In addition, any views, analyses, and forecasts in this article do not constitute investment advice in any form to readers, nor do they assume any responsibility for direct or indirect losses caused by the use of the content herein. Fund investment involves risks, past performance of a fund does not represent its future performance, and the performance of other funds managed by the fund manager does not constitute a guarantee of fund performance. Fund investment requires caution. MACD golden cross signals have formed, and these stocks are performing well!
Comments