Three-month return of 54%, Ping An Fund’s Yao Wenqiang: Optimistic about the AI market in July and August, don’t worry, every round of excess returns comes from macro fluctuations, optimistic about five sub-segments.

Three-month return of 54%, Ping An Fund’s Yao Wenqiang: Optimistic about the AI market in July and August, don’t worry, every round of excess returns comes from macro fluctuations, optimistic about five sub-segments.

```

On June 11, Ping An Fund manager Yao Wenqiang shared his views on AI regarding industry, investment, and strategy at the company's mid-term strategy meeting.

Yao Wenqiang is the head of the TMT group and fund manager at Ping An Fund. He has a short career in the industry, joining Ping An Fund Management Co., Ltd. in November 2025, and serving as the manager of Ping An Technology Select Hybrid Initiated Securities Investment Fund since March 9, 2026. Previously, he was head of investment development at Huajin Fund, investment manager at Blue Arrow Aerospace Space Technology Co., Ltd., and analyst at Northeast Securities Research Department.

The Ping An Technology Select Hybrid A Fund under his management has shown outstanding performance. Since he took over on March 9, 2026, the return during his tenure has been 59.58%, with a three-month return of 53.93% (the fund's overall gain this year is 94.6%).

The representative from “Investment Workbook” compiled the key points as follows:

1. Last year’s trade war, this year’s geopolitical conflicts, and recent discussions of trade crowding, the World Cup opening, Changxin IPO, and macro interest rates.

All these factors, under AI—a wave of the largest industrial revolution for humanity—are external factors.

We focus only on one point: Are current models developing well, fast enough, able to solve practical problems, and driving major companies to continue increasing capital expenditure to form a closed loop?

2. Every wave of excess returns comes from macro volatility. So don’t worry about macro volatility. Every time we see macro volatility we get excited, although the process is tough, macro volatility brings chip exchange, and the trend of technological industry development does not change. Once macro volatility eases even a little, the tech sector reaches new highs immediately.

3. Compute power demand will remain tight for the next 2 to 3 years, this big proposition is very clear. High input in compute power brings rapid model iteration, further drives rapid penetration of applications, and then further boosts compute demand—the closed loop has already formed.

4. Optical communication, why are we so optimistic about it? Optical communication is the only asset in the A-share market that can contend with US stock storage. In US stocks, storage is the best; in A-shares, optical communication is the best. It's our industry advantage, allowing us to sit at the North American AI table and lead technical iteration.

5. May-June is the main volatility period for the AI market. During the volatility, there are various doubts, and all kinds of factors are amplified. I think it’s not about what people around us say, but what the overseas AI giants are actually doing.

Moving into July and August, the overall AI market has performed relatively well in the past three years. Overall, we are optimistic about the AI market in July and August.

In terms of mid- to long-term logic, enterprise-side agent explosion in the second half of this year, and next year’s model self-training explosion, are enough to support very high expectations for the AI industry for quite a long time.

The Ping An Fund service account article also summarized Yao Wenqiang’s core viewpoint: the AI industry is entering a new stage of “capacity is king,” with the market paradigm shifting from “rewarding capital expenditure” to “rewarding expansion after return verification”;

In specific areas, Yao Wenqiang is most optimistic about the entire optical communication industry chain, believes that the 800G product life cycle will exceed market expectations, NPO (Near Package Optics) is the main investment line, CPO (Co-Packaged Optics) progress is anchored by global compute chip giants catalysis, OCS (Optical Switching) has seen explosive orders, and DCI (Data Center Interconnection) demand is beyond expectations.

He especially pointed out the “serious inverted” pricing mismatch opportunity between optical devices and optical modules, and that companies with high silicon photonics technology ratios will continue to benefit.

In addition, he thinks the optical fiber segment is replicating the “cycle to growth” path of optical modules and storage, requiring 1–2 quarters of performance for verification. For storage, PCB (upstream > downstream), MLCC (cycle starting) and other directions, he suggests deep-diving into individual stocks rather than sectoral allocation.

The following are the highlights compiled by Investment Workbook (WeChat ID: touzizuoyeben), shared with everyone:

Three-year review of the AI industry:2023 pretraining→2024 multimodal→2025 strong reasoning

All the investment targets we dig for, even the main investment lines of each year, must have underlying industrial support.

We see it very clearly, in 2023 AI burst onto the scene. At that time the focus was on pretraining working miracles, GPU was at the forefront, so the chain centered around Nvidia was the top priority in our tech industry in 2023.

After Chinese New Year 2024, I remember it vividly. When the Sora model came out, everyone was surprised to find the model could also do multimodal. Multimodal really pushed the boundaries of model capability further. So in 2024, around edge-side AI, including embodied intelligence, our research on robots was also incorporated into the AI application framework.

At the end of 2024, the wave of edge-side AI also widened the investment performance gap between different tech fund products to some extent, thanks to various major model makers focusing on multimodal.

Moving to 2025, the time point is closer to us, easier to review now, coding capability emerged. At first, models replaced programmers to do lots of coding. Later, people found coding was not only programming, but also enabled models to autonomously complete long workflow tasks.

Because humans enter the digital world via code. When models master code, various tasks in the digital world can be completed by models.

So in the first half of this year, agents further drove the explosive growth of tokens. Strong reasoning became a key development direction for AI in 2025.

Of course, in this stage you also see some companies for which multimodal isn’t the focus anymore, but they still push it. For example, the Sora project has now stopped.

In 2025, everyone found the direction for earliest commercial value of AI to land. On the inference side, you have to emphasize high-speed interconnection and lots of long context text. Now million-level context text has become standard. Storage has also become a very important focus since 2025. This is the three-year industrial development thread.

Global AI trading in 2026 seeking new value opportunities

From an investment perspective, I divide it into three stages, as shown in the chart below with three sandpiles: from Q1 2023 when AI just started, to Q3 last year, I characterize it as a continuous sand-addition stage.

AI developed vigorously, large model capabilities rapidly improved, major companies’ capital expenditure accelerated. In this process, any sector with AI elements would have been a major investment direction.

Last year especially, all AI sub-tracks made sure to allocate to all leading companies, and certainly achieved very good investment returns. But turning from last year Q4 to this year Q1, during which my fund officially started operation.

With continued capital expenditure acceleration, major companies faced cash flow pressure, compute power-related sectors saw significant rises, and the market’s AI trading gradually met the slope threshold in the “sandpile model”—AI trading is looking for new value opportunities.

I faced a big challenge as soon as I started managing the product, especially in Q1 this year; a lot of AI giants stalled. Nvidia didn’t rise for over half a year from Q4 to Q1. Google did a bit better, but other companies like Meta, AWS, Amazon all fell. What does this reflect?

A lot of doubts exploded. Meta raised capital expenditure for the first time and its stock price got huge negative feedback. The higher the capital spending, the bigger the cash flow scissors gap. People were worried about whether ROI would form a good closed loop. That was a hesitant phase.

But from Q1 to Q2 this year, we saw macro-level geopolitical conflict. This conflict overshadowed the quantitative data inside the AI industry and what will finally cause qualitative change.

The war happened in March. From that point, you could feel, domestically there was already a lobster trend. Overseas, personal agents, terminal calls, and API calls were rapidly rising; from then, expected annualized revenues for model makers like Anthropic also rapidly rose.

It’s just that the macro situation led to many factors being ignored. In other words, in Q1–Q2 2026, the AI industry was in a good development phase that was accelerated.

So once the war and geopolitical conflicts eased, we immediately emphasized valuation reversion for large-cap targets.

Looking at US stocks, previously when major companies raised capital expenditure, the market basically punished them with negative stock trades. But at the end of Q1 this year, people found high capital expenditure had indeed led to rapid model capability improvement. Especially Meta, which closed the gap greatly with leading model makers, so later the market rewarded them with positive stock feedback, yielding a big rebound.

For A-shares, from that point, some well-known AI targets started to rise.

Here I’ll add some expectations for later data, since overseas model makers’ revenue data will determine cloud makers’ capital expenditure and our judgment of AI models.

Last night, we received new data: Anthropic currently forecasts annualized revenue of about $49 billion, rising to $60 billion by the end of June, $110 billion by year-end, and a target of $550 billion next year. You can sense this growth—it’s very fresh data.

Enterprise-side Agents explode in the second half, model self-training explodes next year

Just now we reviewed past AI development from both industry and investment sides.

Personally, our conviction in the AI industry and optimism about subsectors comes from several points:

First, in the first half of this year, personal-side agents exploded rapidly. I think API calls are just the tip of the iceberg. In the second half, enterprise-side agents will explode widely. You can observe that all model makers are launching enterprise-side agent tools and services.

Another important point is that all our research starts from the model.

From 2023 and 2024, as just described, the model was simple—just a chatbot for conversation. By 2025, it could help with work. Now, many fund managers and colleagues around us use large models, and many are heavy paid users.

How will models develop next? I think from the second half of this year to next year, especially next year, model self-training will be a very important topic. If this year is the age of inference, then next year will turn to model self-training.

What is model self-training? It’s simple. For example, recently a friend returned from North America after investigating the job market. The hottest topic is programmers, with computer graduates facing challenges. Second most in-demand are workers who train models.

Because people found that tuning model parameters and setting algorithms can already be done by models themselves. Model self-training means: models generate data themselves, feed themselves, establish algorithms, their own reward feedback mechanisms, and generate results.

When the trend for model self-training forms, we’ll see the next GPT generation, say GPT-7, iterating much faster and stronger.

Now the model release cycle is one major version per year, one minor version per quarter. This determines how much model capability improves each time, and how the secondary market responds in investment accordingly. In the future, we expect to see a major version update every quarter as the norm, and model capabilities rising exponentially.

If you use agents, you’ll find they can only work continuously for 10–12 hours. But what if the model can work nonstop for 24 hours, or for a month? I think model self-training will bring us…

If next year develops as we expect, then from next year’s perspective, the API call volume today is still just starting out.

What impact will this have?

Big theme is clear: Compute power demand continues tight for 2–3 years

This brings us to the investment report.

Our tech investment approach is very clear: compute power demand remains tight for the next 2–3 years, this big theme is obvious.

High compute power input drives rapid model iteration, which further drives rapid application penetration, and then boosts compute power demand—this closed loop is formed.

Based on these consensuses, if you look at the AI supply chain, you’ll find only one logic behind it: exponential growth in tokens and token calls.

Just mentioned Anthropic’s revenue, with forecast at $110 billion this year and $550 billion next year. That's the demand side. Looking at overall industry supply, all supply is linear growth. One side exponentially growing demand, the other side linear supply.

No supply chain can double its output this year, quadruple it next year, tenfold the year after, no company is that aggressive. Plus, many key supply chains are in overseas hands. Each country and culture decides expansion pace differently.

The core investment theme is simple: find the supply-demand gap. So this year every inflation segment comes from shortage caused by that gap.

With the global AI infrastructure bucket effect, I list familiar directions, divided as follows. On the left: GPU, CoWoS + wafer capacity driving the TSMC semiconductor chain, and AI electricity.

These are the three directions we call certainty growth. If you want certainty, these are must-choose options.

The directions on the right include optical communication (the big category is connectivity): optical communication plus upstream PCB, and CPU and storage. We call these high-elasticity directions.

Why? Because no matter if capital expenditure this year is $700 billion or higher, next year $900 billion or $1 trillion, the structure growth in the right directions far outpaces overall capital expenditure growth.

For example, optical communication, why are we optimistic? It's the only A-share asset that can compete with US stock storage. US stocks’ best is storage, A-shares’ best is optical communication, our industry advantage, sitting at the North American AI table, leading technical iteration.

Growth from last year to this year is over triple. I predicted the next year would double this year, but now expectations have changed.

Next year is still over three times this year, year after next is more than double next year. The inner logic is that optical communication’s share of big capital expenditure was 5%, but will quickly climb to over 20% in the next 2–3 years.

Besides optical communication, CPU and storage also have inner reasons for rapid improvement.

We look for investment opportunities around the bucket effect’s weak links. Here’s our chart this year showing the tightness at each level: moderate tightness, high tightness. Including recent semiconductor equipment categories starting price hikes.

Every wave of excess returns comes from macro volatility,don’t worry,optimistic about five sub-directions

I quickly report the core viewpoint on several directions.

A big premise: tech industry. Tracking over the years, every wave of excess return comes from macro volatility. So don’t worry about macro volatility. Every time we see macro volatility we’re excited, even though the process is tough, it brings chip exchange; tech industry development trends do not change. As soon as macro volatility eases, tech sector hits new highs.

We are also glad that amid this volatility, our products are still net inflows. Last year’s trade war, this year’s geopolitical conflict, recently discussed trade crowding, World Cup opening, Changxin IPO, macro interest rates—all these factors, under AI—the largest industrial revolution for humans—are external factors.

We focus only on one point: Are current models developing well, fast enough, able to solve practical problems, and driving major players to continue increasing capital expenditure to form a closed loop. That’s our only research framework, based on this consensus.

First, the optical direction: over-performing into next year and the year after, growth certainty in A-shares is very clear. Currently, any company’s issues exposed are all at the supply side, not demand.

Supply-side is easiest to resolve, also indicates prominent supply-demand contradiction. Speaking of sub-solutions, here’s my view.

Just now, Mr. Zhang Yu mentioned on the Hong Kong Stock Exchange, global overseas capital has a high value level for domestic quality industry chains. Domestically, relevant optical communication targets also go public in Hong Kong.

So this year’s big board value level will be anchored to Hong Kong Stock Exchange. For example, companies like Accelink listing in Hong Kong, could have much higher premiums than in A-shares. Assume $100 billion profit this year and next; maybe a dozen times PE in A-shares, but in HK? Possible to reach 30 times. I think it’s plausible.

Second, optical fiber: I am very optimistic. Will it go through the cycle-to-growth transition like optical modules and storage did? Later on, AI-driven demand for optical fiber will supersede military and operator main demand.

Third, storage: We deep-dive individual stocks, as good storage names are mainly overseas.

Fourth, PCB: Suggest focusing on the upstream. PCB and optical modules are opposite: optical modules realize value downstream; upstream makes money by expansion but not by price rise; PCB upstream’s entire pricing is growing, making money on price, with stronger bargaining power.

Fifth, we are also optimistic about MLCC, a cycle probably just started.

What’s interesting is MLCC’s key enterprises are in Japan. But Japan is not raising prices in all categories. Japan’s laser drill bits, weaving machines are all out-of-stock nationally, but prices are not up.

Last MLCC cycle was in 2021; industry-wide high/low capacity prices tripled before Japanese players started price hikes. But this time, Japanese firms have already raised prices, so the signal is clear—the cycle may just be beginning.

Optimistic about July-August AI market

Finally, recharge your conviction.

In June, we compared the AI to the NEV market of previous years, and the chart is clear: May-June is the main volatility period. During volatility, doubts abound, and factors get amplified. I think it’s not about what others say, but what the overseas AI giants are actually doing.

So basically in July-August, the overall AI market has performed nicely the past three years. Early July sees earnings and quarterly reports.

Throughout July-August, earnings show it’s not about which internal AI company missed or beat expectations, but that the whole sector’s earnings show a big comparable advantage versus the industry.

Overall, we are optimistic about the AI market in July-August.

Mid- to long-term, as noted: enterprise-side agent explosion in the second half, model self-training explosion next year, both are enough to sustain highly positive expectations for AI for quite a long time.

Source: Investment Workbook Pro, author Wang Li

Risk Warning and DisclaimerThe market has risks, investment needs caution. This article does not constitute personal investment advice, nor does it consider the individual investment objectives, financial status, or needs of particular users. Users should consider whether any opinions, views, or conclusions in this article fit their specific situation. Invest accordingly at your own risk. ```