Eoptolink, Zhongji Innolight, and T&S Communications all surged together today: Eoptolink jumped 16.9%, Zhongji Innolight climbed 15%, and T&S Communications rose 13.9%. Three big bullish candles lined up side by side look like a return to center stage.
But pull back the chart and the story changes. All three companies are still more than 60% below their respective 52-week highs.
They are not alone in the decline. Optical modules, CPO, PCB, servers, and liquid cooling — the entire AI hardware chain has fallen in sync. The same picture applies across the A-share market: broad declines, shrinking volume, and indiscriminate selling of growth stocks. This is not about something going wrong at these three companies; it is an entire industry chain being repriced.
People holding tech stocks have had a brutal two months. Accounts shrank by 40% to 50%, and every close brings the same question: cut losses or hold?
But the earnings picture tells a different story. According to consensus estimates, Eoptolink's net profit is expected to grow 101% in 2026, Zhongji Innolight's by 214%, and T&S Communications' by 66%. Orders are booked into the year after next, capacity is running at full, and customers are still the same handful of overseas cloud providers.
If earnings are so strong, why have the stocks halved? Start with valuation.
The most telling example is Anthropic, whose market valuation shrank from a peak of 2 trillion to 1.2 trillion — a 40% evaporation. Its revenue did not collapse; it actually rose 45% quarter-on-quarter in Q3. Only the valuation collapsed.
High-valuation growth stocks earn money by "discounting many years of future profit into today." That discounting ruler has two scales: one is interest rates, the other is confidence in the endgame. When rates rise or confidence loosens, the ruler shrinks, and valuations collapse — with little connection to current profit.
The same applies to A-shares. ZJ INNOLIGHT (300308) saw its forward P/E fall from over thirty times to twenty times, while Eoptolink dropped from several dozen times to nineteen times. Not a single earnings report deteriorated in between — it was purely a valuation kill.
Earnings did not worsen; the market simply is no longer willing to pay so much for the future.
There is another rule to accept: the longer the duration of an asset, the more sensitive it is to the discount rate. The stock prices of this wave of AI hardware do not reflect this year's profit — they reflect 2028's profit. When the ruler moves, they are the first to collapse.
Now look at policy. For the past two years, it was a nationwide all-in on computing power: big tech poured money into data centers, optical module orders stretched into the year after next, and everyone in the chain dared to expand capacity. Now the wind has shifted: export controls on high-end chips and equipment are being tightened round after round, power and environmental permits for data centers are starting to tighten, and overseas reflection on the pace of AI expansion is growing.
Policy has shifted from pressing the gas to tapping the brake, and the market's first move is to revise computing power demand expectations downward. The three musketeers are precisely the most elastic chips on this front. If cloud providers spend ten billion more in capex, their orders may rise by several billion; conversely, if ten billion less is spent, they are the first to be cut. When expectations are revised, they fall first — and hardest.
However, a policy pivot does not equal an industry pivot. Overseas giants have not yet cut their capex budgets, optical module orders have not disappeared, and what has changed is the slope of growth over the next three years. The market has shifted that slope from steep to gentle and priced it into the stocks all at once. So orders remain while stock prices crash first — these two things are not contradictory.
The third and most critical issue: AI is being forced to do the math. Hyperscale data center construction depends on massive debt and equity financing. High rates make financing expensive, and expansion must slow. On the return side, it gets trickier: ROIC, the most boring indicator, has become the ruler hanging over AI's head. A high-end AI server costs hundreds of thousands of dollars, and at current rental rates and utilization, payback takes years; if computing power supply surges further and rental rates keep falling, the payback period stretches even longer.
Worse is depreciation: billions of dollars of equipment amortized over five to six years makes the first two years' books the ugliest. Slow payback makes ROIC ugly, which makes financing harder, which slows expansion further — a cycle that bites its own tail. AI has turned from a faith back into a business.
So is it cheap now? Eoptolink's 2027 consensus P/E is only around 15 times, ZJ INNOLIGHT (300308) is at 12 times, and T&S Communications is at 28 times, corresponding to net profit growth of 70% to 100%. At the very least, it is not expensive.
But keep one thing in mind: optical modules are an order-driven business — you can see one year out, but not three. After the 1.6T generational shift, no one dares guarantee whose hands the orders will land in. Sell-side analysts have also recently been密集 publishing research reports recommending the optical communications sector. Sell-siders are always optimistic and can only serve as a reference, but at least it shows there is market consensus that "earnings are fine" — the only disagreement left is when the market will be willing to pay.
The tape is also building a bottom. Around November 6, the three musketeers successively hit lows, and the volume-backed bullish candle on November 11 pulled the short-term technical picture into a golden cross. Sentiment heats up first, trend arrives later — this is the typical look of an oversold rebound. This is bottom-building, not a main upward wave.
So will the dragon king return to its throne? My judgment is that the foundation is there, but it cannot rise yet. To truly launch a main upward wave, three things must happen: overseas policy must shift from tapping the brake back to loosening, ROIC must become genuinely calculable with cloud providers willing to endorse the payback cycle, and incremental capital must flow back from value stocks to growth stocks. None of these three can be resolved in three to five trading days.
The batch of core assets in 2021 went through the same process: valuations were killed to the bottom first, then they ground along the bottom for more than a year before the next round arrived. So agonizing over daily candle moves now is meaningless. The question to ask is whether, at this level, what you hold is worth this price. Earnings are doubling, yet the valuation is only in the low teens, while the stock price has already halved and halved again from its high. Cheap is genuinely cheap, but from cheap to rising, there are still three hurdles in between: policy, ROIC, and capital.
Next quarter's earnings report will be the real answer for the three musketeers.
Risk warning: downstream demand falling short of expectations, optical communications technology development falling short of expectations, and geopolitical risks. Note: The industries mentioned in this article are for industrial case analysis only and do not constitute any investment advice. Markets carry risks; invest with caution.
If you are still confused about when the market will stabilize; if you are puzzled about industrial investing; if you find it hard to grasp new hot themes; if you feel at a loss about capital rotation and sector volatility — then scan the code to join us, obtain more detailed industry chain analysis and investment strategies, and follow Gelonghui Research Institute to seize opportunities in the wave of investment and share in the dividends of the industrial revolution!
Comments