Pike时政财经分析搬运号

In-depth analysis of AXTI's 80% surge in five days

Looking ahead to the infrastructure planning from 2026 to 2027, the further development of AI models is constrained by the bottleneck of communication bandwidth within data centers. Nowadays, the status of optical communication has fundamentally changed; it is no longer a peripheral auxiliary component but has upgraded to the neural hub of the AI factory, a supercomputing unit. This transformation has directly ignited strong market demand for compound semiconductor materials such as indium phosphide (InP).

In the process of moving towards the 1.6T optical module era, it is a necessary pathway evolved from 400G and 800G, which imposes extremely stringent performance standards on continuous wave (CW) lasers and electro-absorption modulated lasers (EML). Indium phosphide has the capability to efficiently convert electrical energy into coherent lasers, and with its excellent performance in high-frequency response, low noise interference, and high-temperature stability, InP substrates have become the only mature technological solution in the 1.6T era.

From the market landscape, AXTI and its holding subsidiary, Beijing Tongmei, account for about 60% to 70% of the global indium phosphide substrate production. The company's management has revealed that they receive requests from customers for increased production capacity almost every day. In response to this pressure, AXTI is actively renovating existing plants and adding equipment, planning to double the production capacity of indium phosphide by the end of 2026 compared to the end of 2025, aiming to achieve a profit turning point.

Finally, I would like to share my personal investment operation: I opened a position at a price of 17 on January 27, and in just five days, the return reached 60%.

On how to assess the valuation level of the US stock market

Recently, many investors have discussed the situation of LITE with me. A common focus is its high PE of 130 times, leading to doubts about whether it is worth entering the market at this moment. In fact, the core of stock investment lies in purchasing future expectations, so when we analyze, we should focus on Fwd PE, which is the forward price-to-earnings ratio.

In response to this viewpoint, some friends may raise objections, arguing that even looking at future data, the price is still high; for example, the value displayed on the SA platform has reached 74 times. Can we still buy it? In this situation, we cannot ignore the critical indicator of EPS growth rate. For LITE, its growth rate exceeds 100%, resulting in a PEG value of 0.74.

Of course, the data provided by SA may not be completely accurate. As for the specific forward PE and the actual valuation multiples in market transactions, these details need to be researched and confirmed by everyone personally.

In summary, when making stock investment decisions, it is crucial not to focus solely on the current PE; more importantly, one should look at the forward PE and combine it with the PEG indicator to comprehensively assess the reasonableness of the valuation.

According to the analysis of capital flows in the U.S. stock market released by Goldman Sachs in February, I have summarized the market long and short game logic that investors are currently most concerned about. In terms of positive factors pushing the stock market upward, we observe that the net leverage of hedge funds is decreasing, which means that the process of de-risking has occurred, and the market is currently in a stage of clearing chips, rather than accompanied by high positions and high emotions typical of a top. At the same time, market sentiment indicators are quite extreme, with the GS Panic Index reading around 9.22, a value that has entered the historical high score range and is very close to the maximum fear value of 10. Furthermore, as the VIX event-driven causes risk premiums to be squeezed out, the market is expected to welcome a relief rally, and under the assumption of an upward scenario, the CTA strategy will shift to buying strength.

However, the downside pressure facing the market cannot be ignored either. First, although the CTA strategy has sold off some positions, it has not completely cleared out. If the market weakens, especially if it falls below the S&P index's mid-term technical key level of about 6707, there may be an additional selling pressure of up to $80 billion released within the next month, and the scale of selling in a downward scenario will be significantly greater than in an upward or sideways scenario. Second, the positions of Risk Parity and Vol Control strategies remain at historically high levels of 70% to 80%. Once volatility undergoes a structural increase, these strategies will continuously reduce positions as a slow variable, thereby suppressing the market. Third, the liquidity environment is deteriorating, with the S&P Top-of-Book liquidity having decreased by about 70% compared to the average at the beginning of the year, which will amplify any impact caused by selling. Fourth, the role of market makers' Gamma has transformed from stabilizers to amplifiers, with position nature shifting from long gamma to flat or short gamma, which will lead to passive trend-following selling during downturns, accelerating declines and forming negative feedback. Fifth, seasonal factors combined with the retreat of buying power are also a significant hidden danger. As certain buying powers such as 401k, pensions, and early-year asset allocations withdraw, historical statistics indicate that the market typically performs poorly from mid-February to early March, and under the current systemic selling pressure, this seasonal weakness is more likely to be amplified. Lastly, retail funds are showing a marginal weakening trend, no longer mindlessly bottom-fishing for high-risk assets as in the past, and have turned into net sellers in the short term; considering that Crypto and related assets have already seen significant declines at the beginning of the year, if funds further rotate out of U.S. stocks, it will create a double kill situation for both market sentiment and liquidity.

According to the disclosure by The Wall Street Journal on February 6, public opinion across the United States has been thrown into an uproar due to a sudden incident. Trump uploaded a short video lasting about one minute on his social media platform Truth Social. Although he deleted it shortly after, the content has sparked widespread attention. The ending of the video is particularly controversial, featuring the background melody of "The Lion Sleeps Tonight," with the facial features of Barack Obama and Michelle Obama digitally grafted onto the bodies of cartoon apes. This practice cannot simply be summarized as a crude joke; it touches on long-standing and sensitive racist pain points against the Black community, attacking through dehumanizing stereotypes, and has quickly drawn severe criticism from both Democratic and Republican parties. Notably, there have also been rare dissenting voices within the Republican Party. Senator Tim Scott bluntly pointed out that this is one of the most racist things he has seen in this White House and strongly demanded its retraction; meanwhile, several Republican senators have also come forward to condemn it.

In the face of this incident, aside from feeling outraged and shocked, I have nothing more to say. As the President of the United States, to disseminate content containing racial insults on public social media channels is no longer limited to a matter of personal character but is a desecration of public power. Since April of last year, although I have been criticizing Trump's various policies, this incident undoubtedly further exacerbated my negative view of him: the damage he has done to America's national credibility and democratic image is long-term and structural. The American people electing Trump to the White House is undoubtedly a huge mistake. This conclusion is not about his administrative capability but rather because he completely lacks the corresponding qualifications in terms of personal character.

SemiAnalysis has revised its forecast regarding Micron $MU, adjusting its projected share of Nvidia Rubin HBM down to zero. At present, there are no visible signs that Nvidia is placing orders for HBM from Micron. Instead, the supply landscape for Nvidia HBM4 is expected to consolidate around SK Hynix and Samsung, with an estimated split of 70% and 30% respectively.

SemiAnalysis pointed out that we have adjusted Micron's $MU share expectation in Nvidia's Rubin architecture HBM to zero. At this stage, there are no observed signs of Nvidia ordering HBM from Micron. According to forecasts, the supply of Nvidia HBM4 is expected to further lock in on SK Hynix and Samsung, with the supply ratios estimated to be 70% and 30%, respectively.

From my personal analysis, Micron's focus on HBM business in 2026 is mainly concentrated on the Google TPU ecosystem, AMD camp, and some cloud service providers' self-developed ASIC projects. Since Micron's non-participation in Nvidia's Rubin series GPU supply has long been news, SemiAnalysis's move inevitably raises some suspicions of sensationalism. Additionally, Micron's 380 chip indeed offers a very high cost-performance ratio.

In the face of today's extremely attractive low prices, I decisively increased my positions in BE, MU, Lite, and leaps, and the current price is indeed very cheap.

I bought leaps of BE, MU, and Lite today, and I can't help but sigh that the current prices are really too cheap.

According to Matt Garman, the Chief Executive at AWS, the market appetite for GPUs is unprecedented. He revealed that the cloud giant has kept every A100 unit in operation without retiring a single one, while the hunger for previous-generation processors stays remarkably intense. Matt Garman of AMZN revealed that the demand for GPUs is unprecedented. Because of this, AWS has retained every A100 GPU to date, without any decommissioning. Additionally, the demand for non-latest model GPUs remains astonishingly high. From my perspective, the field of artificial intelligence computing power is facing a structural supply shortage, rather than just a simple iteration of products. Older model graphics cards will not quickly exit the historical stage for this reason; the introduction of new graphics cards serves more as incremental supplementation rather than a mere replacement. If you are still worried about whether there is a bubble in the AI industry, there is no need for concern; the current situation is akin to the explosive internet revolution before 1995, with the peak in 2000 still far off.

Regarding the core idea of investing in lite, we have observed that the scale and complexity of AI models are rapidly increasing, including the emergence of trillion-parameter models, ultra-long context LLMs, and Mixture-of-Experts models. This trend compels future data centers to possess larger computing power reserves and achieve more efficient collaboration of GPUs, thus driving the full-scale explosion of expansion models such as Scale-up, Scale-out, and Scale-across. As the bottleneck of system development gradually shifts from pure computing power to interconnect bandwidth, latency, and power consumption control, the acceleration of data centers transitioning from electricity to optics has become inevitable, and lite's product layout is well-positioned during this critical transformation period. Whether it is InP lasers, namely the optoelectronic chip EML, or OCS optical switches and optical modules, they are all direct beneficiaries of this trend. Reflecting on my practical experience, as early as November last year, I first shared my investment insights about lite in the member group, and completed my initial position and subsequent additions at price points of 220, 310, and 340. Current results show that call returns have reached 200%, and the underlying stock has also achieved a return of 100%.

Regarding the recent panic surrounding the appointment of the new Federal Reserve Chairman Warsh, there is actually no need to be overly anxious. Although he has previously published some papers expressing opposition to QE and excessive market intervention by the Federal Reserve, considering the writing context of that time, his identity then, and the macro environment which has changed significantly compared to now, the reference value of these old statements is quite limited. If investors base their current trades solely on his views on the balance sheet from 15 years ago, this is clearly irrational behavior.

In fact, Warsh not only firmly believes in the tech narrative of Silicon Valley but is also a staunch supporter of artificial intelligence enhancing productivity. We can focus on two details: first, Warsh joined OpenAI's Safety and Strategy Committee in 2024 and assisted OpenAI in lobbying in Washington, emphasizing the importance of AI for American national competitiveness and economic hegemony. Secondly, he has been a researcher at the Hoover Institution at Stanford University and a lecturer at the Stanford Graduate School of Business since 2011, maintaining close ties with the elite circle of Silicon Valley. In his view, AI is a revolutionary technology, comparable to electricity or the steam engine, capable of systematically enhancing total factor productivity. Based on the above context, we have ample reason to believe that he would never disrupt the long-term yield curve nor hinder the development of the U.S. real economy and the AI revolution process. Trump recently stated that we need another Greenspan, his intention is very clear, viewing Warsh as the Greenspan-like figure he anticipates: supporting technological prosperity through low interest rates to drive U.S. economic development.

In addition, Warsh has other advantages. His communication and cooperation with Bessent are extremely smooth; both are trusted confidants of Wall Street legend Stanley Druckenmiller. Moreover, due to Warsh's father-in-law having a decades-long deep friendship with Trump, this relationship makes Trump trust him greatly. Therefore, based on this trust and network of relationships, we have no reason to believe that Warsh would mess up the U.S. stock market and the economy, thereby undermining Trump's midterm election layout.

In summary, I believe that Friday's market decline is an overreaction.

Regarding the recent pullback in Bitcoin's market this weekend, we have observed a clear trend: BTC and other cryptocurrencies are transitioning into independent risk asset attributes. Their correlation with tech stocks is gradually diminishing, no longer presenting the kind of synchronized rise and fall. The current market trends are primarily constrained by the liquidity situation of this asset class, the proportion of leveraged positions, the dynamics of ETF subscriptions and redemptions, and the overall market sentiment. Such short-term price declines often do not stem from a sudden deterioration in fundamentals; rather, they are more the result of forced liquidations and capital reallocation during the deleveraging process that create a chain reaction. In addition to market-level factors, there is one more point that cannot be overlooked: the cryptocurrency mining ecosystem has actually become unsustainable. This has never been an ideal business, and in the future, companies with cheap and abundant electricity resources will no longer waste these energy on Bitcoin mining.

SNDK 620. Here, I would like to review the previous investment decision-making process with everyone. As early as November last year, institutions on Wall Street only set a target price of 230, but after in-depth analysis, I determined that this asset was severely undervalued by the market and clearly pointed out that its reasonable price range should be between 400 and 1000. By sharing this core logic, I am also very happy to have substantively helped all of you members achieve substantial investment returns.

AAOI Investment Value Deep Dive

Currently, the demand for InP-based Continuous Wave (CW) lasers in the optical communication field is surging, directly leading to an extreme shortage in market supply. In this context, the main competitors in the industry are facing severe challenges: Lumentum, constrained by InP supply bottlenecks, has been forced to reduce its client list; while Coherent plans to prioritize its internal laser capacity to meet its own product needs. This situation means that module manufacturers lacking internal chip manufacturing capabilities are facing a real risk of supply disruption. In stark contrast, AAOI has successfully transitioned the majority of its laser production processes to in-house research and production, thus gaining control in the supply chain.

In response to strong demand from core clients like Amazon for 800G and future 1.6T optical modules, AAOI is actively deploying capacity expansion, with its internal laser capacity forming the cornerstone of this strategy. Currently, AAOI's wafer fab in Houston produces approximately 200,000 CW lasers per month. The company has established a clear upgrade roadmap, aiming to significantly increase monthly production capacity to 2.4 million by the end of 2026 by introducing new equipment and upgrading wafer sizes from 2-4 inches to 6 inches. When considering the production capacity of over 3 million SiPho pump lasers each month, this scale means that the laser chip segment will no longer be the bottleneck restricting AAOI's optical module production growth. In contrast, many competitors may find themselves passive due to insufficient upstream chip supply when expanding module assembly lines.

Strategic barriers of domestic manufacturing in the U.S.: In an increasingly complex geopolitical landscape and an environment filled with uncertainty over tariff policies, AAOI possesses not only module manufacturing capabilities but, more importantly, the ability to produce core laser chips domestically in the U.S., which grants the company a high strategic value. AAOI is currently one of the few suppliers that can manufacture EML and CW lasers on U.S. soil. Cloud service giants like Amazon or Microsoft, in order to ensure supply chain diversity and security and to hedge against risks that Asian supply chains may face, are often willing to pay a premium for such domestic suppliers.

Potential Opportunities and Challenges Analysis

Upside Risks

1. Volume growth of 800G/1.6T products driven by large customer orders:
If the capacity expansion plan set by the company can be smoothly implemented, for example, achieving a monthly capacity target of 100,000 800G modules by the end of 2025, this will directly drive significant revenue growth for the company.

2. U.S. manufacturing dividends amid geopolitical dynamics:
Tightening geopolitical relations and tariff pressures are prompting hyperscalers to shift their supply chains away from Chinese manufacturers. With its domestic manufacturing strength, AAOI is expected to capture more market share from Chinese competitors like Innomlight.

3. Cost optimization and profit enhancement brought by automation technology:
The high level of automation at AAOI's Texas facility is remarkable, significantly reducing labor demand by 85%. This move not only effectively addresses the challenge of high labor costs in the U.S. but is also expected to yield gross margin performance exceeding market expectations.

4. Recovery of cable TV CATV business and Charter's upgrade cycle:
AAOI is benefiting from the network upgrade plans promoted by cable TV operators like Charter (involving DOCSIS 4.0 and 1.8 GHz amplifiers). This segment is performing strongly, with revenue exceeding expectations, providing the company with solid cash flow support.

Downside Risks

1. Uncertainty in execution and capacity expansion (core concern):
Regarding yield and delivery: The manufacturing process of optical modules is extremely complex, and fluctuations in yield and cost differences are the norm. Historically, the execution capability of AAOI's management has been inconsistent, and large-scale capacity expansion presents enormous operational challenges. Furthermore, the qualification timeline for customers (especially Amazon) is uncertain, and any delays in any aspect may lead to a postponement of revenue recognition, while the company must bear high expansion costs in advance.

2. Risk of over-reliance on a concentrated customer base:
AAOI's revenue is highly dependent on Microsoft, Amazon, and Charter (through agents like Digicomm). Losing any major customer or a sudden slowdown in demand due to adjustments in AI capital expenditures could severely impact the company.

3. Financial condition and equity dilution risks:
Regarding equity dilution and debt issues: To raise funds needed for capital expenditures (Capex), AAOI has conducted multiple ATM offerings and issued convertible bonds. This has led to a doubling of share capital, inevitably diluting the earnings per share for existing shareholders.

4. Risks of technological iteration and replacement:
Potential threats from CPO co-packaged optics: From a long-term perspective, CPO technology has the potential to replace traditional pluggable optical modules. If this technology is deployed in hyperscale data centers faster than the market expects, AAOI's existing product line will face obsolescence pressure. However, considering that this technology is not expected to mature until after 2028, there is currently no need for excessive concern.

TSMC's announced Capex situation effectively dispelled market concerns about a potential bubble in AI investments, while the explosion in popularity of Clawbot further solidified confidence in the outlook for computing power demand. Driven by these factors, investor sentiment has significantly improved, and the market has fully entered a Risk On state, with IREN and CRWV looking very promising for a return to previous highs.

The market sentiment in the Neocloud sector is undergoing significant recovery, primarily due to the rising rental costs of the first generation H100 graphics cards. In fact, considering the widespread shortage of optical modules, power supplies, and storage devices, it is an inevitable trend for cloud computing providers to raise prices, a prediction I had made earlier. Looking back at November, crwv had already clearly disclosed at the UBS Technology Conference that all current contracts signed are for take-or-pay agreements lasting 5 to 6 years. This should reflect a favorable situation of severe supply-demand imbalance, but unfortunately, the market was heavily pessimistic at that time, and investors even priced in extreme expectations of crwv's impending bankruptcy. In terms of personal operations, when the stock price was in the range of 95 to 100, I did significantly reduce my holdings, but for my position in iren, I have not made any reductions yet.

The call about sndk was made when the stock price was 220 yuan, and I have to say that this profit is really exaggerated.

【Deep Analysis of Sndk's Strong Surge Today】

SNDK surged 9.5% today against the trend, with the market generally believing that this is due to investors gradually realizing that the strategic position of NAND in data center infrastructure during the ACCel era (i.e., the 'After Claude Code' era) is significantly improving.

This trend is primarily driven by the transformation of NVDA's BlueField and DeepSeek's Engram architecture—the role of NAND is undergoing a qualitative change, becoming more like a 'slow RAM', thereby effectively improving its traditional disadvantages in latency. In terms of market sentiment, Citi has shifted to a more positive stance and raised its relevant forecasts and target prices; meanwhile, there is also good news from the supply side, as Asian media reported that Samsung and SK Hynix plan to cut rather than increase NAND production this year.

From an institutional perspective, while SemiAnalysis also holds a positive view on NAND, their stance is relatively conservative, believing that NVDA's ICMSP will only bring limited incremental growth (LSD) to NAND until 2027. We believe this assumption is overly conservative as it does not take into account the demand for backward compatibility and direct access to the Blackwell platform; according to our calculations, the new demand generated from this could approach a double-digit percentage of total NAND demand.
-TMTB News

**【Personal Commentary】**
To be frank, this news from TMTB is actually two months late; I had already anticipated this trend back in November.

What’s more noteworthy is that at the UBS Technology Conference in December, SNDK's management clearly stated: NAND is the only product in the CSP field that can be expanded without being constrained by GPUs, while also enhancing AI inference performance. This statement from management further bolstered my confidence in continuing to hold.

【Analysis on Today's SNDK Surge】

SNDK recorded a 9.5% increase today, indicating that investors are gradually realizing the strategic position of NAND is significantly improving in the data center infrastructure of the ACCel era (i.e., the "After Claude Code" era). This trend is primarily attributed to NVDA's BlueField and DeepSeek's Engram architectural transformation—under these new architectures, NAND's functionality is gradually approaching that of "slow RAM," effectively improving its previous disadvantages in latency.

On the market side, Citibank has shifted to a more positive attitude and has raised its related forecasts and target prices; meanwhile, according to Asian media reports, Samsung and SK Hynix will adopt a strategy of reducing rather than increasing NAND production this year. Although research institution SemiAnalysis shares a positive view on NAND, they believe that NVDA's ICMSP will only have a limited incremental impact on NAND demand until 2027 (LSD). In this regard, TMTB believes this assumption is overly conservative because it does not take into account the demand that is backward compatible and can be directly connected to the Blackwell platform; according to TMTB's estimates, the new demand may approach a double-digit percentage of total NAND demand.
——Source: TMTB

**【Personal Commentary】**
It must be pointed out that TMTB's news is actually delayed by 2 months, as I made a similar prediction back in November last year. In addition, at the UBS Technology Conference in December, SNDK's management also clearly stated: NAND is currently the only product that CSPs (Cloud Service Providers) can expand without being constrained by GPUs and can simultaneously enhance AI inference performance.

OpenAI published a significant article on January 18, with a very clear core message: ChatGPT is at a critical stage of transitioning from a mere 'practical tool' to 'infrastructure', and its business model is undergoing profound evolution.

The article presents a key point: in the current AI era, computing power is both the scarcest resource and the most core strategic asset. Data shows an astonishing consistency between OpenAI's revenue growth and the expansion of computing power—when computing power increases by 3 times, revenue also grows by 3 times. This phenomenon conveys a clear signal: market demand is not a problem at all; the real bottleneck lies on the supply side. In light of this, OpenAI is committed to transforming its computing power strategy from the past approach of 'single reliance and passive limitation' to building a 'multi-vendor, manageable asset portfolio'. This strategy aims to leverage high-end hardware to break through capability limits while using low-cost platforms to enhance efficiency, with the ultimate goal of reducing the marginal cost of AI to a level that is 'acceptable for everyday work'.

Another important point emphasized in the article is that the next stage of AI will be dominated by Agents (intelligent entities) and continuous workflows, rather than being limited to one-time conversational interactions. The future focus will shift to those intelligent agents that can 'run continuously, possess contextual memory capabilities, and execute tasks across tools'. For individual users, this represents an efficient project management and planning execution assistant; for organizations, it equates to an operating system serving knowledge work. This shift means that user engagement frequency will be higher and stickiness stronger, making the business model more predictable.

At the end of the article, OpenAI looks ahead to the future development of AI: the monetization models of AI will not simply stop at the current subscription or API services. As artificial intelligence technology deeply penetrates high-value fields such as scientific research, healthcare, energy, and financial modeling, diverse business forms such as licensing models, IP sharing, and outcome-based pricing will emerge in succession.

CRWV 102! IREN 58! Neocloud has once again become the market's hot commodity! One can't help but exclaim, the changes in investor sentiment are really too fast.