qinbaFrank: AI Computing Power Wave Review and Outlook — From Nvidia's Three Major Debates to Optical Interconnects and SpaceX IPO, How Is Capital Rotating?

Source: Cynthia, Hong Kong Ethereum Community Hub

Guest: qinbaFrank— US stock and crypto secondary market investor, long-term practitioner of deconstructing macro, industry, and individual stock logic from first principles

On June 8, 2026, Futu, SNZ, ETH HK Hub, and Sharplink jointly hosted a VIP event. Veteran investor qinbaFrank delivered a presentation titled "AI Computing Power Wave: Review and Outlook," systematically outlining the complete path the AI market has traveled since 2023: from the three major market debates on "whether computing power is necessary," to how penetration rate dividends determine commercialization efficiency, and finally to the current critical phase of shifting from hardware scarcity to commercialization validation.

He also provided a framework for judging the magnitude of this correction——three scenarios: valuation kill, earnings kill, and logic kill—and explained why this AI wave is "similar in form but different in substance" from the 2000 dot-com bubble.

Disclaimer: The content of this article faithfully presents the guest's shared viewpoints and does not constitute any investment advice, product sales solicitation, or profit guarantee.

I. Why a Risk Alert and Some Position Reduction on June 3rd

Starting from 2023, I have written some thoughts on macroeconomics and this round of AI/computing power market trends. In June 2024, I recommended Palantir on X, believing it had 3-5x upside potential as a representative of defense and military AI. At the time, this view was highly controversial in the market, but looking back, it indeed staged a very substantial rally.

This is my first time doing an offline sharing like this. I want to take this opportunity to systematically organize my overall framework for this AI market cycle: how it evolved, where it stands now, and the possible directions it might take in the future.

Last Wednesday evening (June 3rd), I was interviewed for over two hours on X by the US stock community 168X. The core view was: the market has been a bit "too hot" recently and needs some cooling down and adjustment. The specific reasons are as follows:

First, sentiment is overly crowded, with excessive FOMO. Capital concentration in popular sectors has reached a rather extreme level. Parabolic rises are unsustainable, while orders and earnings reports have yet to fully catch up.
Second, SpaceX's IPO roadshow triggered institutional portfolio rebalancing. During the SpaceX roadshow, many institutions began reducing related holdings and freeing up funds in advance, rather than waiting until the official listing date to act——this capital rotation and siphoning effect often manifests early.
Third, geopolitical tensions are fueling risk-off sentiment. US-Iran negotiations remain volatile, compounded by last Friday's non-farm payroll data and this week's CPI data, leading to an overall decline in market risk appetite.
Fourth, the non-farm payroll data shocked rate cut expectations. If May's new non-farm payrolls significantly beat expectations, the market will re-price a higher interest rate trajectory.
Fifth, this week's CPI data is the real policy variable. Strong non-farm payroll data alone is insufficient to determine whether rate hikes will occur; the real key is core CPI——especially whether rising energy prices will transmit and spread to service sector prices. This is the core variable to watch closely over the next one to two weeks.

The core dividing line for judging the magnitude of this correction is: Pure digestion of funding/crowding typically results in only a minor correction; an upside surprise in inflation data could escalate it to a minor-to-moderate level; only a significant deceleration in AI commercialization or cloud revenue would signal a reset of the entire narrative. Overall, I believe the market needs some time to digest and wait in the short term. Previously overcrowded popular sectors may enter a mild or moderate pullback phase until the next "macro signal" emerges to provide relief.

II. Review: The "Three Great Debates" of the AI Market Over the Past Three Years

To understand the current position, it's necessary to review the complete path of this AI market from 2023 to the present. I believe this was not a simple straight-line rally, but a wave-like market driven by repeated cycles of "market debate—validation—re-debate."

First Debate (Second Half of 2023): Is Capital Expenditure Really Necessary?

In the first half of 2023, this main theme was primarily valuation-driven——earnings hadn't significantly improved yet, but stock prices had already rallied (roughly several times). This coincided with a global semiconductor industry downturn, and the market was deeply divided on "how much computing power AI actually needs." Consequently, the second half of 2023 was characterized by high-level consolidation.

Second Debate (Early 2024 to Early 2025): Will Big Tech's Capital Expenditure Continue to Accelerate?

In Q1 2024, NVIDIA's earnings began to improve sequentially, and major tech companies' capital expenditure also started accelerating. This led the market to gradually confirm that "computing power demand is a real trend." A landmark event was: at the Davos Forum in early 2024, OpenAI's Sam Altman proposed that trillions of dollars would need to be invested in chip manufacturing capacity in the future. This statement was highly controversial within the industry at the time, with management from both NVIDIA and TSMC publicly expressing disagreement, suggesting such a massive scale of investment was unnecessary. However, as subsequent capital expenditure from major cloud providers consistently exceeded expectations, the market gradually accepted this judgment——the scale of power and computing required for new data centers in the US is indeed in the trillions of dollars.

During this phase, capital flowed from Big Tech's capex into NVIDIA and the upstream supply chain, driving the main upward wave of 2024.

Third Debate (Early 2025): Is Computing Power Overestimated?

In Q1 2025, the release of a large model with significantly improved training efficiency triggered market skepticism about whether "so much computing power is really needed," leading to a notable stock price correction. This was followed in February by another sharp drop due to changes in US tariff policy, with related core targets pulling back considerably from their highs——this marked the second major correction of this market cycle.

Third Phase (Second Half of 2025): Consensus Formation

By Q2 and Q3 of 2025, the market broadly felt a significant improvement in the capabilities and practicality of large models. Application scenarios shifted from "training-centric" to "inference-centric," and increases in model parameter scale and multimodal capabilities further drove up computing power demand. During this phase, Big Tech's capital expenditure entered a new round of acceleration, and the market correspondingly entered a new round of rallying.

III. Core Framework: Penetration Rate Determines Commercialization Efficiency

My personal judgment on how far a technology wave can go hinges on the penetration rate, not simply on whether "the trend exists."

Many people compare this AI market cycle to the 2000 dot-com bubble. I believe the two are "similar in form but different in substance": both experienced parabolic price increases where valuations ran ahead of earnings, but the industrial environments are vastly different.

Around the year 2000, US internet penetration was only just over 30%, and business models (advertising, e-commerce, gaming, value-added services) were still in the exploratory phase. Therefore, after the bubble burst, it took the Nasdaq a considerable amount of time to climb out of the trough.
The mobile internet era around 2010 was different: After the iPhone's release in 2007 and the opening of the Android system, mobile internet penetration in China and the US completed the leap from early adoption to mainstream within roughly a decade (2010-2018)——much faster than the two to three decades it took for the internet. This was because the previous generation's infrastructure (internet penetration, information dissemination efficiency) laid a very solid foundation for the next.

Today, we face an environment where billions of people globally are already accustomed to using WeChat, social media, and various apps——the speed of information dissemination and public acceptance of new technologies is completely incomparable to the year 2000. This is precisely the biggest difference between the current AI industrial environment and the internet of 2000.

Specifically regarding the judgment method, I quite agree with a key node in the "Technology Adoption Life Cycle" (Crossing the Chasm theory): a 10% penetration rate is the critical point. Below 10%, the technology is still in the "early validation" phase, and whether it is revolutionary enough determines if it can gain traction; once it crosses 10%, it signifies crossing into the mass market, and the growth slope typically becomes steeper; the 10% to 50% range is the core observation window and the "golden period" for related industry investment——user base expansion and increased willingness to pay occur simultaneously, with token consumption rising accordingly; beyond 50%, the incremental space begins to diminish marginally.

Referencing a survey data point: a major investment bank's survey on enterprise AI procurement willingness shows that this ratio rose from about 10% last September to about 18% by the end of March this year——meaning enterprise AI penetration has crossed the critical point and officially entered a period of rapid growth.

If we compare this AI wave to the previous two generations of technology waves: PC internet took about 20 years from 1990 to 2010 to achieve penetration; mobile internet took less than 10 years from 2010 to 2019; and AI, starting from 2023, may diffuse even faster. The core reason is that the more complete the infrastructure, the shorter the commercialization cycle——in the mobile internet era, smartphones, 4G, app stores, and mobile payments drove mass adoption; today's AI stands on the infrastructure of cloud computing power, model APIs, social dissemination, and Agents, making information diffusion and commercialization methods more mature than any previous generation.

IV. AI vs. Internet: The Essential Difference in Commercialization Logic

The core problem the internet solved was "the efficiency of connection and information dissemination"——it reduced the intermediate costs of information flow, logistics, and capital flow, but it did not directly replace "humans."

AI is different: it directly replaces human cognition and labor. When an AI's capability reaches or exceeds the "societal average" level of a human employee, it brings not just efficiency gains, but genuine substitution——meaning that a company paying for AI is essentially equivalent to the cost it previously paid to employ that portion of labor. This is also why many people (myself included) quickly upgrade their payment for AI tools from free versions to tens or even hundreds of dollars per month, or even pay for multiple large models simultaneously——once you experience that "it indeed does better and faster than me," the willingness to pay rises very decisively. Therefore, once AI surpasses the societal average intelligence level, its commercial value will rise exponentially at a rapid pace.

This also echoes a question raised by a previous guest: under the trend of AI rapidly replacing cognitive labor, how will the "moat" value of an individual's professional knowledge and experience change? This is one of the fundamental reasons why AI commercialization is more complex than that of the internet.

V. Investment Logic of the Computing Power Supply Chain: From "GPU Single-Point Narrative" to Systematic Revaluation

The logic of this round of computing power investment is spreading from simply betting on GPUs to a systematic revaluation of the entire chain, including storage, CPUs, interconnects, power supply, packaging, and edge hardware. The overall framework can be summarized in a three-stage structure: short-term focus on "resource scarcity," mid-term focus on "system upgrades," and long-term focus on "Physical AI penetration rate".

1. Scarcity Pricing: GPU Demand Spills Over to Storage and CPUs

The logic chain is: long-context, multimodal, and Agent applications are driving up storage demand——HBM tightens first, then the pressure transmits layer by layer to DRAM/GDDR, NAND/SSD/HDD, and finally to the CPU scheduling link, and then to power supply.

First came the GPU shortage. The years 2022-2023 were a downcycle for the global storage industry, and a large amount of production capacity was cleared out. Entering 2024, as major cloud providers accelerated their capital expenditures, the impact of this capacity clearance began to show.

Then came the storage/HBM shortage. HBM itself has a complex production process and slow yield improvement. After the brutal overcapacity of the last cycle, major storage manufacturers are very cautious about expanding production, and new capacity will not be gradually released until the second half of 2027. This has led to a significant increase in the bargaining power of storage manufacturers when signing long-term supply agreements——long-term contracts are signed for 5 years, require 10%~30% prepayment, and even demand downstream customers provide financial guarantee instruments. This is why these companies exhibit the characteristic of "performance rising ahead of valuation": performance has consistently exceeded expectations over the past few quarters, but valuations were suppressed by market fears of "repeating the semiconductor cycle," until the existence of long-term agreements gradually convinced the market that cyclical fluctuations would be "smoothed out," and valuations began to repair.

Next is the CPU scheduling shortage, and finally the power shortage. The core reason is that a large number of orchestration and scheduling tasks in data centers are not suitable for GPU processing and must rely on CPUs. Taking the NVIDIA NVL72 rack as an example, the current configuration is roughly 72 GPUs paired with 36 Vera CPUs, meaning the CPU:GPU ratio is about 1:2 (early solutions were about 1:8); market expectations suggest this may move closer to 1:1 in the future, meaning the importance of CPUs (whether Intel, AMD, or custom ARM chips) in computing infrastructure is being repriced. Further down the chain, this transmits to the issue of data center power and grid capacity.

2. Upgrade Pricing: Synchronous Upgrades in Optical Interconnects, Power Supply, and Advanced Packaging

The second main line is the "upgrade logic"——the core is not "whether this module exists," but whether conversion efficiency, power consumption, power supply density, and packaging yield can continue to improve.

Optical Interconnects: Optical modules are evolving towards LPO/NPO/CPO. Co-packaged optics (CPO) integrates optical chips and electrical chips more closely, theoretically reducing power consumption, but it has not yet achieved mass production. Some field research suggests that major cloud providers are unlikely to adopt CPO on a large scale before 2027——the core concern is reliability: traditional optical modules can be directly replaced if broken, whereas if a CPO fails, it involves board-level replacement costs and verification cycles, and major companies still need time to fully verify yield and failure rates.

Power Supply Network: Evolving from 48/54V to 800V HVDC. This is very similar to the high-voltage path in the electric vehicle industry——early EVs generally used lower-voltage power supply architectures with lower efficiency; later, manufacturers including BYD and Huawei successively shifted to higher-voltage DC architectures, with higher voltage, lower current, and less loss. Data center power supply systems are undergoing a similar upgrade path, which is also driving demand in the power semiconductor (such as silicon carbide) and power management related industry chains.

Advanced Packaging: 3D stacking + glass/ceramic substrates. This is similar to the evolution path of smartphone chips in recent years——when the marginal benefit of performance improvement from simply shrinking process nodes becomes increasingly low, the industry turns to breaking physical limits through more advanced packaging methods (such as 3D stacking, glass or ceramic substrates), using better materials and packaging processes to continue improving overall performance.

3. Long-Term Pricing: Edge Computing and Physical AI

The long-term logic is that edge computing and Physical AI are entering the application validation stage——from on-device inference of small models, to robotics and autonomous driving, then to mass production and cost reduction, ultimately forming a new penetration rate curve. The short-to-mid-term tracking focus is on storage, CPU/ARM, optical interconnects, power equipment, and advanced packaging; the long-term focus will be on the mass production curves of robotics and autonomous driving.

VI. Evolution of Investment Main Lines: From Physical Constraints to Vertical AI OS

After the tight supply of computing power eases, the market's focus will undergo a migration path: Physical constraints (computing power/capacity shortage) → Enterprise deployment layer (can enterprises turn AI into a production system) → Vertical AI OS (controlling the industry workflow entry point) → Physical AI (entering the real physical world).

The essence of the enterprise deployment layer is not simply plugging in a chat box, but rewriting the enterprise's workflow: first find high-frequency, high-labor-cost, and verifiable-result workflows, then connect to the enterprise's private data (involving RAG, permission management, data lineage, knowledge graphs), enabling Agents to truly execute actions (calling APIs, SaaS, completing approval and rollback processes), and continuously measure task completion rates, takeover rates, costs, and ROI.

The so-called "Vertical AI OS" can be understood as the industry's intelligent control layer——unlike traditional SaaS where "humans operate software," an AI OS is where "AI calls tools and advances processes, while humans are responsible for supervision, approval, and decision-making," essentially a combination of System of Intelligence + Action + Governance. Core indicators for judging progress at this stage include: whether commercialization continues to accelerate (model ARR, cloud revenue, number of enterprise customers), whether deployment quality has truly passed the production line (task completion rate, human takeover rate, accuracy rate), whether economic viability is closed-loop (unit inference cost, ROI, gross margin), and whether a moat has formed (private data, process depth, compliance auditing).

VII. The Underlying Anchor of the Wave-Like Uptrend: Model ARR and Cloud Revenue

Whether the market narrative can continue does not hinge on "whether valuations are expensive," but on whether the ARR (annualized recurring revenue) of model vendors and cloud business revenue continue to maintain high growth——this determines whether the capital expenditures of large tech companies are reasonable, and whether the prosperity of the entire computing power chain can be sustained. This transmission chain is: Real demand (actual payments from B/C-end) → High growth in model vendor ARR → Cloud business exceeds expectations → Computing power chain continues to benefit.

Around this transmission chain, three scenarios can be discussed:

Scenario One: Growth rate has not slowed, logic has not reversed. If model vendors' ARR is still growing and cloud business continues to exceed expectations, it indicates that the rationale for capital expenditure remains valid, and the order logic for the computing power chain continues to hold. In this case, even if there is a small-to-mid-level pullback due to short-term overbought conditions or valuations being "deemed expensive," the fundamentals are not damaged——declines are often fast, and recoveries are also fast. Once earnings season arrives or new applications emerge, a reversal may quickly follow.

Scenario Two: Growth falls short of expectations, narrative resets. If model vendor performance clearly stalls, or the cloud business demand chain shows a clear slowdown, it indicates the problem is closer to the "commercialization origin"——because much of the cloud computing power procurement comes from these model vendors themselves. In this case, it would be at least a mid-level adjustment, requiring new evidence to prove that scale and growth rate can exceed expectations again before confidence returns.

Scenario Three: Macro/funding factors are "amplifiers," not root causes. Macro and funding factors affect market sentiment and discount rates, but they only escalate to core risks when they truly impact the commercialization level. Specifically, it can be divided into three layers: a simple withdrawal of funds or a single CPI exceeding expectations is usually a small-level adjustment; if combined with persistent inflation, no rate cuts, and geopolitical risks, it may escalate to a small-to-mid-level adjustment; only when model ARR or cloud revenue shows a real slowdown does it enter a mid-level logic reset.

Simply put: As long as large model ARR and cloud revenue have not slowed down, this round of adjustment is more like a repricing at the valuation and funding level, not a 2000-style crash; once fundamentals truly stall, one must wait for new evidence of a reversal.

VIII. Current Phase: Moving from Hardware Scarcity to Commercialization Validation

In the phase from April to June this year, the market's core assumption is: the capital expenditure guidance of major cloud providers will continue to exceed expectations, and the support behind this is the real paid demand from enterprises and consumers for cloud services (i.e., cloud business revenue growth rate). If this assumption holds, it means capital expenditure is "reasonable and sustainable," and the entire supply chain——storage, optics, CPUs, chips, all the way to power and grids——will benefit from it.

Looking ahead, I believe the market's focus will gradually shift from "hardware scarcity" to "commercialization realization". A report in May this year mentioned that in the enterprise service market, the best-selling product category is actually AI implementation/consulting services——that is, the ability to help enterprises truly implement AI into specific business processes. The logic behind this is: the core production processes and experiences of many industries are not public documentation, but are accumulated in the experience of senior employees; the training data of large models themselves does not contain this "tacit knowledge." Whoever can help enterprises combine this industry know-how with AI will seize the opportunity in the next phase.

My personal judgment is: as long as this growth rate itself does not show significant deterioration, any subsequent pullbacks caused by macro factors (such as interest rates, tariffs, etc.) are more likely to be small-to-mid-level phased adjustments, rather than a trend reversal. What truly needs vigilance is a scenario where the overall growth rate of AI commercialization significantly falls short of expectations——only then would the valuation logic of the entire sector truly need to be reassessed.

IX. Historical Reference: A Three-Tier Framework for US Stock Market Adjustments

When judging the level of a US stock market adjustment, looking at the magnitude of the decline itself is meaningless; the key is whether the trigger source overturns the long-term logic——is it a pure impulse to kill valuations, a macro event shock, or a reset of the entire industry narrative. Using the Nasdaq as a benchmark (due to its purer tech attributes), pullbacks over the past 20 years can be roughly divided into three tiers:

L1 Small Level (single-digit percentage decline): The trigger is usually an impulse to "kill valuations" after rising too fast, superimposed with liquidity shocks or disturbances in inflation/rate cut expectations. This kind of adjustment is not a crisis; fundamentals have not changed, and once the disturbance is confirmed to have eased, the reversal is usually very fast. A relatively recent example is the roughly 7%~8% pullback last November, mainly a liquidity shock superimposed with just-sprouting market doubts about AI capital expenditure.

L2 Mid Level (about 15% decline): Usually accompanied by certain major macro events or market mechanism shocks, risk needs to be repriced, but it does not mean the underlying order has collapsed; the market needs to wait for new data to confirm the risk has not spread further. For example, the roughly 15% pullback from August to October 2023 occurred against a backdrop of the 10-year US Treasury yield approaching 5%; the pullback in July-August 2024 was related to carry trade unwinding and market recession fears.

L3 Large Level (over 25% decline): Means the accustomed macro logic has been reset, or the industry's long-term narrative has been overturned; risk appetite undergoes a systematic reassessment, requiring entirely new evidence to rebuild confidence. Historical examples include the 2008 financial crisis (halved), Q4 2018 (about 25%~30%), the March 2020 pandemic shock (about 30%~40%), the 2022 rate hike cycle (about 33%~35%), and the roughly 28% pullback caused by tariff or global trade order shocks.

Applied to the current round of AI market trends, the core dividing line remains whether the AI commercialization growth rate slows down: if model ARR, enterprise user numbers, token revenue, and cloud business revenue continue to exceed expectations, it means the business logic has not been reversed, and pullbacks are more likely small-to-mid-level adjustments caused by funding or macro disturbances; if model vendor performance falls short of expectations, it means we are closer to the commercialization origin, requiring at least a mid-level repricing and waiting for new evidence; only when AI growth slows down, simultaneously superimposed with systemic risks like surging inflation, geopolitical conflicts, or the rupture of global order, could it escalate to a large-level adjustment.

Simply put: As long as AI commercialization has not slowed down, this round of adjustment is more like a "repricing"; only when evidence of commercialization shows a gap does it mean the entire framework needs to be reset.

X. Conclusion: AI is a fundamental leap in civilization’s foundational capabilities

Finally, I’d like to share my personal understanding of the nature of this wave. Historically, gunpowder, the steam engine, electricity, and the internet were essentially “single-point industrial revolutions”—they upgraded a specific tool, energy source, or information channel, solving a key bottleneck before spreading along the industrial chain, presenting an S-curve of a single technology cycle. These revolutions changed “a certain dimensional capability,” rather than directly enhancing intelligence itself.

I believe AI is different—it enhances “intelligence,” the most fundamental foundational capability. It can be compared to humanity’s use of fire: moving from not using fire to using it brought not just “one more tool,” but cooked food changed our physical structure, which in turn affected brain capacity, ultimately leading to an expansion of civilization’s overall capabilities. AI is similarly altering foundational capabilities—the entire suite of perception, reasoning, generation, decision-making, and action is shifting upward as a whole. This is a foundational upgrade at the level of the “civilization production function,” rather than simply making a specific tool more user-friendly.

Precisely because it is a leap in foundational capabilities, new industrial revolutions will continue to emerge in batches at the upper levels: the Agent revolution, the robotics revolution, the drone revolution, and then extending to national defense and military, space technology, and the process restructuring of many more industries. This process will not be realized all at once but will appear wave after wave. Therefore, I believe the main thread truly worth tracking is not betting on a specific application explosion, but continuously observing “how intelligent capabilities spill over into the physical world and various industry processes”—this is the core clue for judging how far this AI wave can go.

Looking ahead one to two years, I think people will continuously feel this “acceleration within acceleration”—technological capabilities and commercialization processes mutually validating and propelling each other. However, the market trend itself will certainly not be a straight line; it will exhibit wave-like characteristics amid the logical shifts between “scarcity—upgrade—long-term realization.”

Disclaimer: This content faithfully presents the views shared by the guest and does not constitute any investment advice, product sales solicitation, or profit guarantee.