Key Drivers and Strategic Recommendations for AI Investment in 2026

Current market conditions show phased volatility: NVIDIA still receives most buy recommendations from analysts despite its high valuation; Microsoft and Google’s cloud+AI strategies are regarded as “wide moat” narratives; while Tesla and Amazon need clearer short-term profit support in their AI boundary exploration. Figure 1 reflects the relative performance of four companies in 2025, highlighting the market’s early response to the divergence among giants [0].

Note: Lines represent the indexed performance of respective closing prices relative to January 1, 2025. The chart marks three event nodes: “Scaling Law Discussion”, “Domestic Computing Power Breakthrough”, and “Space Computing Concept”, indicating the market’s sensitivity to system-level issues; Data and chart source: Jinling API Data [0].

---

1. Power pressure has become a real bottleneck for AI expansion
   : Fab-related surveys indicate that AI data center expansion in 2026 will conflict with the limits of the power system; almost all ultra-large-scale projects in the future must consider the “system urgency” of insufficient power supply/cooling [2]. This constraint is particularly obvious in high-computing-density applications (large model training); any delay in intermediate links will amplify costs and delay output.
2. Scaling Law’s diminishing marginal effect on economic benefits
   : The infrastructure expansion of institutions like OpenAI is also facing the reality of “huge investment but no longer linear returns”; the 2026 outlook mentions that “extreme investment cannot be quickly converted into cash flow growth” [1]. Therefore, relying solely on computing power stacking cannot sustain valuations; it must be upgraded in coordination with power and cooling systems to maintain investors’ confidence in the “system urgency” indicator.
3. Immediate stock price reaction
   : According to Jinling API data, NVIDIA, despite its high valuation, rebounded after a small correction; the market still hopes it will take the lead in high-energy-efficiency accelerators (such as Blackwell); MSFT/GOOGL are supported by high valuations in terms of “cloud+AI integration” system operation capabilities, with short-term volatility lower than pure hardware players [0].

---

1. Google (GOOGL) chain narrative
   : Its “Cloud + Gemini + self-developed TPU for data centers + Bard” narrative is regarded by the market as a representative of “system integration”. Google’s recent combined improvement in AI application landing, advertising efficiency, and cloud subscription growth has allowed it to maintain an annual growth rate of over 60% despite a high PE ratio [0].
2. Microsoft (MSFT) “Azure + Copilot” robustness
   : Its core assets combine “software subscription + cloud infrastructure” with AI efficiency; analysts unanimously believe there is still more than 30% upside potential, especially during the enterprise customer upgrade cycle (e.g., Windows 12/Office Copilot), with low system urgency and strong cash flow generation, helping to buffer the capital expenditure pressure brought by “computing power expansion” [0].
3. NVIDIA as a computing platform
   : Its data center revenue accounts for nearly 90%, most of which comes from CSP customers. If system urgency is constrained by energy or chip production capacity (Nvidia itself/TSMC/foundry), short-term stock prices will amplify volatility (historical data shows NVIDIA’s annualized volatility is as high as over 50%) [0]. This makes it more like a “barometer of system urgency”.
4. Valuation divergence brought by CSP narratives
   : Between targets with clear CSP narratives (e.g., GOOGL, MSFT) and those more dependent on hardware (e.g., NVDA), the market shows obvious “risk aversion preference”: the former bears lower volatility (annualized volatility <25%), while the latter is given high beta but greater potential returns [0]. In the short term, if system supply tension (such as power/chips) is relieved, NVDA’s high-level logic still has room for rebound; otherwise, it will continue to be “highly volatile”.

---

1. Domestic chips and computing power autonomy
   : Bloomberg reports show that Chinese manufacturers accelerated the layout of AI accelerators (such as Huawei Ascend 910C, Dashu Core) and training/inference platforms in 2025, aiming to provide larger-scale, low-power domestic computing power in 2026 to reduce dependence on NVIDIA/AMD [3].
2. Time window: Transition from “existence is reasonable” to “commercially credible”
   : Current domestic computing power still faces problems of limited advanced processes and incomplete software ecology; its “economic efficiency/stability” has not yet matched that of Google/Microsoft-level CSPs. Therefore, the capital market will closely observe whether “system urgency” is buffered by the large-scale deployment of domestic computing power in 2026; if successful, it is expected to form a positive feedback on the valuations of relevant domestic hardware and cloud factories (such as Alibaba Cloud, Tencent Cloud, Huawei Cloud); if failed, it may bring domestic “resource grabbing” risks during the global computing power supply and demand tension period.
3. Investment implications
   : For domestic computing power-themed ETFs or related leaders (e.g., Amlogic, Cambricon), attention should be paid to “deliverability indicators” (prototype mass production time, power/cooling integration capabilities) — the market will remain cautious until these three are continuously verified.

---

1. Energy efficiency advantages and real costs of space data centers
   : According to Reuters/WSJ reports, Blue Origin and SpaceX are advancing orbital AI computing power platforms, emphasizing the strategy of “solar energy + cooling + ground network integration” [4]. Currently, it is still in the proof-of-concept and financing stage (SpaceX plans to raise funds through IPO in 2026), so it is not enough to bring immediate impact on AI investment volatility in 2026.
2. Positive role from the perspective of system urgency
   : If space computing can solve ground power tension, land and water resource constraints, it will become a “system redundancy” buffer pool in the medium and long term. However, in 2026, it still needs to face soft barriers such as launch costs, communication delays, regulation and insurance; short-term valuation driving force is limited.
3. Investment suggestions
   : Allocate space computing as “defensive scarcity” (suggested proportion <5%), focusing on tracking promoters (such as SpaceX financing dynamics, companies with aerospace service suppliers participation), and wait for key tests (first batch of orbital cargo or solar power tests) to be fulfilled before judging expansion.

---

Element	Analysis	Strategy Recommendations
System Urgency	Power supply/chip/data center expansion links are still tight; 2026 risk is higher than 2025	Increase holdings in data center operators with “multi-source power + efficient cooling” capabilities; build positions in batches in high-beta targets like NVDA to control retracement
CSP Narrative	Google/Microsoft’s “chain advantage” built through vertical integration can buffer computing power volatility	Concentrate core allocation on verified AI volume engines like GOOGL and MSFT; use derivatives to hedge short-term volatility
Domestic Computing Power	Existing investors need to wait patiently for supply chain maturity; once mass delivery is achieved, valuations will be repriced	Regularly evaluate domestic accelerator delivery rate and system integration progress; quickly eliminate disjointed targets
Space Computing	High long-term value but unclear short-term returns; need to hedge technical progress risks	Participate through multi-stage financing/cooperation paths (e.g., aerospace service suppliers)

---

1. Short-term (first half of 2026)
   : Focus on system urgency relief signals (e.g., new capacitors for data centers, cooling innovations, chiplet capacity release); conduct spread trading when there is an obvious supply-demand mismatch in hardware platforms like NVDA; build a “CSP narrative” portfolio, using the stable cash flow of GOOGL+MSFT to hedge high-volatility hardware configurations.
2. Mid-term (second half of 2026)
   : Once domestic computing power completes key mass production, consider increasing allocation to relevant cloud factories and upstream supply chains, but need to simultaneously evaluate the maturity of the “independent software ecosystem”; if space computing paths like SpaceX/Blue Origin achieve milestones (e.g., first AI payload test), moderately add relevant aerospace service assets.
3. Risk Control
   : Pay attention to “system urgency” data (power load curves, chip delivery progress, data center PUE changes) and CSP revenue quality (cloud gross profit, AI usage volume).

---

[0] Jinling API Data (real-time stock prices, company overview, technical charts)
[1] Forbes – “Why OpenAI’s AI Data Center Buildout Faces A 2026 Reality Check” (https://www.forbes.com/sites/paulocarvao/2025/12/06/why-openais-ai-data-center-buildout-faces-a-2026-reality-check/)
[2] Forbes – “As AI Booms, Data Centers May Create Electricity Scarcity Among Users” (https://www.forbes.com/sites/kensilverstein/2025/12/15/as-ai-booms-data-centers-may-create-electricity-scarcity-among-users/)
[3] Yahoo Finance/Bloomberg – “Chinese AI Euphoria Masks Long-Term Technological …” (https://finance.yahoo.com/news/chinese-ai-euphoria-obscures-gloomier-000000592.html)
[4] Yahoo Finance/Reuters – “Bezos’ Blue Origin working on orbital data center technology, WSJ reports” (https://ca.finance.yahoo.com/news/bezos-blue-origin-working-orbital-175444569.html)