- Abstract\n\nConclusion first: The current AI cycle is closest to the 'second half of 1999' in the dot-com bubble timeline, characterized by 'accelerating bubble while rate hike expectations turn hawkish,' and has begun to show features of the 'Q1 2000 top window.'\n\nProbability Distribution:\n\nPhase | Current Similarity\n---|---\nSecond half of 1999: Policy turns hawkish but bubble continues to accelerate | 45%\nQ1 2000: Near the valuation top | 30%\nJune 1999: First policy turn to hawkish | 15%\nSecond half of 2000: Orders and capex begin to deteriorate | 10%\n\nCore Judgments:\n\n1. Real AI demand remains strong. AI-related revenue, orders, or capex for NVIDIA, AMD, Broadcom, Arista, Dell, Microsoft, Alphabet, Amazon, and Meta are still in the upward revision range. NVIDIA recently disclosed quarterly revenue of $81.6 billion, up 85% YoY, with a gross margin of about 75%; AMD data center revenue was $5.8 billion, up 57% YoY; Broadcom Q2 AI semiconductor revenue was $10.8 billion, up 143% YoY. (nvidianews.nvidia.com)\n\n2. Capital expenditure has entered a high-pressure state. Microsoft's single-quarter PPE increased by $30.876 billion, Alphabet's Q1 capex was $35.674 billion, Amazon's Q1 PPE purchases were $44.203 billion, and Meta's Q1 capex was $19.84 billion with FY2026 capex guidance rising to $125-145 billion. (microsoft.com)\n\n3. Fed policy expectations are shifting from 'rate cut friendly' to 'rate hike risk.' As of the recent FOMC, the target range is 3.50% to 3.75%; the effective federal funds rate on June 4 was 3.62%. May non-farm payrolls added 172,000, unemployment was 4.3%, wages were up 3.4% YoY, April CPI was 3.8% YoY, April PCE was 3.8% YoY, and core PCE was 3.3%. (federalreserve.gov)\n\n4. Credit spreads have not yet given a 2000-style collapse signal. On June 4, the 2-year Treasury was 4.05%, the 10-year was 4.47%, and the 10-year TIPS real rate was 2.11%; HY OAS was 2.74%, and CCC OAS was 9.46%. This indicates rising valuation pressure, but systemic credit pressure has not yet fully erupted. (fred.stlouisfed.org)\n\n5. AI Storage Conclusion: HBM is a structural bottleneck; DRAM, NAND, enterprise SSD, and nearline HDD are a mix of real AI demand, supply discipline, price hike cycles, and inventory replenishment. TrendForce notes that conventional DRAM contract prices are expected to rise 58% to 63% QoQ in Q2 2026, and NAND Flash contract prices are expected to rise 70% to 75% QoQ, driven by server and AI demand and NAND allocation to enterprise SSDs. (trendforce.com)\n\n6. Current Trading Conclusion: Not suitable for unprotected chasing of the overall AI beta. Suitable for holding cash-flow leaders, reducing exposure to second- and third-tier assets and those dependent on financing, and using Nasdaq or SOX to hedge left-tail risk. For AI storage, shift to a 'long leaders plus high beta hedge' trading style.\n\n2. Current AI Cycle Positioning\n\n2.1 Cycle Positioning\n\nPhase | Dot-com Bubble Features | Current AI Status | Judgment\n---|---|---|---\n1996-1998 | Early diffusion of infrastructure and narrative | Already passed this stage | No match\n1999 | Simultaneous acceleration of narrative and valuation | AI capex, semiconductors, storage, power chain diffusion | High match\nPost-June 1999 | First rate hike, bubble not over | Market repricing hike risk, AI fundamentals strong | High match\nLate 1999 | Topping while hiking | Current closest match | Main Conclusion\nQ1 2000 | Near valuation top, fundamentals not yet deteriorated | Some high beta stocks show topping features | Secondary Conclusion\nLate 2000 | Orders and capex begin to deteriorate | Not yet confirmed by public data | Not yet confirmed\n2001-2002 | Credit risk and bankruptcy wave | Early risk only in 2nd/3rd tier financing platforms | No match\n\nCurrent closest: Second half of 1999. Reason: Hawkish policy expectations, accelerating capex, strong leader earnings, increased volatility in 2nd/3rd tier stocks, and diffusion of storage/hardware price hike chains.\n\nCurrently showing Q1 2000 top window features. Reason: Extremely high capex intensity, massive AI storage price increases, worsening sensitivity of some stocks to rates and good news, and significant single-day semiconductor drops. Reports show a sharp decline in the semiconductor sector on June 5, with SOX down about 10.3% and Micron, AMD, Broadcom, and NVIDIA under pressure. (reuters.com)\n\n3. Dot-com Bubble Historical Timeline Reconstruction\n\n3.1 Nasdaq Timeline\n\nTime | Event | Data/Metric | Trading Meaning\n---|---|---|---\nEarly 1995 | Nasdaq ~751 | Historical price sequence | Bubble starting point\nMarch 10, 2000 | Nasdaq Composite closed at 5048.62 | Historical peak | Valuation top\nOctober 2002 | Nasdaq fell to ~1114-1140 range | ~78% drawdown from high | Fundamental/credit clearing\nPost-2003 | Surviving companies re-expand | Amazon, Google, Microsoft enter next compounding round | Low-price asset transfer phase\n\nNasdaq rose sharply from 1995 to 2000, peaked in March 2000, and then retraced about 78% by October 2002. (fred.stlouisfed.org)\n\n3.2 Fed Policy Sequence\n\nDate | Action | Target Rate | Description\n---|---|---|---\nSept 29, 1998 | Cut 25bp | 5.25% | Easing after Russia crisis, LTCM, financial stress\nOct 15, 1998 | Cut 25bp | 5.00% | Fed cited cautious lenders and unstable financial conditions\nNov 17, 1998 | Cut 25bp | 4.75% | Fed cited 75bp cumulative cuts since Sept; stress remains\nJune 30, 1999 | Hike 25bp | 5.00% | Fed cited 1998 stress eased; easing no longer fully necessary\nAug 24, 1999 | Hike 25bp | 5.25% | Continued withdrawal of easing\nNov 16, 1999 | Hike 25bp | 5.50% | Continued tightening\nFeb 2, 2000 | Hike 25bp | 5.75% | Continued tightening\nMarch 21, 2000 | Hike 25bp | 6.00% | Continued hiking 11 days after Nasdaq top\nMay 16, 2000 | Hike 50bp | 6.50% | Final hike, ~67 days after Nasdaq top\nJan 3, 2001 | Cut 50bp | 6.00% | Fundamentals deterioration begins to dominate\n\n3.3 Relationship Between First Hike, Top, Final Hike, and First Cut\n\nNode | Date | Relationship to Nasdaq Top\n---|---|---\nFirst Hike (1999) | June 30, 1999 | ~254 days before Nasdaq top\nNasdaq Top | March 10, 2000 | Baseline\nFinal Hike (2000) | May 16, 2000 | ~67 days after Nasdaq top\nFirst Cut (2001) | Jan 3, 2001 | ~299 days after Nasdaq top\n\nTrading Conclusions: 1. Do not equate the first hike with an immediate top. 2. Do not wait for the final hike to sell. 3. Do not treat the first cut as a bottom-fishing signal; Nasdaq continued to fall as fundamentals worsened.\n\n3.4 Bubble Bursting Order\n\nOrder | Variable | 1999-2003 Performance | Current AI Corresponding Variable\n---|---|---|---\n1 | High valuation, no profit stocks | Fell first | Unprofitable AI apps, private AI platforms\n2 | High beta hardware/networking | Order expectations compressed first | 2nd/3rd tier AI servers, networking, optical, storage\n3 | Earnings leaders | Valuation fell first, orders verified later | NVIDIA, Broadcom, Arista, Dell, Meta, Microsoft\n4 | Telecom/Fiber financing chain | Debt and financing windows worsened | CoreWeave, Nebius, GPU leasing, data center project finance\n5 | Capital Expenditure | Telecom capex revised down | Hyperscaler AI capex revised down\n6 | Credit Risk | WorldCom, Global Crossing exposed | Private credit, leasing, project finance, convertible bond risk\n7 | Survivor asset acquisition | Low-price transfer of fiber/data center assets | Low-price transfer of AI data centers, power, GPU, storage assets\n\nWorldCom and Global Crossing collapsed under debt and capex pressure in 2002. (sec.gov)\n\n3.5 Cisco, Lucent, Nortel, Fiber, and Bandwidth\n\nCompany/Segment | Role in Bubble | Key Issue\n---|---|---\nCisco | Earnings network leader | Real earnings, but valuation and order cycle still pierced\nLucent | Telecom equipment supplier | Receivables, inventory, supplier/customer financing risk amplified\nNortel | Telecom and optical networking | Orders amplified by telecom capex, then sharply revised down\nGlobal Crossing, WorldCom | Fiber and long-haul assets | Debt-driven, bandwidth prices collapsed\nJDS Uniphase, Corning | Optical components/fiber | Price and demand reversed after overheating\nEMC, Sun, Brocade | Enterprise storage/servers | Real demand, but amplified by internet capex and IT cycles\n\nCisco's backlog fell from ~$3.4 billion in Sept 2000 to ~$2.03 billion in Sept 2001. (newsroom.cisco.com)\n\n4. Current AI Cycle Timeline Reconstruction\n\n4.1 Narrative Launch\n\nTime | Event | Impact\n---|---|---\nNov 30, 2022 | OpenAI releases ChatGPT | Generative AI shifts from research to consumer/developer product\n2023 | Cloud providers expand AI infra investment | GPU, HBM, server, network, and data center chain starts\n2024-2025 | AI capex diffuses from GPU to power, cooling, optical, storage | Infrastructure rally diffuses\n2026 | AI capex enters high-intensity phase, rates turn hawkish | Trend remains strong, but top window risk rises\n\n4.2 Current AI Infrastructure Data Table\n\nCompany | Latest Public Data | Trading Meaning\n---|---|---\nNVIDIA | Q1 FY2027 revenue $81.6B, +85% YoY, GM ~75% | AI compute demand remains strong; leader is core variable\nAMD | Q1 2026 revenue $10.3B, Data Center $5.8B, +57% YoY | No. 2 GPU and CPU data center beneficiary\nBroadcom | Q2 FY2026 AI semiconductor revenue $10.8B, +143% YoY | Strong custom ASIC and network-related AI demand\nArista | Q1 2026 revenue $2.709B, +35.1% YoY | AI networking still in uptrend\nDell | FY2026 Q4 AI optimized server revenue $9B, +342% YoY | AI servers strong; traditional storage weaker\nSuper Micro | Q3 FY2026 revenue $10.2B, GM ~9.9% | High growth but high working capital risk\nOracle | FY2026 Q3 guidance: FY2026 capex $50B | Data center financing and fulfillment pressure needs monitoring\nCoreWeave | Q1 2026 total debt ~$24.86B | Asset sensitive to rates and credit spreads\n\n4.3 Hyperscaler Capex Pressure Table\n\nCompany | Latest Capex and Cash Flow | Capex Pressure\n---|---|---\nMicrosoft | Q3 FY2026 PPE +$30.876B, OCF $46.679B, FCF $15.8B | Capex is ~66% of OCF; AI investment pressing FCF\nAlphabet | Q1 2026 PPE purchases $35.674B, OCF $45.790B | Capex is ~78% of OCF; cloud growth strong but heavy investment\nAmazon | Q1 2026 PPE purchases $44.203B, OCF $26.032B | Quarterly capex exceeds OCF; FCF pushed very low\nMeta | Q1 2026 capex $19.84B, OCF $32.23B, FCF $12.39B | Capex supported by cash flow, but intensity is extremely high\n\n5. AI Storage Mainline\n\n5.1 Essence of AI Storage\n\nLayer | Essence | AI Demand Source | Structural Alpha?\n---|---|---|---\nHBM | High bandwidth memory for GPU | Training/inference accelerator binding | High\nServer DRAM | Data active memory for CPU/GPU clusters | AI servers, high-capacity DIMM, CXL | Medium-High\nNAND | Non-volatile flash bit supply | Enterprise SSD, AI server local storage | Medium\nEnterprise SSD | Low latency, high throughput hot data | Training data reads, checkpoints, inference cache | Medium-High\nNearline HDD | Low-cost capacity storage | Cold data, object storage, data lakes | Medium\nStorage Systems | Enterprise reliability/management | All-flash arrays, parallel file systems | Medium-High\nCloud Storage | Hyperscaler internal/external services | Objects, blocks, files, data lakes | High (not split out)\nData Infra Software | Governance, vector DB, RAG | Enterprise knowledge base, embedding | High (valuation sensitive)\n\n5.2 Who Pays, Who Benefits, Who Suffers\n\n- Who Pays: Hyperscalers, model companies, GPU clouds, enterprise customers, governments.\n- Who Benefits: HBM suppliers, DRAM/NAND suppliers, enterprise SSD, HDD, storage systems, cloud storage, data platform software.\n- Who Suffers: AI apps unable to raise prices, low-margin model providers, unprofitable AI companies dependent on low-cost inference/storage, weak GPU cloud platforms.\n\n5.3 Where AI Storage Price Hikes Come From\n\n- Real AI Demand: Capacity shifting to HBM/servers.\n- Traditional Server Recovery: Server DRAM and enterprise SSD strengthening simultaneously.\n- Mobile/PC Inventory Repair: Client SSD and mobile NAND also benefit.\n- Supplier Discipline: Supply contraction after the cycle bottom.\n- HBM Crowding Out DRAM: HBM demand squeezes regular DRAM supply.\n- Customer Pre-purchasing: Occurs during rapid price rises; a top risk signal.\n\n5.4 AI Storage Company Data Verification\n\n- Micron: Q2 FY2026 revenue $23.86B, GM 74.4%. Strong earnings but receivables rising.\n- SK hynix: Q1 2026 revenue 52.58T KRW, operating margin 72%. Structural HBM benefit.\n- Samsung: Q1 2026 Memory Business record sales driven by AI and price hikes.\n- Seagate: FY2026 Q3 GM 46.5%, FCF $953M. Strong HDD capacity cycle.\n- SanDisk: Q3 FY2026 Datacenter revenue +233%. High beta NAND/SSD play.\n\n6. Fed Policy, Real Rates, and Liquidity\n\n6.1 Current Macro State\n\n- Fed Target Range: 3.50% to 3.75% (April 2026). Policy remains restrictive.\n- 2-Year Treasury: 4.05%. Sensitive to hike expectations.\n- 10-Year TIPS Real Rate: 2.11%. Valuation pressure on high-duration assets.\n- HY OAS: 2.74%. No systemic credit pressure yet.\n- CCC OAS: 9.46%. Low-rating tightening, but not out of control.\n- ON RRP: $761M. Buffer essentially exhausted.\n\n6.2 Shift to Rate Hike Expectations?\n\nYes, the market has begun repricing hike risks, but has not entered the 'tightening plus credit deterioration' peak risk state. 2-year yields and real rates are constraining high-duration assets, but credit spreads (HY OAS) do not yet show the clearing seen in 2001-2002.\n\n7. Similarities to Dot-com Bubble\n\n- Strong tech narrative and accelerating infra investment.\n- Leaders' earnings stronger than 2nd/3rd tier stocks.\n- High market concentration; index reliance on few stocks.\n- Hawkish policy expectations lowering valuation ceilings.\n- Infra chain diffusing from core compute to networking, power, and storage.\n- Questioning of capex ROI (shifting from growth to 'return per dollar').\n\n8. Differences from Dot-com Bubble\n\n- Leader Profitability: NVIDIA, Microsoft, etc., have stronger cash flows than 1999 firms.\n- Capex Funding: Hyperscalers rely more on internal cash flow; 2nd tier platforms rely on debt.\n- Asset Depreciation: GPU depreciation and tech iteration are faster than fiber.\n- Demand Verification: AI/cloud revenue and token usage provide partial verification.\n\n9. Cycle Scoring Model (0-5, 5=High Risk)\n\n- Valuation Bubble: 4.0 (Leaders high, 2nd tier volatile)\n- Capex Overheating: 4.5 (Hyperscaler capex/OCF ratio extremely high)\n- Financing Vulnerability: 3.5 (High debt/lease burdens for platforms like CoreWeave)\n- Real Demand Realization: 1.5 (Strong revenue for NVIDIA/AMD; low risk score)\n- Supply Excess Risk: 3.0 (HBM tight, but NAND/SSD/HDD expansion risk)\n- Market Width Deterioration: 3.5 (High concentration, sharp semi drops)\n- Fed/Liquidity Headwinds: 3.0 (Hike expectations rising, high real rates)\n\nOverall Judgment: Late stage, ~3.5 to 4.0. Positioned like the second half of 1999 with Q1 2000 top window features.\n\n10. Three Future Scenarios\n\n10.1 Scenario 1: Bubble Continues (40% Probability)\nAI revenue continues high growth; capex revised up; credit stable. Action: Keep core longs, avoid chasing 2nd tier, use options for protection.\n\n10.2 Scenario 2: Topping Volatility (35% Probability)\nLeader revenue strong but valuation expansion stops; real rates rise; good news doesn't move stocks. Action: Reduce net exposure, increase pairs trading, hedge with NDX/SOX put spreads.\n\n10.3 Scenario 3: Bubble Bursts (25% Probability)\nHyperscalers cut capex; GPU orders canceled; credit spreads widen; financing fails. Action: Shift to cash/short-term bonds, short 2nd tier hardware and unprofitable AI.\n\n11. Trading Action Framework\n\n- Core Longs: Keep, but concentrate on cash-flow leaders with verifiable orders.\n- AI Hardware: Reduce 2nd/3rd tier beta.\n- AI Storage: Shift from single-sided theme to leader-long/beta-short.\n- AI Software: Screen for companies converting AI to paid revenue/NRR.\n- Index Hedging: Use Nasdaq or SOX protection.\n- Cash Ratio: Increase to handle 15-25% drawdowns.\n\n12. 3-6 Month Watchlist\n\n- AI Fundamentals: Hyperscaler capex guidance, NVIDIA data center revenue/margins, GPU leasing prices.\n- AI Storage: HBM prices/orders, DRAM/NAND contract prices, storage company inventory/receivables.\n- Fed/Liquidity: FedWatch hike probabilities, 2Y/10Y yields, HY/CCC OAS spreads, ON RRP levels.\n\n13. Fed Policy Response Matrix\n\n- Strong AI Revenue + Rising Hike Expectations: 1999 H2 risk; keep leaders, reduce high-duration/financing-dependent assets, hedge with NDX.\n- Weak AI Revenue + Fed Rate Cuts: Recessionary cut risk; do not bottom-fish on the first cut; wait for order/credit/earnings repair.\n- Strong AI Revenue + Widening Credit Spreads: Infra financing chain risk; avoid high-leverage data centers/leasing platforms.\n\n14. AI Storage Trading Matrix\n\n- Capex Up + Price Up + Healthy Inventory: Mainline continues; hold leaders.\n- Capex Up + Price Up + Rising Inventory: Pre-purchasing risk; reduce high beta.\n- Capex Slowing + Price Falling: Cycle reversal; exit storage chain, avoid high-inventory/high-valuation firms.\n\nFinal Direct Answers:\n- Should you wait for the Fed to actually hike to reduce positions? No, the market trades on expectations.\n- Should you wait for the final hike to short? No, the 2000 top preceded the final hike.\n- Should you wait for the first cut to bottom-fish? No, Nasdaq continued to fall after the 2001 cut.\n- Current optimal action: Reduce positions, add hedges, and rotate; do not just short everything blindly.\n- Which assets are like high-duration bonds? Unprofitable AI, high-valuation data platforms, high P/S software.\n- Which assets are like short-duration cash flow? Microsoft, Meta, Amazon, Alphabet, Broadcom.
Comparison of the 2000 Dot-com Bubble and the 2026 AI Investment Cycle
- Views
- 185K
- Likes
- 382
- Reposts
- 76
- Comments
- 50
- Bookmarks
- 568
TL;DR
This analysis maps the AI investment cycle against the 2000 dot-com bubble, identifying current trends as late-stage acceleration. It provides data-driven insights on AI storage, hyperscaler capex, and strategic trading frameworks.
Reading the ENGLISH translation
