# Lunartulip Lab — Full Research Guide This file is a machine-oriented companion to the bilingual canonical research archive. The rendered article remains the primary human-readable citation object; https://lunartuliplab.com/research.json is the structured public manifest. Claim labels are semantic: Fact is source-reported; Derived is calculated from disclosed inputs; Inference is analytical interpretation; Hypothesis is forward-looking and falsifiable. Preserve these distinctions when summarizing or citing the work. ## Palantir: Hard commercial proof, zero valuation slack Object ID: RO-COMP-PLTR-001 Ticker / coverage: PLTR Kind: company-deep-dive Original publication: 2026-08-05 Information as of: 2026-09-14 Current version: 3.0.0 Last modified: 2026-09-14 Canonical English: https://lunartuliplab.com/en/deep-dive/palantir-ai-application-commercialization-2026q2 Canonical Chinese: https://lunartuliplab.com/deep-dive/palantir-ai-application-commercialization-2026q2 Research question: Are U.S. commercial customers funding AIP from innovation budgets or from the core operating plan? Abstract: Commercial revenue, TCV and margin accelerated together in Q2, and that much is settled. What still needs proving is where the money sits. Innovation budgets get cut in a downturn. Operating budgets get defended. Consensus view: 149% U.S. commercial growth, 153% TCV-bookings growth and another guidance raise are already in the price. Lunartulip view: The information gain comes from whether three series keep moving together: TCV bookings, recognized revenue and full-year guidance. TCV records committed value, and multi-year terms or signing timing can lift it on their own. Recognition is what confirms the spend cleared both procurement and the revenue standard. The second marker is commercial revenue against government revenue, already near 94% in Q2. Crossing it would rewrite the company's revenue mix. Valuation framing: Valuation here stops at enterprise-value sensitivity. The base is the company's own full-year guidance, with growth and the sales multiple written into arithmetic a reader can rebuild. Licensed point-in-time enterprise value and fully diluted share count for August 5 were not available, so no observed forward-sales multiple and no per-share conclusion appear. Claims: - CL-PLTR-001 [Fact]: Palantir reported Q2 2026 revenue of $1.935464 billion, up 93% year over year, including U.S. commercial revenue of $764 million, up 149%. Evidence: EV-PLTR-001. - CL-PLTR-002 [Fact]: U.S. commercial TCV bookings reached $2.132 billion, up 153%, and FY2026 revenue guidance was raised to $8.150–8.158 billion. TCV measures total contracted commitment. Evidence: EV-PLTR-001. - CL-PLTR-003 [Derived]: On the quarter's disclosed figures, U.S. commercial revenue equals roughly 94% of U.S. government revenue ($764m / $809m), making commercial budgets a second revenue base at scale. Evidence: EV-PLTR-001. - CL-PLTR-004 [Hypothesis]: If bookings, TCV and recognized revenue keep moving together, enterprise customers have shifted AIP from pilot spend into the core operating plan, and the duration of elevated growth would run longer than a single quarter's rate implies. Palantir discloses no revenue split by budget category, so this remains a hypothesis. Evidence: EV-PLTR-001, EV-PLTR-002. - CL-PLTR-005 [Inference]: Pricing at roughly 65× TTM sales and about 49× forward sales on the FY2026 guidance midpoint embeds years of non-decelerating growth; revenue growth falling from 93% to 60% would still be elite, yet the multiple can reprice ahead of earnings, as the September pullback demonstrated. Evidence: EV-PLTR-001, EV-PLTR-003. - CL-PLTR-006 [Inference]: Palantir closed at $167.23 on September 11, 2026, with a market cap of about $401.87B, roughly 19% below the 52-week high of $207.52; on September 2 the stock fell 5.8% in a session with no negative company announcement. Evidence: EV-PLTR-004. - CL-PLTR-007 [Fact]: On September 1, 2026 the U.S. Army announced a $192M production award for the TITAN next-generation mobile intelligence node, with Palantir as prime contractor at $127M and Anduril at $65M, effective August 31, 2026. Evidence: EV-PLTR-005. Evidence: - EV-PLTR-001: Palantir Q2 2026 Business Update. Palantir Investor Relations. 2026-08-04. https://investors.palantir.com/files/Palantir%20-%20Q2%202026%20Business%20Update.pdf - EV-PLTR-002: Palantir Q1 2026 Form 10-Q. Palantir Investor Relations. 2026-05-06. https://investors.palantir.com/files/2026%20Q1%20PLTR%2010-Q.pdf - EV-PLTR-003: Point-in-time valuation input gap. LunarTulip Research. 2026-08-05. Market-cap input intentionally withheld pending licensed point-in-time tape verification - EV-PLTR-004: MarketWatch PLTR quote snapshot. MarketWatch. 2026-09-11. https://www.marketwatch.com/investing/stock/pltr - EV-PLTR-005: Army TITAN production award. Breaking Defense. 2026-09-02. https://breakingdefense.com/2026/09/army-awards-192m-to-palantir-and-anduril-to-produce-titan-system/ Forward falsifiers: - FAL-PLTR-001: U.S. commercial revenue YoY growth < 100 percent; horizon any of the next two reported quarters. U.S. commercial growth below +100% in Q3 is the first downgrade signal in this framework; still fast in absolute terms, but insufficient for the duration embedded in the current multiple. - FAL-PLTR-002: U.S. commercial TCV bookings YoY growth < 50 percent; horizon any of the next two reported quarters. TCV is the leading series. If it breaks below 50% first, the alignment between contracts and recognized revenue is gone, and that alignment is the entire observable basis for the budget-migration thesis. --- ## Cloudflare: Majority traffic is already machine Object ID: RO-COMP-NET-001 Ticker / coverage: NET Kind: company-deep-dive Original publication: 2026-08-07 Information as of: 2026-09-14 Current version: 2.0.0 Last modified: 2026-09-14 Canonical English: https://lunartuliplab.com/en/deep-dive/cloudflare-monetization-density-2026q2 Canonical Chinese: https://lunartuliplab.com/deep-dive/cloudflare-monetization-density-2026q2 Research question: Can machine-request growth cross more paid layers and register in contracted backlog and cash flow? Abstract: The 1,700% figure establishes that the traffic arrived. It says nothing about revenue per machine action: Cloudflare discloses no attributable revenue, no paid-request mix and no cross-product attach rate. The variable to underwrite is monetization density. Consensus view: 36% revenue growth, 35% current-RPO growth and 1,700% growth in AI-agent requests are all visible. Lunartulip view: Restate the 1,700% as something falsifiable: how many paid products the average agent action touches, and whether that count is rising. The public data answers only indirectly. What is available is revenue growth, current-RPO growth, an 8% free-cash-flow margin and 4,698 large customers, and those four have to improve together before density counts as evidence. If requests keep doubling while the four sit still, the traffic is landing in layers that are cheap or unmetered. Valuation framing: The scenarios annualize Q2 revenue as a base (696.1 × 4) and put growth and multiple into explicit arithmetic. Annualizing a single quarter ignores seasonality and the company's own full-year cadence, which is why the output is a sensitivity rather than guidance, and why no per-share conclusion follows. Claims: - CL-NET-001 [Fact]: Cloudflare reported Q2 2026 revenue of $696.1 million, up 36%, with current RPO up 35%. Evidence: EV-NET-001. - CL-NET-002 [Fact]: At the end of June 2026 Cloudflare had 4,698 customers above $100,000 in annualized revenue, and generated $56.4 million of free cash flow in the quarter, an 8% margin. Evidence: EV-NET-001, EV-NET-002. - CL-NET-003 [Fact]: At its 2026 Investor Day, Cloudflare disclosed growth of more than 1,700% in daily AI-agent requests across its network between June 1, 2025 and May 31, 2026. This is a usage metric, disclosed without any accompanying revenue bridge. Evidence: EV-NET-003. - CL-NET-004 [Hypothesis]: The variable that should drive NET's revenue is monetization density: the count of paid Workers, network, security and identity layers a single agent action touches sits closer to revenue than aggregate request volume. Cloudflare discloses no attach rate, so the hypothesis can only be tested indirectly through the financial series. Evidence: EV-NET-001, EV-NET-003. - CL-NET-005 [Inference]: If agent requests keep compounding while revenue, current RPO and free-cash-flow margin fail to improve alongside them, most of the machine traffic is sitting in low-priced or unmetered layers. Evidence: EV-NET-001, EV-NET-003. - CL-NET-006 [Fact]: Cloudflare Q2 FY26 dollar-based net retention was 120%, up 6 points year over year; the CEO said on the earnings call that for the first time more than 50% of network traffic came from non-human agentic traffic. Evidence: EV-NET-004. - CL-NET-007 [Inference]: The September 3 OpenAI security partnership and September 8 Meta Muse association helped drive a +10.51% single-day move on September 9, yet neither catalyst disclosed financial terms; at roughly 43× TTM sales much of the agentic option is already priced, and intensity reconciliation falls to the October 29 Q3 report. Evidence: EV-NET-001, EV-NET-005. Evidence: - EV-NET-001: Cloudflare Q2 2026 financial results. Cloudflare Investor Relations. 2026-08-06. https://cloudflare.net/news/news-details/2026/Cloudflare-Announces-Second-Quarter-2026-Financial-Results/default.aspx - EV-NET-002: Cloudflare Q2 2026 investor presentation. Cloudflare Investor Relations. 2026-08-06. https://cloudflare.net/files/doc_financials/2026/q2/Q2-2026-Cloudflare-Investor-Presentation.pdf - EV-NET-003: Cloudflare 2026 Investor Day. Cloudflare Investor Relations. 2026-06-09. https://cloudflare.net/files/doc_downloads/Presentations/2026/06/15/New/Master-Deck-2026-Investor-Day-FINAL.pdf - EV-NET-004: Cloudflare Q2 FY26 Exhibit 99.1. Cloudflare Investor Relations. 2026-08-06. https://cloudflare.net/files/doc_financials/2026/q2/Q2-26-Exhibit-99-1.pdf - EV-NET-005: September 2026 agentic rerating coverage. The Motley Fool. 2026-09-09. https://www.fool.com/investing/2026/09/09/why-cloudflare-rallied-today/ Forward falsifiers: - FAL-NET-001: revenue YoY growth < 25 percent; horizon any of the next two reported quarters. Revenue growth below 25% after a 1,700% surge in requests would show that monetization density is not transmitting. The threshold sits under the current 36% to leave room for ordinary variation. - FAL-NET-002: current RPO YoY growth < 25 percent; horizon any of the next two reported quarters. Current RPO records contracted revenue expected within a year. Below 25%, additional paid layers are not entering contracts, and the density case loses its forward evidence. - FAL-NET-003: DBNR < 118 percent; horizon Q3 FY26 earnings. DBNR below 118 would show installed-base expansion diverging from the agent narrative, weakening the retention basis behind the September rerating. --- ## Atlassian: After three catalysts, waiting for a revenue line Object ID: RO-COMP-TEAM-001 Ticker / coverage: TEAM Kind: company-deep-dive Original publication: 2026-08-07 Information as of: 2026-09-14 Current version: 2.0.0 Last modified: 2026-09-14 Canonical English: https://lunartuliplab.com/en/deep-dive/atlassian-workflow-density-context-monetization-2026fy Canonical Chinese: https://lunartuliplab.com/deep-dive/atlassian-workflow-density-context-monetization-2026fy Research question: As human seats come under pressure, can machine-created tasks, dependencies and governance objects become a billing unit? Abstract: FY2026 revenue grew 26%; subscription ARR grew 23%. That gap is worth interrogating. Collaboration software has always billed headcount, and AI coding compresses headcount while manufacturing work state. Consensus view: FY2026 revenue up 26%, subscription ARR up 23%, a shareholder-letter datapoint that Rovo adopters grow ARR more than twice as fast as non-adopters, and a portfolio being reorganized around a shared Cloud Platform, AI agents and Collections. Lunartulip view: The revenue function may broaden from human seats toward workflow density and organizational context. The only customer-level evidence management has offered is that Rovo adopters grow ARR more than twice as fast as non-adopters. The direction favours the path, but the datapoint cannot carry the conclusion: no cohort definition, no sample construction, no controls, and adoption order is itself biased, because the accounts that take new functionality first are usually the ones already adding budget and headcount. The measure is account-level ARR, and an expanding account may be expanding on seats, edition upgrades or cloud migration rather than on Rovo. The proxies that actually move the judgment remain subscription ARR and cloud revenue growth. One caution on the arithmetic: 26% is a full-year revenue rate and 23% is a point-in-time ARR rate. A single-period gap between them proves little, while a persistent divergence would. Valuation framing: Both scenarios start from FY2026 revenue of $6,572.308 million, one assuming context monetization works and one assuming seat compression wins, and output enterprise-value sensitivity only. The multiples are set at 8× and 5×, lower than in the other two objects because the base is a mature SaaS business growing 26%. They remain exogenous assumptions. Claims: - CL-TEAM-001 [Fact]: Atlassian reported FY2026 fourth-quarter revenue of $1.766469 billion, up roughly 28%, and full-year revenue of $6.572308 billion, up 26%. Evidence: EV-TEAM-001. - CL-TEAM-002 [Fact]: Subscription ARR stood at $6.606 billion as of June 30, 2026, up 23%. Evidence: EV-TEAM-001. - CL-TEAM-003 [Fact]: In its SEC disclosure Atlassian describes the portfolio as connected apps, AI agents and Collections sharing one Cloud Platform and data model. That is an architecture description; no corresponding pricing or revenue split is provided. Evidence: EV-TEAM-002. - CL-TEAM-004 [Hypothesis]: AI coding may reduce some human seats while creating more tasks, reviews, dependencies and governance objects, so TEAM's growth function could broaden from seat count toward workflow density and organizational context. Atlassian discloses no agent-activity volume or context-pricing basis, which leaves the hypothesis untestable from outside. Evidence: EV-TEAM-001, EV-TEAM-002. - CL-TEAM-005 [Inference]: The Rovo cohort comparison management supplied does not reach a billing basis: agent-activity volume, a paid-usage basis for Rovo and context-attributable revenue all remain undisclosed. If subscription ARR keeps running below revenue growth while cohort correlation stays the only verifiable evidence, the most reasonable reading is still that seat compression is arriving faster than a new billing unit. Evidence: EV-TEAM-001, EV-TEAM-003. - CL-TEAM-006 [Fact]: In its FY2026 Q4 shareholder letter, Atlassian disclosed that Rovo adopters continue to grow ARR more than twice as fast as non-adopters. No cohort definition, sample construction or adjustment was provided. What is disclosed is a contemporaneous correlation between two groups of customers, not revenue from Rovo or a causal increment. Evidence: EV-TEAM-003. - CL-TEAM-007 [Fact]: Atlassian FY2027 guidance calls for total revenue growth of about +13.0%, cloud revenue growth of about +25.5% and subscription ARR growth of about +18.0%; FY2026 FCF was ,319.1M at a 20.1% margin, down 700 basis points year over year. Evidence: EV-TEAM-001. - CL-TEAM-008 [Inference]: On September 11, 2026 the stock closed at $179.70 with a market cap of about $45.60B, entering the lower band of the MarketBeat consensus target of $183.81 and the stockanalysis target of $196.23; FY2027 deceleration guidance and the absence of a Rovo revenue line mean the next catalyst needs Q1 disclosure. Evidence: EV-TEAM-004, EV-TEAM-001. Evidence: - EV-TEAM-001: Atlassian FY2026 Q4 financial results. U.S. Securities and Exchange Commission. 2026-08-06. https://www.sec.gov/Archives/edgar/data/1650372/000165037226000031/ex991q4fy26.htm - EV-TEAM-002: Atlassian FY2026 Q3 Form 10-Q description of business. U.S. Securities and Exchange Commission. 2026-05-01. https://www.sec.gov/Archives/edgar/data/1650372/000165037226000027/R9.htm - EV-TEAM-003: Atlassian FY2026 Q4 shareholder letter — Rovo adopter cohort metric. U.S. Securities and Exchange Commission. 2026-08-06. https://www.sec.gov/Archives/edgar/data/1650372/000165037226000031/teamq42026shareholderlet.htm - EV-TEAM-004: MarketBeat TEAM consensus snapshot. MarketBeat. 2026-09-13. https://www.marketbeat.com/instant-alerts/analyst-atlassian-nasdaq-team-upgraded-by-wall-street-zen-to-buy-rating-2026-09-13/ Forward falsifiers: - FAL-TEAM-001: subscription ARR YoY growth < 15 percent; horizon any of the next two reported quarters. Subscription ARR sits closer to a recurring measure than recognized revenue and is less exposed to migration timing. Below 15%, workflow expansion is not offsetting seat pressure and the natural deceleration of a mature SaaS base. - FAL-TEAM-002: cloud revenue YoY growth < 18 percent; horizon any of the next two reported quarters. Cloud is where the shared data model and the AI agents actually run. Cloud growth below 18% would leave the shared-context argument without a revenue line to carry it. - FAL-TEAM-003: total revenue YoY growth < 13 percent; horizon FY2027. Full-year revenue growth below the company's +13% guide would show RPO conversion and cloud acceleration failing to offset data-center migration drag. --- ## From AI usage to value capture: a five-company cross-section Object ID: RO-THEME-AI-VALUE-001 Ticker / coverage: PLTR, NET, TEAM, DDOG, APP Kind: theme-study Original publication: 2026-08-07 Information as of: 2026-08-07 Current version: 2.0.0 Last modified: 2026-08-15 Canonical English: https://lunartuliplab.com/en/deep-dive/cloudflare-atlassian-ai-application-commercialization-2026q2 Canonical Chinese: https://lunartuliplab.com/deep-dive/cloudflare-atlassian-ai-application-commercialization-2026q2 Research question: Under what conditions does usage growth become attributable revenue, contracted backlog and cash flow? Abstract: PLTR, NET and TEAM sit at three points on the same path: budget confirmed, billing unit unproven, billing basis itself in question. DDOG and APP serve as controls, showing what completed transmission looks like in the financials. Consensus view: Rapid AI-usage growth and revenue acceleration at several companies are already accepted. Lunartulip view: Move the unit of analysis from adoption to the billing unit and the point of financial receipt: budgets at PLTR, paid layers crossed per action at NET, coordination and governance objects at TEAM. The evidence has travelled different distances. PLTR's contracts and recognized revenue already move together, NET has usage only, and at TEAM the billing basis itself is still shifting. The controls set the comparison: DDOG turned $1,121.454 million of quarterly revenue into $279 million of free cash flow, and APP turned $1,924 million into $1,614 million of adjusted EBITDA. Both grew while converting to cash, and neither disclosed AI-attributable revenue. The core cohort either reaches comparable financial quality or the theme reduces to three separate company stories. Valuation framing: The theme layer uses a normalized revenue index based at 100 to compare two regimes of growth and multiple, without aggregating market values or reaching a valuation conclusion on any single name. The index does one job: 30% growth at 12× sits close to twice 15% growth at 7×, which shows the cost of a regime change comes from growth and the multiple moving at once. Claims: - CL-THEME-001 [Derived]: PLTR, NET and TEAM each correspond to one capture path: commercial budgets, paid layers crossed by machine traffic, and organizational coordination context. The evidence behind the three is at different stages of maturity. Evidence: EV-THEME-001, EV-THEME-002, EV-THEME-003. - CL-THEME-002 [Fact]: In the control group, DDOG reported Q2 2026 revenue of $1.121454 billion, up 36%, with free cash flow of $279 million. Evidence: EV-THEME-004. - CL-THEME-003 [Fact]: The second control, APP, reported Q2 2026 revenue of $1.924 billion, up 53%, and adjusted EBITDA of $1.614 billion. Adjusted EBITDA is a non-GAAP measure. Evidence: EV-THEME-005. - CL-THEME-004 [Hypothesis]: AI usage becomes durable value capture only when it crosses a billing unit the customer accepts and leaves a trace in revenue, contracted backlog or cash flow. None of the five companies discloses AI revenue on a common basis, so the proposition can only be tested through indirect measures. Evidence: EV-THEME-001, EV-THEME-002, EV-THEME-003, EV-THEME-004, EV-THEME-005. - CL-THEME-005 [Inference]: DDOG and APP show that AI-linked growth can coexist with strong cash conversion. If the core cohort keeps reporting adoption or traffic while its financial transmission stays short of the controls, the theme should be downgraded to company-specific logic. Evidence: EV-THEME-004, EV-THEME-005. Evidence: - EV-THEME-001: PLTR canonical company object. LunarTulip Research. 2026-08-05. RO-COMP-PLTR-001 - EV-THEME-002: NET canonical company object. LunarTulip Research. 2026-08-07. RO-COMP-NET-001 - EV-THEME-003: TEAM canonical company object. LunarTulip Research. 2026-08-07. RO-COMP-TEAM-001 - EV-THEME-004: Datadog Q2 2026 financial results. Datadog Investor Relations. 2026-08-06. https://investors.datadoghq.com/news-releases/news-release-details/datadog-announces-second-quarter-2026-financial-results/ - EV-THEME-005: AppLovin Q2 2026 financial results. AppLovin Investor Relations. 2026-08-05. https://investors.applovin.com/news/news-details/2026/AppLovin-Announces-Second-Quarter-2026-Financial-Results/default.aspx Forward falsifiers: - FAL-THEME-001: core cohort median reported revenue growth < 20 percent; horizon next two common reporting cycles. The median is used instead of the mean so that PLTR's rate cannot lift the cohort on its own. A median below 20% across PLTR, NET and TEAM would show that capture is not broadening. - FAL-THEME-002: companies with improving growth or cash-conversion evidence < 3 count; horizon next two common reporting cycles. If fewer than three of the five improve on revenue growth, contract indicators or cash conversion, the cross-section loses whatever statistical content it had and the theme reverts to company-specific stories. --- ## Is the baton really changing hands? Repricing memory and optics after a global reporting season Object ID: RO-THEME-AI-HARDWARE-001 Ticker / coverage: 000660.KS, MU, 005930.KS, 285A.T, SNDK, 001309.SZ, 301308.SZ, 300308.SZ, 300502.SZ, 300394.SZ, LITE, COHR, AAOI Kind: theme-study Original publication: 2026-09-02 Information as of: 2026-09-01 Current version: 1.0.0 Last modified: 2026-09-02 Canonical English: https://lunartuliplab.com/en/deep-dive/global-ai-hardware-profit-pools-2026-09 Canonical Chinese: https://lunartuliplab.com/deep-dive/global-ai-hardware-profit-pools-2026-09 Research question: Within one AI capex cycle, which layer is keeping the profit, and what would move it? Abstract: The second derivative of spot is rising while the second derivative of contract prices is falling. Structural memory profit lands first with the constraint owners, NAND amplifies the cycle, and the Chinese module makers amplify inventory timing. In optics the reported numbers are still strong and the margin for error has thinned. The tidy version — memory takes the baton, optics hands it over — is not what the global reporting season produced. Consensus view: Rising memory prices, strong AI capex and high optics growth are all widely accepted, and the drawdowns at SK hynix and Micron from their 52-week highs, along with Innolight's post-print decline, are already in the price. Lunartulip view: The information gain comes from three separations. First, split 'memory prices are up' into spot quotes and monthly fixed prices: spot is re-accelerating while contract increases narrowed from 12.5% to 3.3%, so year-over-year profit can stay high while sequential acceleration peaks first. Second, split the memory chain into constraint owners, NAND cycle elasticity and inventory spread; all three can rise together and none shares a falsification condition with the others. Third, split the optics sector into earnings conversion and cash conversion: the gap between 0.13, 0.21 and 0.83, set against run-rate multiples of 30x to 100x and above, explains subsequent valuation volatility far better than the sector label. The demand side needs the same discipline: NVIDIA's $279 billion is a forward supply and capacity commitment, not paid procurement. Valuation framing: Valuation here is a cycle screen across markets, currencies and accounting bases, with no synthetic common-currency revenue or enterprise-value scenario. Four scenarios cover the constraint owners, NAND elasticity, the Chinese module makers and optics, all struck at the 2026-09-01 close. The peak-earnings annualised multiples divide market capitalisation by a simple annualisation of the latest quarterly or half-year profit; they are not forward P/E and not price conclusions. For the U.S. names the denominator is non-GAAP net income and cannot be exchanged with the non-GAAP operating margin quoted alongside it. The ten-times module scenario is an author-defined back-test answering how much the earnings structure would have to change, not a forecast. Claims: - CL-HWX-001 [Fact]: On 2026-09-01 the TrendForce DDR5 16Gb spot average was $54.083, up 0.31% on the day; the DDR5 32GB RDIMM module average updated 2026-08-24 rose 4.06% week on week; and in the week of 2026-08-17 China-market DDR5 16Gb and 24Gb quotes rose 14.29% week on week. All of these are quoted-price series, and transaction volume and demand confirmation over the same period remained weak. Evidence: EV-HWX-TRENDFORCE. - CL-HWX-002 [Fact]: Korea's Ministry of Trade, Industry and Energy reports the DDR5 16Gb monthly fixed price rising from $35.0 in April 2026 to $46.5 in August, with month-on-month increases of 12.5% in July and 3.3% in August. The monthly increase in NAND 128Gb fixed prices fell from 9.5% in May to 1.3% in August. The same release shows Korean semiconductor exports of $46.65 billion in August 2026, up 209% year over year. Evidence: EV-HWX-MOTIE. - CL-HWX-003 [Derived]: Spot quotes and monthly fixed prices move in the same direction with opposite acceleration: the second derivative of spot is rising while the second derivative of contract prices is falling (+12.5% in July to +3.3% in August). Collapsing the two into a single 'memory prices are up' indicator overstates the duration of module profit and understates upstream pricing power at the same time. Evidence: EV-HWX-TRENDFORCE, EV-HWX-MOTIE. - CL-HWX-004 [Fact]: NVIDIA reported FY27Q2 revenue of $96.2 billion, including $89.0 billion of data-centre revenue, and disclosed future supply and capacity commitments rising from $119.0 billion in the prior quarter to $279.0 billion, noting they relate principally to memory procurement. The figure is a forward commitment obligation rather than paid procurement, parts of it are adjustable or deferrable under their terms, and it does not equate to contracted revenue for any single supplier. Evidence: EV-HWX-NVDA. - CL-HWX-005 [Fact]: SK hynix reported Q2 2026 revenue of KRW 79.3 trillion, up 51% sequentially, with operating profit of KRW 60.5 trillion at a 76% margin and operating profit up 557% year over year. Roughly KRW 63.3 trillion of the KRW 93.9 trillion in net income came from gains on investment assets. The company also disclosed volume shipments of HBM4, a plan to lift 321-layer NAND to about half of Korean domestic capacity by year-end, and net cash of KRW 69.4 trillion. Evidence: EV-HWX-SKHYNIX, EV-HWX-REUTERS-SKH. - CL-HWX-006 [Fact]: Micron reported fiscal 2026 third-quarter revenue of $41.46 billion, up 73.7% sequentially, with GAAP operating profit of $33.32 billion at an 80.4% margin and operating cash flow of $25.39 billion. The Cloud Memory and Core Data Center units posted quarterly operating margins of 78% and 83%, and FQ4 guidance calls for $50.0 billion of revenue at roughly 86% gross margin. Evidence: EV-HWX-MU. - CL-HWX-007 [Fact]: Samsung Electronics reported Q2 2026 group revenue of KRW 171.5 trillion, with the Device Solutions segment contributing KRW 127.5 trillion of revenue and KRW 89.2 trillion of operating profit. The group basis includes handsets, displays and foundry alongside memory. Evidence: EV-HWX-SAMSUNG. - CL-HWX-008 [Fact]: Kioxia reported fiscal 2026 first-quarter revenue of JPY 1.767 trillion, up 76.2% sequentially, with non-GAAP operating profit of JPY 1.326 trillion at a 75% non-GAAP operating margin. The non-GAAP basis is company-defined and differs from the GAAP result. Evidence: EV-HWX-KIOXIA. - CL-HWX-009 [Fact]: SanDisk reported fiscal 2026 fourth-quarter revenue of $8.97 billion, up 51% sequentially with roughly two-thirds of the increase from price, data-centre revenue up 103% sequentially, and GAAP operating profit of $7.04 billion. Evidence: EV-HWX-SNDK. - CL-HWX-010 [Fact]: Demingli reported first-half 2026 revenue of RMB 16.91 billion, up 311.6% year over year, with attributable net profit of RMB 6.02 billion at a 47.1% blended gross margin. Operating cash flow was RMB -6.74 billion and inventory stood at RMB 21.41 billion, or 65.0% of total assets. Evidence: EV-HWX-DEMINGLI. - CL-HWX-011 [Fact]: Longsys reported first-half 2026 revenue of RMB 24.09 billion, up 136.3% year over year, with attributable net profit of RMB 10.58 billion. Operating cash flow was RMB -3.15 billion and inventory stood at RMB 25.78 billion, or 60.1% of total assets. Evidence: EV-HWX-LONGSYS. - CL-HWX-012 [Fact]: Innolight reported first-half 2026 revenue of RMB 41.78 billion, up 182.5% year over year, and attributable net profit of RMB 13.65 billion, up 241.7%. First-half operating cash flow equalled 0.13 times attributable net profit. Evidence: EV-HWX-INNOLIGHT. - CL-HWX-013 [Fact]: Eoptolink reported first-half 2026 revenue of RMB 20.91 billion, up 100.3% year over year, and attributable net profit of RMB 7.53 billion, up 91.0%. First-half operating cash flow equalled 0.21 times attributable net profit. Evidence: EV-HWX-EOPTOLINK. - CL-HWX-014 [Fact]: T&S Communications reported first-half 2026 revenue of RMB 2.83 billion, up 15.2% year over year, and attributable net profit of RMB 1.20 billion, up 33.9%. First-half operating cash flow equalled 0.83 times attributable net profit. Evidence: EV-HWX-TFC. - CL-HWX-015 [Fact]: Lumentum reported fiscal 2026 fourth-quarter revenue of $1.006 billion, up 109.3% year over year, with a 36.6% non-GAAP operating margin, next-quarter non-GAAP operating margin guidance of 39.5%–40.5% and a next-quarter revenue guidance midpoint near $1.25 billion. GAAP net income is affected by a one-time, non-cash loss on debt extinguishment, so operating comparisons use the company's disclosed non-GAAP basis. Evidence: EV-HWX-LUMENTUM. - CL-HWX-016 [Fact]: Coherent reported fiscal 2026 fourth-quarter revenue of $2.046 billion, up 33.8% year over year, with datacom and communications up 58.6% and a 21.8% non-GAAP operating margin. Full-year operating cash flow was $79.5 million against capex of $1.103 billion, inventory rose from $1.438 billion to $2.581 billion, and the next-quarter revenue guidance midpoint is near $2.3 billion. Evidence: EV-HWX-COHERENT. - CL-HWX-017 [Fact]: AAOI reported second-quarter 2026 revenue of $192 million, up roughly 86.4% year over year, and guided to a third-quarter revenue midpoint near $273 million. The quarter produced $5.5 million of non-GAAP profit while remaining loss-making on a GAAP basis. Evidence: EV-HWX-AAOI. - CL-HWX-018 [Derived]: On their own guidance midpoints, Lumentum, Coherent and AAOI imply roughly 24%, 12% and 42% sequential revenue growth for the coming fiscal quarter (12.5/10.06, 23/20.46, 2.73/1.92). All three guiding to sequential growth indicates the optics demand slope had not turned as of 2026-09-01. Evidence: EV-HWX-LUMENTUM, EV-HWX-COHERENT, EV-HWX-AAOI. - CL-HWX-019 [Derived]: The peak-earnings annualised multiples used in this object divide market capitalisation by a simple annualisation of the latest quarterly (or half-year) profit. They are point-in-time screens at the 2026-09-01 close, not forward P/E and not price conclusions. For the U.S. names the denominator is non-GAAP net income, which is a different basis from the non-GAAP operating margin quoted alongside it and cannot be substituted for it; the Chinese names use attributable net profit. Market capitalisations and 52-week drawdowns were not reconciled to fully diluted share counts and therefore stay out of the fact layer. Evidence: EV-HWX-MARKET, EV-HWX-MU, EV-HWX-KIOXIA, EV-HWX-SNDK. - CL-HWX-020 [Derived]: The global memory chain separates into three profit layers of different character: constraint owners (SK hynix, Micron, Samsung, holding the capacity, process and customer qualification behind HBM, server DRAM and enterprise NAND); NAND cycle elasticity (Kioxia and SanDisk, directly exposed to NAND pricing and data-centre SSD); and inventory spread (Demingli and Longsys, where much of the profit sits between low-cost inventory and realised selling prices). All three can benefit at once, and their falsification conditions are entirely different. Evidence: EV-HWX-SKHYNIX, EV-HWX-MU, EV-HWX-SAMSUNG, EV-HWX-KIOXIA, EV-HWX-SNDK, EV-HWX-DEMINGLI, EV-HWX-LONGSYS. - CL-HWX-021 [Hypothesis]: Within one AI capex cycle, structural profit should accrue first and persist with the constraint owners, because tight supply reaches their ASPs, gross margin and cash flow before anyone else's. The core variable from September to November is therefore whether profit keeps landing with the companies carrying cleaner balance sheets and tighter customer lock-in. None of these companies discloses segment profit for HBM or enterprise product on a common basis, so the proposition can only be tested indirectly through operating margin, cash conversion and the acceleration of contract prices. Evidence: EV-HWX-SKHYNIX, EV-HWX-MU, EV-HWX-MOTIE, EV-HWX-NVDA. - CL-HWX-022 [Inference]: The 7–8x static annualised multiples at Demingli and Longsys rest on a denominator containing peak profit formed jointly by low-cost inventory, sharply rising ASPs and light impairment. If Q3 gross margin keeps falling, operating cash flow keeps running negative and inventory-to-assets and turnover keep deteriorating, the apparent low P/E is an inflated cyclical denominator rather than a valuation discount. Evidence: EV-HWX-DEMINGLI, EV-HWX-LONGSYS, EV-HWX-MOTIE. - CL-HWX-023 [Inference]: The valuation dividing line in optics has moved from growth to cash conversion. First-half operating cash flow covered only 0.13 and 0.21 of attributable net profit at Innolight and Eoptolink against 0.83 at T&S Communications, and Coherent generated $79.5 million of full-year operating cash flow against $1.103 billion of capex. Against run-rate multiples of 30x to 100x and above, delayed cash recovery amplifies valuation volatility quickly. Evidence: EV-HWX-INNOLIGHT, EV-HWX-EOPTOLINK, EV-HWX-TFC, EV-HWX-COHERENT. - CL-HWX-024 [Inference]: Humanoid robotics lacks a profit vehicle in global public markets comparable to HBM or 1.6T optics: most leading whole-machine developers are private or held inside large groups, and listed component suppliers rarely disclose humanoid revenue as a separate line. Until announcement-grade orders, separately disclosed revenue and gross margin appear together, it remains a long-dated technology option rather than a member of a like-for-like profit-pool comparison. Evidence: EV-HWX-HUMANOID-GAP. - CL-HWX-025 [Inference]: The bear case is supply-side normalisation compressing excess margin: long-term agreements raise demand visibility while capping the incremental ASP elasticity manufacturers capture from rising spot, and 2027 capacity additions narrow the spread further. Separately, press reporting states that Chinese manufacturer CXMT has begun small-volume production of HBM3E. That report is not company-confirmed, no pricing or share effect is inferred from it here, and it is recorded only as an unverified supply-side risk. Evidence: EV-HWX-SKHYNIX, EV-HWX-SAMSUNG, EV-HWX-MOTIE. Evidence: - EV-HWX-TRENDFORCE: TrendForce DRAM spot and module quotes. TrendForce. 2026-09-01. https://www.trendforce.com/price/dram/dram_spot - EV-HWX-MOTIE: Korea August 2026 export release with DRAM and NAND fixed prices. Ministry of Trade, Industry and Energy, Republic of Korea. 2026-09-01. https://www.motir.go.kr/kor/article/ATCL3f49a5a8c/172145/view - EV-HWX-NVDA: NVIDIA FY2027 Q2 CFO commentary. U.S. Securities and Exchange Commission. 2026-08-26. https://www.sec.gov/Archives/edgar/data/1045810/000104581026000073/q2fy27cfocommentary.htm - EV-HWX-SKHYNIX: SK hynix Q2 2026 business results. SK hynix Investor Relations. 2026-07-28. https://news.skhynix.com/en/q2-2026-business-results/ - EV-HWX-MU: Micron fiscal 2026 third-quarter results. Micron Technology Investor Relations. 2026-06-24. https://www.globenewswire.com/news-release/2026/06/24/3317151/14450/en/micron-technology-inc-reports-record-results-for-the-third-quarter-of-fiscal-2026.html - EV-HWX-SAMSUNG: Samsung Electronics Q2 2026 results. Samsung Electronics Newsroom. 2026-07-31. https://news.samsung.com/global/samsung-electronics-announces-second-quarter-2026-results - EV-HWX-REUTERS-SKH: Reuters report on SK hynix Q2 profit and the market reaction. Reuters. 2026-07-28. https://www.reuters.com/world/asia-pacific/sk-hynix-q2-profit-jumps-557-ai-chip-demand-misses-forecasts-2026-07-28/ - EV-HWX-KIOXIA: Kioxia Holdings FY2026 Q1 results presentation. Kioxia Holdings Investor Relations. 2026-08-07. https://www.kioxia-holdings.com/en-jp/ir/library/presentation.html - EV-HWX-SNDK: SanDisk fiscal 2026 fourth-quarter results. U.S. Securities and Exchange Commission. 2026-08-05. https://www.sec.gov/Archives/edgar/data/2023554/000162828026053346/sndkq4-26ex991xpressrelease.htm - EV-HWX-DEMINGLI: 德明利 2026 年半年度报告 / Demingli 2026 interim report. 巨潮资讯网 CNINFO（中国证监会指定信息披露平台）. 2026-08-28. https://www.cninfo.com.cn/new/fulltextSearch?keyWord=%E5%BE%B7%E6%98%8E%E5%88%A9 - EV-HWX-LONGSYS: 江波龙 2026 年半年度报告 / Longsys 2026 interim report. 巨潮资讯网 CNINFO（中国证监会指定信息披露平台）. 2026-08-28. https://www.cninfo.com.cn/new/fulltextSearch?keyWord=%E6%B1%9F%E6%B3%A2%E9%BE%99 - EV-HWX-INNOLIGHT: 中际旭创 2026 年半年度报告 / Innolight 2026 interim report. 巨潮资讯网 CNINFO（中国证监会指定信息披露平台）. 2026-08-22. https://static.cninfo.com.cn/finalpage/2026-08-22/1225491752.PDF - EV-HWX-EOPTOLINK: 新易盛 2026 年半年度报告 / Eoptolink 2026 interim report. 巨潮资讯网 CNINFO（中国证监会指定信息披露平台）. 2026-08-28. https://www.cninfo.com.cn/new/fulltextSearch?keyWord=%E6%96%B0%E6%98%93%E7%9B%9B - EV-HWX-TFC: 天孚通信 2026 年半年度报告 / T&S Communications 2026 interim report. 巨潮资讯网 CNINFO（中国证监会指定信息披露平台）. 2026-08-28. https://www.cninfo.com.cn/new/fulltextSearch?keyWord=%E5%A4%A9%E5%AD%9A%E9%80%9A%E4%BF%A1 - EV-HWX-LUMENTUM: Lumentum fiscal 2026 fourth-quarter and full-year results. Lumentum Investor Relations. 2026-08-11. https://investor.lumentum.com/financial-news-releases/news-details/2026/Lumentum-Announces-Fourth-Quarter-and-Full-Fiscal-Year-2026-Results/default.aspx - EV-HWX-COHERENT: Coherent fiscal 2026 fourth-quarter results. Coherent Corp. Investor Relations. 2026-08-12. https://www.coherent.com/content/dam/coherent/site/en/documents/investors/financial-releases/2026/august-12/earnings-release-fy26-q4.pdf - EV-HWX-AAOI: Applied Optoelectronics second-quarter 2026 results. Applied Optoelectronics Investor Relations. 2026-08-06. https://investors.ao-inc.com/news-releases/news-release-details/applied-optoelectronics-reports-second-quarter-2026-results - EV-HWX-MARKET: Point-in-time market snapshot, 2026-09-01 close. Google Finance, MarketWatch and exchange closing data. 2026-09-01. https://www.google.com/finance/quote/MU%3ANASDAQ - EV-HWX-CONSENSUS: Third-party sell-side consensus snapshot for MU. Seeking Alpha. 2026-09-01. https://seekingalpha.com/symbol/MU/earnings/revisions - EV-HWX-HUMANOID-GAP: Humanoid robotics segment-disclosure gap in global public markets. Lunartulip Research. 2026-09-01. Disclosure survey: no listed component supplier in the reviewed sample reports separately disclosed humanoid robotics revenue or gross margin Forward falsifiers: - FAL-HWX-001: DDR5 16Gb monthly fixed price, month-on-month change < 0 percent; horizon any monthly fixed-price release from September to November 2026. The month-on-month increase in fixed prices has already fallen from 12.5% in July to 3.3% in August. Turning negative would mean upstream ASP acceleration has not merely peaked but reversed, which directly weakens the premise that constraint ownership captures structural profit first. The threshold sits at zero rather than at some positive rate so that a break does not depend on a subjective view of how much deceleration counts. - FAL-HWX-002: Micron reported GAAP gross margin, FQ4 FY2026 < 80 percent; horizon the FQ4 FY2026 result reported in September 2026. The company's own FQ4 guidance is roughly 86% gross margin. The threshold sits at 80% rather than 86% to leave about six points of buffer, so ordinary variance against guidance is not mistaken for a break. An actual print below 80% would mean ASP or mix is converting materially below the company's own basis, and the pricing power of the constraint owners would need to be reassessed. - FAL-HWX-003: Number of Demingli and Longsys simultaneously improving on Q3 gross margin, operating cash flow and inventory-to-total-assets < 2 count; horizon the Q3 2026 reports filed in October 2026. The research framework for these two requires four signals to appear together, and three of them read directly off the Q3 report: gross margin stops falling, negative operating cash flow narrows materially, and inventory-to-assets and turnover stop deteriorating. If neither company clears all three, profit is still driven by the inventory spread rather than operating quality, and the 7–8x static multiple is an inflated cyclical denominator. - FAL-HWX-004: Innolight and Eoptolink full-year 2026 operating cash flow ÷ attributable net profit < 0.4 multiple; horizon the FY2026 annual reports. The first-half readings were 0.13 and 0.21 against 0.83 at peer T&S Communications, which brackets the range. Staying below 0.4 for the full year would show profit is not converting to cash rather than reflecting half-year timing, and against run-rate multiples of 30x to 100x and above that is enough to overturn the ranking as furthest along on earnings conversion. The threshold sits between the two ends rather than close to the half-year reading, so that seasonal collections are not mistaken for structural improvement. - FAL-HWX-005: Number of Lumentum, Coherent and AAOI reporting next-quarter revenue at or above their own guidance midpoint < 2 count; horizon the fiscal quarters reported from October to November 2026. The three guidance midpoints imply roughly 24%, 12% and 42% sequential growth, and that is the entire public basis for saying the demand slope is intact. Fewer than two of three clearing their own midpoint would push the evidence back to single-company stories and out of cross-sectional reach. The test uses each company's own midpoint rather than external consensus so that it depends only on first-party disclosure. - FAL-HWX-006: Number of listed humanoid robotics component suppliers in the reviewed sample disclosing separate humanoid revenue and gross margin < 1 count; horizon the Q3 2026 reports filed in October 2026. Separately disclosed revenue and gross margin are the minimum threshold for moving humanoid robotics from narrative to profit pool; below it there is no comparable financial object at all. The threshold is one rather than something higher because the current count in the sample is zero, so a single company beginning to break the line out is enough to put the judgment into revision. --- ## Can HBM break the traditional cycle valuation? The next alpha may sit in the revenue-to-cash capture gap Object ID: RO-THEME-MEMORY-CASH-CAPTURE-001 Ticker / coverage: MU, 000660.KS, 005930.KS, SNDK, 285A.T Kind: theme-study Original publication: 2026-09-10 Information as of: 2026-09-10 Current version: 1.0.0 Last modified: 2026-09-10 Canonical English: https://lunartuliplab.com/en/deep-dive/hbm-memory-cash-capture-alpha-2026-09 Canonical Chinese: https://lunartuliplab.com/deep-dive/hbm-memory-cash-capture-alpha-2026-09 Research question: Has HBM contracting raised the cycle floor, or are peak profits financing the next supply wave? Abstract: Demand, pricing, yield and capacity are already crowded debates. The more testable expectations gap is whether product-revenue growth still leaves repeatable cash after customer funding, joint-venture contributions and new capacity investment. Consensus view: AI-memory demand, the HBM4 ramp and long-term agreements improving revenue visibility are widely accepted. Lunartulip view: The work moves from aggregate demand to the revenue-to-cash capture gap and to the second-order supply response induced by high profit: capex, process migration and new entrants. Valuation framing: Valuation is decomposed into attributable product revenue, adjusted cash conversion and competitive-advantage period rather than a single revenue-growth or gross-margin point. The SanDisk bridge is a methodological control, not a proxy for HBM product profit. Claims: - CL-MEMTH-001 [Fact]: SanDisk reported FY2026 operating cash flow of $11.671 billion, purchases of property, plant and equipment of $177 million, company-presented free cash flow of $11.494 billion, and revenue of $20.248 billion. Evidence: EV-MEMTH-SNDK. - CL-MEMTH-002 [Derived]: Reported free-cash-flow margin was 56.77%. After a further $275 million of net joint-venture contributions and $2.476 billion of cash effects related to the new business model, adjusted free cash flow was $8.743 billion, or 43.18% of revenue, a gap of 13.59 percentage points. Evidence: EV-MEMTH-SNDK. - CL-MEMTH-003 [Fact]: Micron disclosed that its HBM4 12-high production ramp was running at roughly twice the pace of HBM3E 12-high and that cumulative HBM4 shipment revenue had exceeded $1 billion. SK hynix reported volume HBM4 shipments, while Samsung reported expanding HBM4 sales and HBM4E samples delivered to major customers. Evidence: EV-MEMTH-MU, EV-MEMTH-SKH, EV-MEMTH-SAMSUNG. - CL-MEMTH-004 [Hypothesis]: The next repeatably testable HBM alpha may sit in the product-revenue-to-adjusted-cash-capture gap. Rising product revenue, stable cash conversion and slow qualified supply would support a higher cycle floor; rising revenue and profit alongside weakening adjusted cash conversion would signal earnings quality peaking ahead of headline profit. Evidence: EV-MEMTH-SNDK, EV-MEMTH-MU, EV-MEMTH-SKH, EV-MEMTH-SAMSUNG. Evidence: - EV-MEMTH-SNDK: SanDisk FY2026 Results and Form 10-K. SanDisk Investor Relations. 2026-08-05. https://investor.sandisk.com/node/8136/pdf - EV-MEMTH-MU: Micron FY2026 Q3 prepared remarks and Form 10-Q. Micron Investor Relations. 2026-06-24. https://s25.q4cdn.com/621799436/files/doc_financials/2026/q3/Q3-FY26-Prepared-Remarks.pdf - EV-MEMTH-SKH: SK hynix Q2 2026 business results. SK hynix Newsroom. 2026-07-30. https://news.skhynix.com/en/q2-2026-business-results/ - EV-MEMTH-SAMSUNG: Samsung Electronics Q2 2026 results. Samsung Newsroom. 2026-07-30. https://news.samsung.com/global/samsung-electronics-announces-second-quarter-2026-results Forward falsifiers: - FAL-MEMTH-001: Adjusted cash-conversion direction while HBM product revenue grows < 0 percent; horizon each of the next two reporting cycles. If adjusted cash conversion falls sequentially while HBM product revenue rises, the gap is widening and the structural rerating hypothesis should be downgraded. - FAL-MEMTH-002: Supplier programs advancing from plan to qualified sellable output >= 2 count; horizon each of the next two reporting cycles. If at least two supplier programmes reach post-qualification sellable output in the same observation window, supply is arriving faster than assumed and competitive duration should shorten. --- ## Financials at the peak, capital flooding in: Micron's escape-from-cycle thesis enters a falsifiable window Object ID: RO-COMP-MU-CAPITAL-CYCLE-001 Ticker / coverage: MU Kind: company-deep-dive Original publication: 2026-09-10 Information as of: 2026-09-07 Current version: 1.0.0 Last modified: 2026-09-10 Canonical English: https://lunartuliplab.com/en/deep-dive/micron-hbm-capital-cycle-2026-09 Canonical Chinese: https://lunartuliplab.com/deep-dive/micron-hbm-capital-cycle-2026-09 Research question: Can HBM contracting offset the supply response arriving in 2027–2028? Abstract: Near-term price and allocation dynamics can still deliver, but the variable that decides whether the cycle is broken is not the next print. It is what happens to price, margin and commodity linkage once capex becomes qualified supply. Consensus view: Record FQ3-26 revenue and margin, FQ4 guidance and long-term HBM agreements improving visibility are already in the price. Lunartulip view: The information gain is the four-way supply response and its concentrated 2H27–2028 arrival window. The thesis must pass contract-price, gross-margin and commodity-linkage tests together. Valuation framing: The reverse DCF reads market requirements of roughly 20% five-year revenue CAGR and roughly 30% steady FCF margin; it does not equate a low forward P/E with undervaluation. Claims: - CL-MUCC-001 [Fact]: Micron reported FQ3-26 revenue of $41.46 billion, up roughly 346% year over year, and a non-GAAP gross margin of 84.9%; FQ4 guidance called for revenue of $50.0 billion ± $1.0 billion and a non-GAAP gross margin of roughly 86%. Evidence: EV-MUCC-Q3. - CL-MUCC-002 [Fact]: Micron disclosed that FY27 capex would run above the roughly $10 billion FQ4 level in every quarter, with more than half of the increment allocated to cleanroom construction. Evidence: EV-MUCC-Q3. - CL-MUCC-003 [Derived]: Under a simplified reverse DCF using a 9.5% WACC, 3% terminal growth, a ten-year window and enterprise value approximated by market capitalisation, the 2026-09-04 snapshot implied 21.57% five-year revenue CAGR and a 28.99% steady FCF margin. The output is sensitive to the price date and model assumptions. Evidence: EV-MUCC-MODEL. - CL-MUCC-004 [Hypothesis]: The thesis that HBM contracting lets memory escape the cycle requires three tests to pass together: 2027 HBM4/4E contract prices, HBM gross margins across all three suppliers, and decoupling between HBM and commodity DRAM. The research note assigns a subjective probability of roughly 25% that all three survive through 2028. Evidence: EV-MUCC-Q3, EV-MUCC-MODEL. Evidence: - EV-MUCC-Q3: Micron FY2026 Q3 prepared remarks. Micron Investor Relations. 2026-06-24. https://s25.q4cdn.com/621799436/files/doc_financials/2026/q3/Q3-FY26-Prepared-Remarks.pdf - EV-MUCC-MODEL: Micron point-in-time research uniform and reverse-DCF-lite output. LunarTulip Research. 2026-09-07. MU capital-cycle uniform · 2026-09-07 Forward falsifiers: - FAL-MUCC-001: 2027 HBM4/4E contract-price YoY decline > 15 percent; horizon calendar 2027. A year-over-year decline greater than 15% would mean custom-product price competition has broken the first condition of the escape-from-cycle thesis. - FAL-MUCC-002: HBM gross margin across all three suppliers < 60 percent; horizon full-year 2027. If any supplier's full-year HBM gross margin falls below 60%, the joint three-condition thesis fails. --- ## Contracts raise the floor; the market prices the peak: Stress-testing SanDisk and the NAND capital cycle Object ID: RO-COMP-SNDK-CAPITAL-CYCLE-001 Ticker / coverage: SNDK Kind: company-deep-dive Original publication: 2026-09-10 Information as of: 2026-09-07 Current version: 1.0.0 Last modified: 2026-09-10 Canonical English: https://lunartuliplab.com/en/deep-dive/sandisk-nand-capital-cycle-2026-09 Canonical Chinese: https://lunartuliplab.com/deep-dive/sandisk-nand-capital-cycle-2026-09 Research question: Do NAND take-or-pay contracts raise the cycle floor, or capitalise peak cash flow in advance? Abstract: Contracting is already visible on the balance sheet, but the floor has not been stress-tested through oversupply. The gap between reported and adjusted free cash flow is the first bridge in judging whether the annuity narrative can hold. Consensus view: Long-term agreements, RPO, customer deposits and the FY28–30 medium-term model support a higher cycle floor, and the market already accepts that direction. Lunartulip view: The market primarily requires peak FCF margin to persist. The 2H27 supply window, floor gross margin and changes in contract liabilities are the more consequential adjudication variables. Valuation framing: The reverse DCF produces roughly 9% five-year revenue CAGR and roughly 57% steady FCF margin, concentrating the valuation burden on margin perpetuity. Reported cash margin and the roughly 43% adjusted reading must remain separate. Claims: - CL-SNDKCC-001 [Fact]: SanDisk reported FY2026 revenue of $20.248 billion, up 175% year over year, operating cash flow of $11.671 billion and purchases of property, plant and equipment of $177 million. Evidence: EV-SNDKCC-10K. - CL-SNDKCC-002 [Fact]: The company disclosed ten long-term agreements with eight datacentre and edge customers, $93.9 billion of expected revenue at floor pricing, $59.8 billion of RPO, $1.242 billion of contract liabilities and $1.5 billion of customer deposits. Evidence: EV-SNDKCC-10K. - CL-SNDKCC-003 [Derived]: Reported FCF margin was roughly 56.77%; after removing net joint-venture contributions and cash effects related to the new business model on the research note's basis, adjusted FCF margin was roughly 43.18%. Both are cycle-point readings and should not be extrapolated directly as perpetuities. Evidence: EV-SNDKCC-10K. - CL-SNDKCC-004 [Hypothesis]: Contracting has genuinely raised the NAND-cycle floor, but whether it becomes an annuity depends on customers continuing to perform at the floor after supply loosens and on the floor gross margin being sufficient to support the medium-term model. Evidence: EV-SNDKCC-10K, EV-SNDKCC-MODEL. Evidence: - EV-SNDKCC-10K: SanDisk FY2026 results and Form 10-K. SanDisk Investor Relations. 2026-08-17. https://investor.sandisk.com/static-files/47c2209d-7ef3-488d-b9f4-62d371ad8740 - EV-SNDKCC-MODEL: SanDisk point-in-time research uniform and reverse-DCF-lite output. LunarTulip Research. 2026-09-07. SNDK capital-cycle uniform · 2026-09-07 Forward falsifiers: - FAL-SNDKCC-001: 4Q26 NAND contract-price change <= 0 percent; horizon October-November 2026. A non-positive contract-price change is the high-frequency signal that the first pillar of the supply-discipline narrative has broken. - FAL-SNDKCC-002: Disclosed floor gross margin < 40 percent; horizon upon first company disclosure. A floor gross margin below 40% would directly weaken the claim that contracting can support the 80% medium-term gross-margin model as an annuity. --- ## NVIDIA: Hard growth, costs on the balance sheet Object ID: RO-COMP-NVDA-FY27Q2-001 Ticker / coverage: NVDA Kind: company-deep-dive Original publication: 2026-09-14 Information as of: 2026-09-14 Current version: 1.0.0 Last modified: 2026-09-14 Canonical English: https://lunartuliplab.com/en/deep-dive/nvidia-fy27q2-supercycle-2026-09 Canonical Chinese: https://lunartuliplab.com/deep-dive/nvidia-fy27q2-supercycle-2026-09 Research question: Can the roughly 70% FY2028 total revenue outlook hold without gross margin breaking the lower band? Abstract: FY27 Q2 revenue of $96.2B and Data Center growth of +117% settle the demand debate; the dispute is gross-margin path, $279B supply commitments and $108.5B guarantees. Consensus view: Q3 guidance of $108B ± 2% beat pre-report consensus and Vera Rubin entered full production. Lunartulip view: We move the weak spot from demand to cost and statements: DSO 45→60 and 97% of guarantees on one project. Valuation framing: Scenarios output enterprise-value sensitivity only; no per-share conclusion or target price. Claims: - CL-NVDA-001 [Fact]: NVIDIA FY27 Q2 revenue was $96,221M, up 106% year over year; Data Center revenue was $89.0B, up 117%. Evidence: EV-NVDA-001. - CL-NVDA-002 [Fact]: Q3 FY27 guidance is revenue of $108.0B ± 2% with no assumed China data-center compute revenue; Q3 gross margin (GAAP and non-GAAP) is 74.0% ± 50bps. Evidence: EV-NVDA-001. - CL-NVDA-003 [Fact]: The 10-Q discloses supply purchase commitments of $279B, guarantee obligations capped at $108.5B (including $105B for the SB Energy/OpenAI-related campus), and receivable days sales outstanding rising from 45 to 60 days. Evidence: EV-NVDA-002. - CL-NVDA-004 [Derived]: On the FY27 Q2 call the CFO guided to approximately 70% total company revenue growth in fiscal 2028, subject to supply constraints; pricing at roughly 26× forward earnings embeds a cyclical form rather than pure software duration. Evidence: EV-NVDA-001, EV-NVDA-003. - CL-NVDA-005 [Inference]: Directional demand validation is complete; the debate shifts to statement shape: the gross-margin path, guarantee concentration and longer DSO change how the balance sheet absorbs any demand slowdown. Evidence: EV-NVDA-002, EV-NVDA-003. - CL-NVDA-006 [Hypothesis]: If Q3 gross margin holds 74.0% ± 50bps and DSO stops rising in tandem, the roughly 70% FY2028 total revenue outlook can still support the multiple; otherwise balance-sheet costs reprice ahead of demand falsification. Evidence: EV-NVDA-001, EV-NVDA-002. Evidence: - EV-NVDA-001: NVIDIA Q2 FY27 press release. U.S. SEC. 2026-08-26. https://www.sec.gov/Archives/edgar/data/1045810/000104581026000073/q2fy27pr.htm - EV-NVDA-002: NVIDIA Q2 FY27 Form 10-Q. U.S. SEC. 2026-08-26. https://www.sec.gov/Archives/edgar/data/1045810/000104581026000074/nvda-20260726.htm - EV-NVDA-003: NVDA market multiples snapshot. 24/7 Wall St via public coverage. 2026-09-10. https://247wallst.com/investing/2026/09/10/nvidia-nvda-stock-price-prediction-where-analysts-see-the-stock-going-by-2030/ Forward falsifiers: - FAL-NVDA-001: Q3 gross margin (GAAP midpoint) < 73.5 percent; horizon Q3 FY27 earnings. Gross margin below the guidance band would unwind the margin assumptions embedded in FY2028 EPS revisions. - FAL-NVDA-002: days sales outstanding > 65 count; horizon Q3 FY27 earnings. DSO continuing higher while revenue growth slows would upgrade statement shape from observation to a first-order variable. --- ## Snowflake: Hard acceleration, soft attribution Object ID: RO-COMP-SNOW-AIDC-001 Ticker / coverage: SNOW Kind: company-deep-dive Original publication: 2026-09-14 Information as of: 2026-09-14 Current version: 1.0.0 Last modified: 2026-09-14 Canonical English: https://lunartuliplab.com/en/deep-dive/snowflake-agentic-data-cloud-2026-09 Canonical Chinese: https://lunartuliplab.com/deep-dive/snowflake-agentic-data-cloud-2026-09 Research question: Will Q3 disclose AI product revenue for the first time? Abstract: Three quarters of accelerating product revenue near +37% is verifiable; the half-from-AI line is management attribution only. Consensus view: The stock rose 16.5% the day after earnings and FY2027 guidance was raised to +36%. Lunartulip view: We set monetized AI revenue disclosure as the falsification switch, not CoCo account counts. Valuation framing: Scenarios use the metrics path for multiples and implied CAGR; no target price. Claims: - CL-SNOW-001 [Fact]: Snowflake Q2 FY27 product revenue was $1,491.9M, up 36.8% year over year, accelerating for a third straight quarter; FY2027 product revenue guidance was raised to $6.07B (+36%). Evidence: EV-SNOW-001. - CL-SNOW-002 [Fact]: NRR was 126% with 828 customers above $1M (+27%); CoCo AI coding agent accounts were disclosed for the first time at more than 9,100. Evidence: EV-SNOW-001. - CL-SNOW-003 [Inference]: The CEO attributed roughly half of the acceleration to AI products, yet source materials disclose no AI revenue amount; under consumption billing the 37% growth reflects real usage while the attribution share cannot be verified independently. Evidence: EV-SNOW-001. - CL-SNOW-004 [Derived]: Reverse-solving the September 4 close implies about 37.8% five-year revenue CAGR and a 23.18% steady-state FCF margin (9.5% WACC, 3% terminal growth); that CAGR exceeds the FY2027 +36% guide and prices near-zero deceleration. Evidence: EV-SNOW-002, EV-SNOW-001. - CL-SNOW-005 [Hypothesis]: If Q3 discloses AI product revenue for the first time and product revenue holds the $1.588B guidance floor, the AI data cloud re-acceleration narrative can persist; otherwise the duration embedded in roughly 18.5× forward sales must come down first. Evidence: EV-SNOW-001, EV-SNOW-002. - CL-SNOW-006 [Inference]: Consumption billing offers no deferred cushion on the way down; falling RPO sequentially plus a forward PE near 113× means a Q3 miss below $1.588B punishes narrative, not just earnings. Evidence: EV-SNOW-001, EV-SNOW-002. Evidence: - EV-SNOW-001: Snowflake FQ2-27 press release. Business Wire / Snowflake IR. 2026-09-02. https://investors.snowflake.com/news/news-details/2026/Snowflake-Reports-Financial-Results-for-the-Second-Quarter-of-Fiscal-2027/default.aspx - EV-SNOW-002: Yahoo Finance SNOW snapshot. Yahoo Finance. 2026-09-07. https://finance.yahoo.com/quote/SNOW/ Forward falsifiers: - FAL-SNOW-001: product revenue YoY growth < 35 percent; horizon Q3 FY27 earnings. Product revenue growth below 35% would undermine three-quarter acceleration and zero-deceleration pricing. - FAL-SNOW-002: NRR < 125 percent; horizon Q3 FY27 earnings. NRR below 125 would show consumption expansion failing to offset optimization-cycle risk. ## Services - Always-On Research Desk: https://lunartuliplab.com/en/desk - AI-native Research Framework Workshop: https://lunartuliplab.com/en/workshop - Machine-readable Research API and point-in-time alternative datasets: https://lunartuliplab.com/en/institutional-access Public metadata supports discovery and citation but does not grant a commercial data license. Research is not investment advice.