FUNDA

FUNDA now offers the Market Top and Bottom Indicator, a mid to long-term contrarian signal system. Low scores signal buy zones, high scores signal sell zones. Top and Bottom Score A 0 to 10 contrarian composite weighted across 11 sub-components, updated daily. Low readings correspond to fear, high readings to greed. As a high-conviction indicator, it triggers approximately 2 signals per year. Component Breakdown 11 dimensions scored on a 0 to 10 scale, covering credit spreads, yield curve, equity flows, bond flows, hedge fund positioning, CTA positioning, long-only positioning, sentiment, greed, equity breadth, and credit technicals. Each score reflects that dimension's current contrarian reading. Market Top and Bottom Indicator is available to all paid Substack subscribers. Start here: funda.ai/zh/agent-chat?…

Funda has been flagging that there is upside in $smtc ACC numbers since early May. The NPO opportunity has also been getting alot of interest.

Deep| $ATI : The Materials Chokepoint of the Starship Era; Superalloy Content Per Vehicle Is 25-30x a Falcon 9 ATI is framed as the public-market chokepoint for Starship-era superalloy demand: ~39 Raptors per stack at 800kg-1t each imply 30-40t of superalloy, or 25-30x a Falcon 9. Space demand could reach ~900-1,100t/yr by 2028 and 2,300-2,900t/yr by 2030, while tight VIM capacity and 10-12-year LTAs drive repricing. Despite this, ATI trades at 31.9x trailing EV/EBITDA versus CRS 38.4x and HWM 43.3x, suggesting re-rating upside. The generational turnover in rocket engines is driving a ~25-30x step-change in superalloy content per launch vehicle - one Starship stack carries roughly as much superalloy as 25-30 Falcon 9s combined. 1. Core thesis: a structural step-change in space demand. A full Starship stack carries ~39 Raptor engines (~33 on the Super Heavy booster plus 6 on the ship). Each Raptor consumes 800kg-1t of nickel superalloy, 7-8x the prior-generation Merlin (~125kg), putting superalloy content per stack at roughly 30-40 tonnes, or 25-30x a Falcon 9. The step-change is dictated by the physics of full-flow staged combustion on methane; BE-4 and Archimedes are converging on the same architecture, making the trend industry-wide and irreversible. Layering in refurbishment of turbine wheels, seals, and thrust chambers every ~10 flights under the reuse economy, we see space superalloy demand of roughly 900-1,100 t/yr by 2028 and 2,300-2,900 t/yr by 2030 (with a conservative floor scenario, excluding refurbishment, of ~480t in 2028; see Section 4). 2. The materials layer is one of the few structural entry points through which public-market investors can express this step-change. SpaceX’s vertical integration has swallowed nearly every link in the rocket value chain; specialty melting is the lone exception, protected by a time-based barrier that capital cannot compress. SpaceX still outsources melt production of its proprietary SX alloys to an external mill under an NDA, and its revealed preference has been to sign a 10-year supply agreement with Korea’s Spear rather than build captive capacity. As volumes ramp, the incremental materials demand must spill over to incumbent mills - and rockets share the same mills and the same specification system as commercial aerospace. ATI is the largest, most full-line listed name in that layer (see “Origin of This Report”). 3. The market prices ATI as a commercial-aero cyclical; we believe it is a structural asset occupying the deepest-moat link in the aerospace materials chain. ATI sits in the specialty alloy melting layer (L2): the nickel round-billet market is ~90% held by ATI, Carpenter, and PCC (the industry’s “Big Three,” with broadly similar shares), and ATI is #1 in nickel plate. At the very top end, only 2-3 US mills can supply rotating-grade melt, and a greenfield entrant needs 7-10 years to reach qualified delivery. As of the July 10, 2026 close (S&P Global MI data), ATI trades at 31.9x trailing EV/EBITDA (~26x on FY26 guidance) - the cheapest of the three listed specialty melt/casting leaders on every valuation metric: Howmet 43.3x and Carpenter 38.4x on same-day trailing figures, with forward P/E likewise the lowest (ATI 40.3x vs. HWM 51.9x and CRS 48.0x). The deepest-moat layer trading at a 15-30% discount is, in our view, a systematic mispricing. 4. ATI’s core moat is a ~40-year qualified-supplier record with zero quality escapes - a time-based moat that capital can neither replicate nor accelerate. Per our channel checks, even a new entrant with unlimited capital would need at least ~10 years to earn Boeing-grade Ti-6Al-4V qualification. FAA full-traceability qualification, once earned, remains effective indefinitely, and customers switching suppliers face multi-year requalification of their own - a two-way lock-in that gives revenue exceptional stickiness. 5. Five structural demand drivers are stacking, extending the industry’s demand growth across multiple cycles: (i) chemistry-locked titanium intensity in composite airframes (~15% titanium by weight vs. ~5% for aluminum designs); (ii) MRO (maintenance, repair and overhaul) demand at record levels with a multi-year backlog (counter-cyclical); (iii) the structural step-change in space: ~39 Raptors per Starship stack at 800kg-1t of nickel superalloy per engine (vs. only ~125kg for the prior-generation Merlin), i.e., ~25-30x a Falcon 9 per vehicle; (iv) AI-datacenter-driven industrial gas turbine (IGT) superalloy demand (the three major turbine OEMs hold over 200 GW of firm orders plus capacity reservations, with delivery slots sold out through 2028-2030); (v) reactor-grade zirconium/hafnium demand from naval nuclear and SMRs (ATI is the Western Hemisphere’s primary supplier). 6. Tight supply and long-term contract repricing are sources of earnings upgrades not yet embedded in Street models. Industry VIM melting capacity is running flat out; announced expansions (ATI 11,000t of VIM and Carpenter 9,000t, both commissioning in 2028 with qualified output from 2029) still fall short of the ~30,000t aerospace supply gap in 2029-30, with IGT demand adding another ~10,000t on top - and mills are deliberately restraining capex to defend pricing. Since 2023-24, OEMs have shifted to 10-12-year LTAs with two-way fixed-volume commitments; base prices step up roughly 10% per year at re-signing and spot has moved faster - “security of supply over price” is the prevailing OEM posture. Detailed Report fundaai.substack.com/p/deepati-the-…

We are delighted that the FUNDA team will be participating in the TomorrowX Summit 2026! This will definitely be the most important and insight-rich conference of the year. Heading to TomorrowX Summit 2026 in Austin, Nov 17-18. Two days with the CEOs, investors, and operators moving TMT. @TomorrowXSummit #tomorrowXsummit2026

Bit of travel with Funda colleagues: Santa Clara: 29 July to 8 August Paris: 1 to 4 September New York: 7 to 15 September SF / Austin: 13 to 20 November

Rate cuts repriced duration into AI infrastructure, and AI infrastructure created a structural bid on storage. The front end of that chain already moved. What comes next has not, and that is the question a macro-to-sector-to-stock rotation funnel is designed to address. 10 new SKILLs are now live on FUNDA: M&A pattern recognition and premium benchmarking, patent litigation network mapping and IP portfolio valuation, product roadmap execution tracking, macro-to-stock rotation funnels, insider ownership signals, employee incentive structure analysis, and power market fundamentals. 6 of these are designed for institutional clients only — for a limited time, full access to the entire set is open to all paid Substack subscribers through July 20, 2026. The rest are available to all paid Substack subscribers.

Weekly|Meta Next-Year Capex Defies Cut Fears, Grok 4.5 Brings SpaceXAI Back, Memory, $TSMC and $ASML Previews, Kioxia ASP Hike, Expert Call Database Last week the tape punished the Q2 winners. This week, we saw a sharp snapback in the latter half. The biggest narrative change came from Meta. Last week's Meta Neocloud news had the market reading it as a prelude to capex cuts, and semis sold off on it. This week, Bloomberg’s reporting on Meta’s capex plans for next year ran counter to the narrative: the company is still hungry for compute, still moving forward with expensive new data centers, and recently signed major compute deals with CoreWeave, Google and Oracle. The sell side moved fast, with some lifting their FY27 Meta capex estimate to $220bn against Street estimates of $160bn or more, and peers calling 2027 spend “shockingly high” after Meta contracted over 5 GWW of data center capacity in 1H26 alone. Add the c. $13bn, 1GW-and-growing Alberta data center announced this week, and the message is consistent. The market spent a week worrying Meta might cut; Meta is planning to spend more. We argued the unwind said more about crowded positioning than fundamentals, and the price action this week backed that view. Grok 4.5 reinforced the same conclusion from the model side. After months of the market writing xAI down to a neocloud, the release puts SpaceXAI back at the frontier-lab table with near-frontier capability at a clearly lower cost per task, trained on the Cursor data flywheel and tens of thousands of GB300s. Compute remains the binding constraint, and we stay bullish on it. Semis earnings season starts next week with TSMC. We published our previews ahead of the prints: another beat-and-raise from TSMC, and higher EUV shipment numbers for ASML as supply bottlenecks ease. Details below. Starting this issue, the Weekly also includes our Weekly Expert Interviews Summary, a snapshot of the expert calls hosted by our analysts this week across optics, AI servers, cloud infrastructure and more. You will find it below. Please reach out to sales@funda.ai for more details. This Week’s Reports SPCX - Grok 4.5 puts SpaceXAI back at the frontier-lab table. The five-player race is back on, and the Cursor coding-agent data flywheel validates the model + product surface + workflow data loop. Our conclusion is unchanged: stay bullish on compute demand. fundaai.substack.com/p/deepspcx-gro… TSMC and ASML - previews into earnings season. We expect TSMC 26Q2 to deliver another beat-and-raise, with Q2/Q3 gross margins above guidance and consensus, and we lift our ASML EUV shipment forecasts to 66/95/109 units for 2026/27/28 as supply bottlenecks ease against strong memory and logic demand. fundaai.substack.com/p/previewtsmc-… fundaai.substack.com/p/previewasml-… Premium Report Snapshot A portion of our research is reserved for Premium subscribers and is not distributed via Substack. Below is a snapshot of what Premium subscribers received this week beyond the Substack feed. Research | Kioxia: Raised ASP Forecasts and Continued Validation of Our Thesis Weekly Expert Interviews Summary A snapshot of the expert calls conducted for Premium subscribers this week. Full transcripts and takeaways are available on the FUNDA platform. Silicon Photonics — AI I/O Bottlenecks and the CPO Roadmap (CRDO, AVGO, MRVL) Former Co-founder, Photonics technology company AI Rack-Scale Servers — Taiwan ODM Share and High-Voltage DC Power Upgrade (NVDA, MSFT, AMZN, GOOGL, META, VICR) Power supply industry professional Nebius — GPU Cloud Competitive Positioning and Data Center Build-Out Constraints (NBIS, ORCL, CRWV, NVDA) Former Senior Data Center Manager, OCI Data Center Optical Interconnects — ZR Demand, Coherent Lite Power Consumption, and the 1.6T Bottleneck (NOK, MRVL, COHR, LITE, GLW, MSFT) Former Professional at CTO Office, Nokia HVDC — 800V Data-Center Power Distribution Density and Coexistence with SST (NVDA, META, BABA) Orbital Data Centers — Droplet Radiators and SpaceX Launch Economics (SPCX) Former Quality Assurance, SpaceX Planet Labs — Satellite Hardware Supply Chain and Pelican Constellation Progress (PL) Former Tech Lead, Planet Labs Power Semiconductors — High-Voltage System Integration for AI Data Centers Former Global Foundry Operation Lead, Vishay AI Server Power Supplies — 800 VDC Migration and PSU Power Shelf Expansion (NVTS, STM, TXN) Former PM, Delta Electronics Detailed Report fundaai.substack.com/p/weeklymeta-n…

Deep| $SPCX : Grok 4.5 Brings SpaceXAI Back at the Frontier-Lab Table; Remain Bullish on Harness Data Flywheels and Compute Demand What matters most about the Grok 4.5 release is that xAI / SpaceXAI is back at the table among North American frontier labs. At the beginning of the year, the North American frontier-model race still looked like a five-player game: OpenAI, Anthropic, Google, Meta, and xAI. But for a stretch, xAI had clearly fallen behind, and the market was more willing to treat it as a new neocloud, so the narrative shrank from five players to four. More recently, reports that Meta wanted to build “Meta Compute” and sell spare AI compute were read as frontier-lab consolidation and compute oversupply, bearish for neoclouds and AI infrastructure. Detailed Report fundaai.substack.com/p/deepspcx-gro…

Preview| $ASML 2Q26: Raising EUV Shipment Expectations With ASML’s supply bottlenecks easing, we raise our 2026/27 EUV shipment forecasts to 66 and 95 units. For 2028, we now expect 109 units. On the demand side, both memory and logic remain strong. In memory, Micron raised its CapEx guidance for this fiscal year from USD 25bn to about USD 27bn at its late-June earnings call, and next year’s CapEx should exceed the previously communicated USD 40-45bn range. Shortly after, the South Korean government, together with Samsung and SK Hynix, announced an investment plan of roughly USD 3.1tn. The plan mostly pulls forward projects originally slated for 2047 and 2045 (Samsung and SK Hynix, respectively) to 2040 and 2033 rather than adding new near-term spending. However, it still gives equipment makers more confidence in long-term orders. In advanced nodes, we expect TSMC to accelerate its capacity build-out as ASML increases EUV supply. We also see Intel possibly raising its 2026 CapEx guidance from USD 18bn to USD 20bn at its Q2 earnings call; if it does, the move matters even more for Intel itself. Detailed Report fundaai.substack.com/p/previewasml-…

Preview| $TSMC 26Q2: Expect Another Beat-and-Raise Q2 and Q3 Gross Margin to Beat Market Expectations We expect Q2 gross margin of 68.4%, above the company’s 65.5- 67.5% guidance, and Q3 gross margin of 69.9% versus the 68% consensus. A favorable product mix and a heavy volume of rush orders should keep margins ahead of market expectations. Detailed Report fundaai.substack.com/p/previewtsmc-…

An ESG rating gets downgraded and the stock sells off that morning. But the metric behind it had been getting worse for quarters. And the governance score was already flagged weeks before the downgrade came out. The data was all public. The signals were already there. IT hiring budgets surge well before the revenue inflection. Meanwhile, a company posts jobs for a tech stack that doesn't match its roadmap. It's all on job boards, out in the open. Yet most go unread until after the move. 10 new SKILLs now live on FUNDA. Full access to the entire set is open to all paid Substack subscribers.

@zephyr_z9 @insane_analyst @zephyr_z9 is correct

@insane_analyst I think ur wafer costs are wrong 4 inch wafers are around 1.5k-2k these days

Two hyperscaler NPO programs and AMD's advanced optics program are all starved for lasers. Compare my InP model with SA model.

13F filings contain position-level data for institutional managers required to report. The raw filings are public on EDGAR, but structuring them into usable form takes manual work every cycle. FUNDA now offers 13F Holdings Preview, a Play that automates this structuring layer. Trade Changes Presents each quarterly position adjustment, covering new entries, additions, reductions, and full exits, with dollar and percentage change attached, ranked by magnitude. The largest capital movements surface first. Holdings Detail The complete position list, sorted by current market value, displays holding amount, portfolio weight, and quarterly change for each security. Designed for examining a manager's core positions and allocation structure. Sector Map Holdings are mapped by weight as area and color-coded by sector. Concentration levels, allocation tilts, and relative position sizing are presented in a single visualization of the overall portfolio landscape. 13F Holdings Preview is available to all paid Substack subscribers. Start here funda.ai/zh/agent-chat?…

$META 's ad recommendation system has, at least up until now, not consumed much in the way of inference compute. For how to scale up the inference system and the reasoning behind it, please refer to our reports 10 months ago. Detailed Report fundaai.substack.com/p/deepmeta-evo…

Research| $META : Building a NeoCloud and Continuing to Rent Capacity are not Contradictory Not a contradiction: Meta considering external commercialization of surplus AI compute does not mean its AI compute demand has peaked. Meta is still locking in major new capacity, including roughly 1.6 GW from Crusoe data centers in Childress, Texas, and Warrenton, Missouri. Generational compute shift: Meta’s large H100/H200 fleet remains valuable for inference, fine-tuning, enterprise model serving, image/video generation, and traditional ML. But for 3T+ parameter MoE models, long context, multimodal training, and RL-heavy post-training, GB200/GB300 and future Vera Rubin systems offer better economics for frontier training. Tiered AI infrastructure: The AI compute market is moving from a single GPU shortage into multi-generation, tiered pricing and usage. GB300/Rubin-class systems remain scarce for frontier model training, while H100/H200 can shift toward inference, hosted models, agent workloads, and external compute monetization. Supply chain implication: Meta building a NeoCloud is not a bearish signal that demand has collapsed. It suggests GPU fleets are becoming financialized, multi-generation assets: older GPUs do not go to zero, and next-generation training compute remains scarce. Detailed Report fundaai.substack.com/p/researchmeta…

I guess what you mean is something like DeepSeek V4’s KV cache compression and NAND-based storage. No one was talking about it during the last round of DeepSeek breakthroughs, because DeepSeek had only just reached Claude 4.5’s level. But the technical breakthrough is obvious. If it were OAI or Anthropic this time, it would indeed be different. fundaai.substack.com/p/deepdeepseek…

Research|Korea's AI Investment Plan: Short-to-Medium Term Supply Tightness Cannot be Alleviated Spot market signals may be misleading: Recent Ornn and Silicon Data datasets show lower GPU spot prices, especially for H100, but these sources mainly aggregate fragmented, smaller-scale compute capacity. Funda’s checks suggest they may not reflect the broader mid-to-large enterprise market. AWS H100 pricing remains firm: For AWS H100 p5.48xlarge in US East / N. Virginia, June spot pricing was higher than in prior months. This contrasts with the softer Ornn H100 data and points to continued strength in large-cloud demand. B300 pricing has not materially declined: The market is shifting toward B300 as the lead accelerator generation, while Ornn does not yet provide B300 data. AWS B300 p6-B300.48xlarge pricing across operating systems has remained relatively stable, with current levels broadly in line with May despite increased supply. Enterprise demand still exceeds supply: Checks with Baseten, FireworksAI, and neocloud providers indicate scaled GPU demand remains tight. One inference customer needing 3,000 GPUs could wait until year-end, while neocloud providers cited demand at roughly 20x supply, early-June GPU price increases of about 30%, and a potential further 20% rental price increase by year-end. Detailed Report fundaai.substack.com/p/researchkore…

Guess who the #1 Substack in Singapore is… it’s Funda.

Thanks man! Looking forward to your thoughts.

Deep| $NVDA : Rethinking NVIDIA's Moat in the AI Stack Apr 23, 2026 Here’s our in-depth report on CUDA published in April. I think it’s worth reading more carefully now. Margin premium under scrutiny: NVIDIA’s Data Center business now dominates revenue, with GPU chip-level gross margin around 84% and GAAP gross margin reaching 75% in FY2026 Q4. Compared with AMD’s 57% gross margin, the roughly 18-point premium is not just hardware strength; it reflects CUDA’s ecosystem pricing power. CUDA’s three-layer moat: The report breaks CUDA into developer ecosystem, closed-source libraries, and closed-source compiler. After 25 years, CUDA has 6 million developers and 300+ acceleration libraries, but super-buyers are diversifying compute, and China’s open-source model ecosystem is shifting toward Huawei Ascend and CANN. AI agents are compressing migration costs: Moving frontier models from CUDA to alternatives like CANN or ROCm used to take quarters of engineering work. With AI coding agents, cycles may compress to weeks or days, and Claude Code has already shown CUDA-to-ROCm translation in around 30 minutes. Kernel and compiler defenses are being repriced: Triton and TileLang make cuDNN/cuBLAS less mandatory, while AI-generated kernels are starting to beat expert-written libraries on selected workloads, including Standard Kernel’s H100 GEMM results and doubleAI’s cuGraph replacement. NVIDIA still has near-term advantages in hardware know-how, system design, and supply chain, but the source of its premium may be shifting from structural lock-in to execution speed and accumulated tuning. Detailed Report fundaai.substack.com/p/deepnvda-ret…