Investment Thesis Database

Sivers is the bleeding-edge independent DFB/CW laser supplier for Co-Packaged Optics (CPO) — the next-generation photonic architecture being deployed by every major hyperscaler. While LITE and COHR are fully allocated to NVIDIA, Sivers has emerged as the Tier-1 laser source for AMD CPO (via GFS), Marvell Celestial, Jabil 1.6T LRO, Ayar, and likely Apple silicon photonics. At a ~$1.5B MC vs. LITE at $60B+, the valuation gap is extraordinary for what is essentially the same critical chokepoint in the laser supply chain. CHIPS Act funding and a pending NASDAQ listing unlock US institutional capital that fund mandates previously barred.

AXT is not just an InP substrate maker — it controls the entire upstream InP supply chain from raw materials (indium, gallium, germanium) through pBN crucible processing to final InP substrate duopoly with Sumitomo. This multi-layer chokepoint ownership means that even if competitors try to scale, they cannot without AXT's upstream materials. Retail short sellers consistently misunderstand this, treating AXT as a simple substrate company. Institutions have been silently accumulating. If China weaponizes the supply chain via Vital/AXT's JV structure, Western hyperscalers have no alternative at scale.

Soitec's Photonics-SOI is deployed in 100% of next-generation AI data centers, per their own EVP. That single quote defines the thesis — this is a functional monopoly on the substrate layer required for silicon photonic chips (the foundational layer of CPO). When Serenity published the SOI thesis at ~$40, institutions were publishing analyst reports saying it was overvalued. One month later Morgan Stanley quietly picked up 6.5% of the float. Up 200%+ since the original thesis.

AAOI is the largest US-based 1.6T optical transceiver and laser fab by projected capacity. Called at $30, now $150+. Amazon has purchase agreements and warrants with AAOI, making them a direct beneficiary of hyperscaler 1.6T validation. If they execute on becoming the largest 1.6T capacity in the US, Serenity sees $30B+ valuation long term.

IQE makes InP epiwafers — the processed wafer layer between AXT's raw substrates and final laser chips. Landmark Optoelectronics (competitor) is reporting output 'far below customer needs,' validating the epiwafer bottleneck. Called at ~$100M MC, now up 260%+. MTSI invested $45M and took a board seat to secure long-term epiwafer supply — the strongest possible institutional validation of the thesis.

AEHR makes wafer-level burn-in test systems — a critical yield step for optical transceivers and silicon photonics chips before volume deployment. Called at the bottom when investors sold on negative 'current earnings' without understanding that qualification-stage companies should be valued on volume ramp indicators, not TTM revenue. After going long, AEHR almost tripled. Now has a T1 optical transceiver customer starting volume burn-in.

Nebius is Serenity's favorite Neocloud — the only one worth owning long-term. Built from Yandex's cloud infrastructure spin-off, Nebius has Avride (autonomous driving), Clickhouse (analytics DB), a strong AI cloud balance sheet, and partnerships with ORCL and MSFT that de-risk the platform. Called at ~$87, now returned 100%+. In a world where IREN and SLNH dilute shareholders into oblivion, Nebius is a compounder.

HPS.A is the largest North American dry transformer and switchgear manufacturer — a critical bottleneck in AI data center power infrastructure. Transformer lead times are 2-5 years. The White House invoked the Defense Production Act declaring a 'national emergency' in grid infrastructure including transformers. Called at $186 in the thesis post, up 47%+ in 3 weeks. Long-term thesis: as AI capex continues and power bottlenecks persist, HPS.A compounds.

LPKF holds a functional monopoly over Laser Induced Deep Etching (LIDE) — the key manufacturing process for glass core substrates, the next-generation advanced packaging substrate after organic. Glass core substrates are required for CPO architectures and advanced chiplet packaging. 80%+ of major global semiconductor players have selected LPKF equipment for process validation. Mass production inflection point: 2027. Volume orders H2 2026 — not included in any current analyst forecasts. Called as potential 10× at $362M MC.

Riber makes Molecular Beam Epitaxy (MBE) machines — the ultra-precise deposition equipment used to grow quantum structures and photonic semiconductor layers. As a duopoly supplier to Microsoft's quantum computing program and to IQE/Fujitsu quantum dot photonics arms, Riber is positioned at the intersection of two multi-decade growth themes. At ~$300M MC, it's a gem that institutions haven't discovered yet — but Apollo could buy the entire float for what is pocket change relative to their dry powder.

MSSCorps holds a functional monopoly over CPO inspection — specifically non-destructive infrared (IR) leakage detection for Co-Packaged Optics yields. The CEO explicitly stated 'the company's goal is to seize a 90 percent share of the CPO inspection market' (translated: 100%, they just can't say antitrust). Customers include TSM, NVDA, AAPL, AMAT, LRCX, ASML, INTC. They've aggressively used litigation (Enli Tech lawsuit) to lock out rivals. At ~$1.2-1.4B MC for what could become a $5-9B monopoly.

Marvell's Celestial CPO architecture positions them as the 'mini-Broadcom' of AI networking ASICs. MSFT Maia ASIC ramp H2 2026 + Google TPU development talks for 2028 provide multi-year revenue visibility. The Celestial acquisition was transformative. Post-POET NDA breach, MRVL is likely sourcing lasers directly from SIVE — actually improving Sivers' position.

AI inference bottleneck is shifting back to CPUs — inference ratio moved from 1:8 (CPU:GPU) to 1:1. AWS Graviton, Google Axion, MSFT Cobalt — all ARM-based. $15B annual revenue target starting to look conservative. Called long when ARM was at lower levels; up 45%+ this month.

Intel serves two theses simultaneously: (1) CPU bottleneck — server CPUs up 10-20% since March, shortages 6-12 months. (2) US foundry sovereignty — Intel Foundry is the only large-scale Western alternative to TSM. Called at $35 PT from Bernstein; now $100. The CHIPS Act anchor role makes INTC politically protected.

SanDisk (spun off from WD) is the pure-play NAND memory supercycle vehicle. Q3 2026: revenue $5.95B (+252% Y/Y), EPS $23.41 (+62% beat), 78.4% gross margin. Q4 guidance: $7.75-8.25B revenue (23% above consensus). Memory companies reprice ahead of earnings via third-party NAND price hike reports — the earnings just confirm. Up 300% YTD.

Towa holds a monopoly over compression molding — the critical packaging step for HBM memory. Every major memory maker (MU, SK Hynix, Samsung) uses Towa. With all three printing from HBM3e and now ramping HBM4, their capex cycles feed directly into Towa equipment orders. Flat YTD despite record memory capex — markets haven't connected the dots yet. Earnings May 11 as catalyst.

Auros is an undiscovered Korean company at ~$270M MC building IR metrology and thin-film thickness measurement for HBM4 hybrid bonding — a segment that KLA previously monopolized. Now qualifying at Samsung factories with SK Hynix likely qualifying too. Two products simultaneously volume ramping H2 2026. A decade of R&D finally paying off.

Reddit is a high-growth money printer that algorithms incorrectly bundle with AI-disrupted social media. Q1 2026: $663M revenue (+69% Y/Y), 91.5% gross margins, $204M GAAP net income. Still down 28% YTD after blowout earnings. Reddit's data licensing for AI training + growing ad platform creates durable FCF. Long from ~$140-148.

HIMS is a DTC health platform with 36%+ short interest despite NVO partnership, peptide arc expansion, multiple global acquisitions, and improving fundamentals in a friendlier macro climate. The short squeeze thesis is structural — short interest can only be so high before covering creates inherent buying pressure. Long around $28.

EWY is the cleanest vehicle for exposure to Samsung and SK Hynix through a US-listed ETF. When everyone was doomposting 'KOSPI crash bubble' and 'Silver Crash', Serenity went long and held conviction. Now up ~3× on 2028 LEAPS. Korean memory makers are printing from HBM3e and ramping HBM4.

AMSC makes HTS superconducting wires and power electronics — explicitly named in the White House Defense Production Act executive order as beneficiaries of grid infrastructure emergency funding. Also has links to MSFT for HTS technology. Serenity built a mini-ETF: AMSC, CLF, PLPC, HPS.A as DPA beneficiaries.

Shunsin is Foxconn's optical packaging and test arm — literally a free piggyback ride to NVDA's CPO supply chain. Foxconn was selected as NVDA's exclusive rack-scale supplier for Groq accelerators, vertically integrating networking including Shunsin's optical packaging. Valued at less than LWLG despite being a production-stage company, not an R&D project.

Win Semi is the compound semiconductor foundry handling Sivers' volume scaling for DFB laser production. Already in supply chains for AVBO, SpaceX, QCOM, Mediatek, MTSI, NXPI, AAPL. When every frontier photonics and space company uses Win Semi, they are the invisible shovel seller of the CPO supercycle.

York Space acquired ALLSPACE — SIVE's lead defense customer. YSS is a national defense prime contractor with links to Space Force, Space Development Agency, DoD, and the Golden Dome initiative. This means SIVE has a backdoor to US space buildout through YSS.

POET packages SIVE lasers into optical interposers for Marvell Celestial — but Marvell cancelled purchase orders after POET CFO violated the NDA. Small positions held. The packaging layer is easier to vertically integrate away than laser IP, making POET the most likely to be designed out long-term. $420M cash provides downside buffer.