Subscribe now

Then came the bottleneck bros. Many companies’ fortune changed. Small, unsexy, unheard-of names put up historic stock runs. A toilet maker turned into a toll collector for the latest and greatest AI build (looking at you & your ceramics crafts, Toto) Murata is one such company that has had a great run.

What do they do? Why are they a bottleneck? Who are their customers? When is their price justified? Should you buy the stock? Read further to know more…

To read my full disclaimer please click here

TL;DR: World #1 in MLCC that is the passive part every phone, car & AI server needs. Elite franchise (materials moat, net cash, real pricing power) coming off a record year.

1.Brief History

Murata was founded in Kyoto in October 1944 by Akira Murata, who started by making ceramic capacitors. To this day, ceramics as material choice is the core offering from the company. Most electronics firms are defined by a product, design or process know-how, Murata is defined by the material and the required craft around it.

Their story since then has been about riding every wave of electronics miniaturization: radios, televisions, the PC, the mobile phone, and now the AI data centre. Each wave demanded more passive components, smaller, and more reliable. Murata's answer has always been to go deeper into the ceramic.

They went on a bit of acquisition spree where they bought Peregrine Semiconductor Corporation in 2014, Sony’s battery division in 2016-17 and then Resonant Inc in 2022. In between, in 2021, they also co-developed a tire monitoring RFID system along with Michelin.

They have complete in house expertise and believe in vertical integration from making their own dielectric powder to manufacturing to delivery. They embody the Japanese concept of monozukuri - the craft of making things.

Murata considers the SCM axis, which delivers value from suppliers to customers through their business, and the ECM axis, which conducts product development and the related development of processes and equipment, to be its monozukuri (manufacturing) domains in a broad sense.

Share

2.Product Portfolio / Business Verticals

Murata’s product portfolio includes capacitors, inductors / EMI filters, radio frequency device & communication modules, battery & power modules and other functional devices. Murata reports in two big buckets: Components; Devices & Modules; plus "Others." The important shift of the last three years is that the profitable, defensible bucket is growing its share of the mix while the cyclical, competitive bucket shrinks.

Why are they in the spotlight now?

One word: MLCC. Every electronic device needs a steady flow of power. Chips, though, gulp power in sudden spikes, and a spike or a dip can make them glitch or fail. A capacitor is a tiny reservoir that sits right beside the chip, storing a little charge and releasing it in an instant to smooth those spikes. It is kind of a shock absorber for electricity.

Capacitor used for this application is made of stacking numerous layers of ceramics on top of each other, hence MLCC: Multi Layer Ceramic Capacitor. They are ubiquitous:~1000 in a smartphone and 30000+ in a single AI server. Murata is the world's largest maker of them, at about 40% of the global market.

Components - crown jewels (~63% of revenue, rising)

This is the capacitor division. MLCC belong here along with EMI filters and inductors. Considering the recent surge in stock prices driven by capacitors, it should be no surprise that capacitors alone are now over half of total company revenue.

They have built themselves a product leadership there that is not easy to break giving them pricing power. Owing to AI demand, they have a very healthy (fat) margin from this segment. Based on recent disclosure, company’s entire profit essentially comes from MLCC.

Devices & Modules — the drag (~36% of revenue, falling)

This is high-frequency modules and SAW filters (the radio-frequency parts inside smartphones), the battery business, and functional devices (sensors). The problem child here is the RF/communications module line, which is tied to the smartphone cycle, faces brutal competition, and has been shrinking

Murata took a goodwill impairment on the SAW-filter business in FY March 2026. This segment barely breaks even; on the company’s disclosure it swung to a small operating loss in FY March 2026. It is the anchor holding back an otherwise pristine margin story.

Two lenses make the transition clearer.

• By application: communication (smartphones) has fallen from ~43% of sales to ~36%, while computers (that includes AI servers) jumped to ~17% and mobility (automotive) holds at ~26%.

• By geography: Greater China is still ~47% of sales — a number to keep in the back of your mind for the risk section.

  

Share Glocal Investor

3.USP

Murata’s structural MOAT is unquestionable. It primarily relies on 4 factors:

The material itself: Murata formulates and makes its own ceramic dielectric powder. This is the hardest part of the MLCC to get right and has a high entry barrier owing to process know-how and decades of accruing experience. Vertical integration into materials is the foundation.

Miniaturization at yield: Anyone can make a large capacitor. Making a reliable one at the 008004 size (smaller than a full stop in this article) with hundreds of internal electrode layers, at yields high enough to sell profitably, is a decades-long manufacturing accumulation. Murata’s scale here (~40% of the world’s MLCCs) offers them a cost and quality advantage that compounds.

Breadth as a system sell: Murata can supply the capacitor, the inductor, the EMI filter and the RF part into the same design. For a customer building an AI board or a car, one qualified supplier across many passive parts lowers integration cost and risk.

Pricing power: The real-time proof of a moat is whether you can raise prices into a shortage without losing the customer. In March 2026 Murata announced 15–35% list-price increases on high-capacitance AI-server MLCCs, high-end automotive-grade parts and RF/microwave MLCCs, effective April. High-end utilization is running above 80%. Any company cannot push through hikes like that unless the customer has no option but to accept.

4.Corporate Governance

Murata is, refreshingly, a clean, widely-held company. There is no dual-class structure and no controlling shareholder. Top holders are the usual Japanese trust banks and insurers, with a long institutional tail (roughly 191,000 shareholders).

The founding family is stepping back rather than digging in. President and CEO Norio Nakajima was reappointed at the June 2026 AGM; Tsuneo Murata, grandson of the founder and long-time chairman, moved to an advisory role and is now retiring. The board is chaired by an outside director (Takashi Nishijima), and the 2026 reshuffle added outside directors to the Audit and Supervisory Committee. A family member (Takaki Murata) remains a senior director, so influence has not vanished, but the trajectory is toward more external oversight, not less.

The balance sheet is the loudest governance signal of all. An 85% equity ratio, a net-cash position (over ¥650B of cash against negligible debt), and capital returns that are generous without being reckless. They have planned a raised ¥70/share dividend and a buyback of up to ¥150B for FY March 2027. This is a conservatively run business. No flags.

5.Financial Analyses

Fundamental

FY March 2026 was a record year: revenue of ¥1,830.9B (+5.0%) and net income of ¥233.9B, both ahead of the company's own February guidance. The engine was capacitors for servers; the brake was RF modules for smartphones.

Murata's operating margin ran 23.4% in FY March 2022, collapsed to 13.1% by FY March 2024 as the post-Covid inventory glut worked through, and has recovered to 15.4%. That is a swing of ten margin points in three years on a business people are now valuing as a secular grower. Passives are cyclical. The chart proves it. The AI cycle can be real and the through-cycle volatility can still be real.

Split the profit by segment, quarter by quarter, and the whole thesis shows up in one picture. Components EBIT are large but swinging with the cycle, from a trough near ¥45B in early 2023 to fresh records approaching ¥86B by late 2025. Devices & Modules is erratic & repeatedly loss-making, gouging out quarterly losses of roughly ¥54B and ¥36B in the RF- and smartphone-heavy periods.

In practice, Murata's profit is Components' profit, minus whatever Devices & Modules gives back that quarter and it has been cyclical in the past.

Leave a comment

Read more

It was very challenging and confusing, as expected, to understand various terminologies like substrates, wafers, ABF, silica sand et al. It took a lot of time for me to declutter all this in pre GPT era.

The aim of the article is to make this journey easier for you by walking through various layers of manufacturing and list down crucial names across the value chain. Let’s begin

To read my full disclaimer please click here

Before everything, this video from is a very good place to begin your journey.

Why the wafer?

In the pre bottleneck era, wafers were just the silent, studious set of companies who just silently went about their job with little attention. The importance were always understood and respected. Transistors weren’t grown out of trees, rather they were literally grown on polished disc of crystals aka wafer.

Properties of that disc set the ceiling for everything above it: how fast the chip can switch, how much heat and voltage it survives, whether it can carry light instead of electricity at all. The material decides what is possible, and only then does the clever design get to do its work.

Wafer layer is one of the quietest and most durable oligopolies in all of technology. Growing a defect-free crystal disc, atom by atom, at industrial scale, is one of the genuinely hard manufacturing problems left on earth. It requires decades of process craft that is difficult to buy off a shelf.

That kind of moat is exactly the sort that survives a downturn. It is a place worth understanding before investing. They are essentially the foundation of semiconductors.

What "substrate" means?

Substrate is the foundational base material upon which integrated circuits and transistors are fabricated. It provides crucial physical support, thermal management, and electrical connectivity for microelectronic devices.

“Wafer” is the physical form factor, while “Substrate” is the functional role.

The industry uses the word "substrate" for two completely different things, and confusing them is the fastest way to get lost, so let me draw the line before we start.

Front end substrates: Wafer in itself on top of which chips are made. The polished discs of single-crystal material upon which transistors are physically built inside the fabrication plant (fab). This material defines the physical and electrical ceiling of the completed chip.

Back end substrates: High-density organic or inorganic carriers upon which finished silicon chips are mounted during assembly to interface with a printed circuit board (PCB). This is an interconnect industry structurally closer to advanced electronics manufacturing.

This post covers primarily front end substrates. I have laid this out as a chain, the same way I did the nuclear ecosystem, for easier breakdown.

The first section walks the process: how a fistful of purified sand becomes a finished wafer, step by step, from raw silicon through to the epitaxial coating, with the companies that own each step.

The second section walks the variations: the engineered and exotic substrates, SOI, silicon carbide, gallium nitride, indium phosphide and a few honorable mentions, that do the jobs plain silicon cannot, and where the tightest monopolies in the whole industry happen to live.

The intention is not to teach you materials science. It is to show you where the bottlenecks and the pricing power sit.

Section I: The Base Silicon Process

Silicon, passes through four stages before it is ready to become a wafer chip. For every material, there will be “names to watch” and “nature of business”

Stage 1: Polysilicon Purification

It begins with silicon refined to a purity that is genuinely difficult to picture. The electronic grade that becomes a chip is eleven nines pure, 99.999999999%, meaning one stray foreign atom for every hundred billion silicon atoms.

This is a different animal from the solar-grade polysilicon. There is abundance there, and some producers make both, but the gap between the two grades is huge, which is precisely why the electronic grade carries such pricing power. It is the feedstock for every silicon wafer downstream.

Nature of business: Defensive and unglamorous. In 2025 with a Section 232 investigation into polysilicon imports, US reemphasized its importance.

Names to watch: ~75% of market is held by 2 companies: Wacker Chemie (WCH.DE) and privately held Hemlock Semiconductor (majority Corning). Japan’s Tokuyama (4043.T) and Mitsubishi Materials (5711.T) are other big names. Leader is Wacker who also have a broader chemicals and solar business under same roof.

Stage 2: Single-Crystal Ingot Growth

The polysilicon from Stage 1 is pure, but it is in the wrong physical form. The clue is in the name: polysilicon is a jumble of many small crystals fused together. A chip cannot be built on a jumble. It needs one continuous, flawless crystal with a single lattice orientation running end to end. Stage 2 is where that transformation happens through a method called Czochralski pulling.

High-purity polycrystalline silicon is melted in a quartz crucible [A] inside a Czochralski puller under a controlled argon atmosphere, heated by a resistance element. Once the melt stabilizes near silicon's melting point of ~1,412°C, a rotating single-crystal seed is dipped into it [B]. A slight temperature drop starts silicon crystallizing onto the seed [C].

As the seed is drawn slowly upward a cylindrical single-crystal ingot forms beneath it [D]. Controlling the pull speed, the crystal and crucible rotation, and the temperature gradient holds the ingot at a constant diameter while it inherits the seed’s lattice orientation [E–F]. Germanium uses the same method with a graphite crucible instead of quartz and lower temperatures, matched to its ~938°C melting point.

Nature of business: Process craft, and the hardest, least visible part of the moat in the whole chain. A single misplaced atom or trapped impurity turns every wafer sliced from that section into scrap. This is the reason “silicon-wafer startup” aren’t common: no amount of capital cuts the decades of furnace know-how it takes.

Names to watch: No standalone players. Growing the crystal and slicing it are a single integrated craft, folded inside the wafer majors in Stage 3.

Stage 3: Slicing, Etching, and Polishing (Prime Wafers)

Then ingots are then sliced, processed and polished to form Prime silicon wafers. Almost entire chip industry runs on the base of this chips. It comes mainly in 2 diameter variations: 200mm & 300mm.

The 300mm wafer carries the advanced money: the logic in CPUs, GPUs and AI accelerators, and the DRAM, NAND and high-bandwidth memory those systems need. That is the segment the AI build is pulling hardest.

The older 200mm wafer still runs an enormous business underneath, carrying the power-management, analog, microcontroller, image-sensor and display-driver chips that fill every car, phone and appliance.

Nature of business: Oligopolistic in nature with 5 players dominating it. MOAT is deep owing to high entry barrier & decades of process know-how and customer relationship. Qualifying a new wafer supplier can take years. It is cyclical.

Names to watch: Shin-Etsu Chemical (4063.T) and SUMCO (3436.T) hold roughly half the 300mm market between them, with GlobalWafers (6488.TWO), Siltronic (WAF.DE) and Korea's SK Siltron (inside SK Group, not separately listed) making up the rest (industry estimates, flagged). Leader: Shin-Etsu.

Stage 4: Epitaxial Coating (Epi Wafers)

Last step: A layer of coating is grown on top of the prime polished silicon wafer. It carries most of the electrical performance while the base wafer offers mechanical support. Unlike steps 1 - 3, this is done not on all chips because epitaxy is an extra, expensive step, and a plain polished wafer is already good enough for many chips.

At the end of stage 2, causes some impurity / defects stay on the wafer. For many applications like memory or standard logic, this imperfection is acceptable. Epi skin are essential when the device physics demands something just prime silicon wafer cannot give: a layer far cleaner and more precisely doped.

Applications like image sensor in phone, power devices, CMOS require this. Epitaxy is also the bridge into the variations below, because growing gallium nitride on a silicon wafer is itself an epitaxy problem.

Nature of business: Value-add layer sold at a premium over bare polished discs, and often captured captively inside the device makers. The one listed pure-play behaves like the small, cyclical, single-technology bet it is.

Names to watch: IQE (IQE.L) is the listed specialist, growing the compound-semiconductor films that go into smartphone RF front-ends and photonics. The Stage-3 majors also sell epi wafers at a premium. Leader: fragmented.

Share

Section II: Material Variations & Structural Moats

When industry applications outgrows standard silicon, specialized compound materials are added to the base chips. This is where hyper-concentrated monopolies emerge. Each variation will have “Application” and “Problems it solves” section in addition to give context

SOI, silicon-on-insulator: the French near-monopoly

Problem it solves: Ordinary silicon is slightly conductive, so at radio frequencies the signal leaks down into the bulk wafer, wasting power and adding noise. It also enables a phenomenon called “parasitic capacitance” (check glossary in the bottom). Plain silicon is a leaky, sluggish foundation for a high-frequency or ultra-low-power chip.

Parasitic capacitance = an unwanted capacitor that forms between a transistor & the conductive silicon wafer beneath it. Every switch has to charge/drain it first, which wastes time and power.

SOI fixes these problems with a clever sandwich. It kills the leakage, and by pushing the transistor away from the substrate, shrinks the parasitic capacitor to almost nothing. There are two ways to build it:

• Bonding (the dominant method): oxidise one wafer's surface into glass, bond it face-down onto a handle wafer so the glass is trapped between them, then thin the top wafer to a sliver of silicon.

  

• Oxygen implantation (SIMOX): fire oxygen ions deep into a single wafer and anneal, so the buried glass forms in place.

Either way, the same stack results, top to bottom: thin working silicon, buried glass, thick support silicon. The result is a chip that switches faster while sipping far less power. The same buried insulator also confines light, which is why it matters for photonics.

Application: SOI are used when higher switching speeds, better power efficiency & Monolithic integrations are needed. Two main applications for SOI Wafer: RF SOI & Silicon photonics.

1. RF-SOI: used in high-frequency wireless communications to build front-end modules. It powers components like antenna tuners, RF switches, and power amplifiers. Are used in smartphones, automotive, wearables etc

2. Silicon Photonics: Silicon-on-Insulator (SOI) technology is the core foundation of silicon photonics. It traps infrared light like tiny fiber optic cables right on a chip, combining high-speed electronics and optical logic at the speed of light.

Nature of business: Near-monopoly of SOITEC built on a proprietary process called Smart Cut, Soitec's bonding & splitting method that sits behind nearly every SOI wafer made. Others make SOI too (Shin-Etsu, GlobalWafers, SUMCO), but mostly under a Smart Cut license and in photonics-grade SOI it is the only qualified volume supplier. Licensed IP plus sole-supplier status: Rare double moat

Name(s): Soitec (SOI.PA) estimated to make the large majority of the world's SOI wafers. Leader: Soitec, decisively. Shin-Etsu (4063.T) (the main Smart Cut licensee), GlobalWafers (6488.TWO) and SUMCO (3436.T) make mostly general-purpose SOI and sit well behind

Silicon carbide, SiC: the battlefield

Problem it solves: silicon breaks down under high voltage and gives up at high temperature. To handle serious voltage a silicon power device has to be built thick, and thick silicon conducts poorly, which wastes energy as heat every time it switches. That is dead weight in an EV, where every watt lost is range lost.

Silicon carbide handles high voltage and high temperature far better than silicon, tolerating roughly ten times the breakdown field and switching with far lower losses.

Application: EV powertrain. Used in traction inverter & onboard charger, where switching power more efficiently translates into range. Beyond cars, it also serves DC fast-charging stations, solar/wind inverters, industrial power supplies, rail and grid electronics.

However, a critical second structural growth curve has emerged in AI Datacenters. Specifically its use as thermal interposers in next-generation AI platforms (like NVIDIA's Rubin) to manage extreme GPU heat loads emerging .

Nature of business: Most contested substrate on the board, and the cautionary tale of this whole piece. Chinese share climbed from around a tenth in 2021 to near forty percent by 2024, collapsing substrate prices by a reported thirty percent in a single year and breaking the sector. A superb material is not a good investment if the supply curve runs away from the price. The safer exposure is through diversified device makers that grow their own substrates and are not naked to the substrate price war.

Names to watch: STMicroelectronics (STM) and Infineon (IFX.DE) are the big Western automotive-power names, with Rohm (6963.T) and Resonac (4004.T) in Japan (Resonac expanding non-Chinese capacity by choice). Substrate makers: Coherent (COHR),China’s SICC (688234.SH) and privately held TanKeBlue. Wolfspeed (WOLF) for restructuring specialists only. Leader: contested, no clean monopoly, which is the whole problem.

Gallium nitride, GaN: two very different games

Problem it solves: Silicon switches too slowly and lossily for power conversion that is both compact and efficient, so chargers stay bulky and run warm, and at 5G/radar frequencies it cannot deliver the power at all. GaN switches far faster with lower losses, shrinking a charger to a fraction of the size

GaN almost never gets its own wafer. It is grown as a thin film on a base, and the base you choose defines the cost, the quality and the job:

• GaN-on-silicon: grown on a cheap silicon wafer. Lowest cost, but a capped quality. The volume power play.

• GaN-on-silicon-carbide: grown on a pricier SiC base that dissipates heat far better, which suits high-power RF. The defence and telecom play.

• Bulk GaN (GaN-on-GaN): a wafer made of GaN crystal itself, no foreign base and no mismatch. Hardest and priciest to make, highest quality, aimed at next-generation vertical power devices.

  

GaN vs SOI: SOI was about isolation (cleaning up silicon’s parasitic losses so low-power and signal-level chips run efficiently), GaN is about power: moving watts efficiently at speed. In an RF chain the two are complementary. SOI does the switching and control, GaN is the power amplifier.

Application: Power side (GaN-on-silicon): phone and laptop fast chargers, 48V AI-datacenter supplies, consumer and industrial power. RF side (GaN-on-silicon-carbide): 5G base stations, military radar, satellite links, defence.

Nature of business:

• GaN-on-silicon power is a crowded scale-and-cost race, not a monopoly.

• GaN-on-silicon-carbide RF is a concentrated oligopoly of a few specialists, now pulled more by defence and radar than by soft 5G telecom.

• The higher-value bulk GaN corner (native crystal for vertical power devices) is an estimated 85%-plus Japanese oligopoly and gets interesting if vertical GaN scales.

Names to watch:

• On-silicon: Navitas (NVTS), Innoscience (2577.HK) the Chinese IDM, Infineon (IFX.DE), STMicro (STM) and epi from IQE (IQE.L). fragmented

• Bulk: Sumitomo Electric (5802.T), Mitsubishi Chemical (4188.T), Sciocs. Leader: Sumitomo Electric

• On-SiC: Qorvo (QRVO), MACOM (MTSI), NXP (NXPI) and Sumitomo Electric (5802.T). Leader: Qorvo/MACOM

  Share Glocal Investor

Indium phosphide, InP: the AI-optics dark horse

Problem it solves: silicon cannot make light (an indirect bandgap, if you want the term), so it will not emit a usable laser or detect light at speed. As AI datacenters move data as light (Photonics), that is a hard wall, and InP is the material that clears it, emitting and detecting light at very high speed.

Application: the lasers and photodetectors inside 800G and 1.6T optical modules are built on InP, so as hyperscalers push to those speeds, substrate demand follows. Also telecom long-haul, high-speed RF, sensing and LiDAR, and photonic integrated circuits.

Nature of business: the purest listed expression of AI optics at the material level, small market today, real pull. The other name, alongside Soitec, that I most want to deep-dive. The pure-play carries central China exposure, a question for that deep dive.

Names to watch: Sumitomo Electric (5802.T) the anchor, Coherent (COHR) the integrated device-and-substrate play, AXT (AXTI) the US-listed pure-play (via a China subsidiary it is separately floating). Leader: Sumitomo Electric.

Honorable mentions: sapphire and the rest

Sapphire solves a narrow problem: you cannot grow an efficient blue LED on silicon (it absorbs the light and the lattices mismatch), so you need a transparent, insulating, lattice-friendly base. It doubles as scratch-proof cover glass on lenses and watch faces. Mature market, fragmented across mostly private names like Monocrystal, Rubicon and CoorsTek. Engineered and reclaimed silicon get a nod too. Named for completeness, not weighting.

Share

One more substrate, and it is not a wafer

Everything above was the front-end substrate, the wafer the transistor is built on. The back-end substrate is where we finish.

Once cut from the wafer, the die cannot be soldered straight onto a board. It first mounts onto a package substrate, a dense multi-layer carrier.

The premium end (CPU/GPU substrates) is a tight group: Japan’s Ibiden (4062.T) and Shinko Electric (6967.T), Taiwan’s Unimicron (3037.TW) and Nan Ya PCB (8046.TW), Austria’s AT&S (ATS.VI), Korea’s Samsung Electro-Mechanics (009150.KS).

But the name nobody expects: the insulating film at the heart of virtually every high-end substrate is ABF, Ajinomoto Build-up Film, made by Ajinomoto (2802.T), the MSG-and-seasoning company. A food business with a near-monopoly on a material computing depends on.

This layer is being disrupted by glass, a story I have already told. This material pivot introduces a new group of supply chain beneficiaries: established glass materials giants (Corning, SKC/Absolics, SCHOTT) and Tier-1 display manufacturers (such as BOE Technology) pivoting legacy LCD glass-processing lines into advanced semiconductor packaging units.

Packaging substrate is a different industry that deserves its own map. Consider this an entree

Vote whether you would love the second part covering back end substrates too.

Conclusion

I am not telling you to buy a basket of wafer stocks, only where to look. The map sorts into three buckets, the “secure the base while you aim for the moon” logic.

Compounders (the base): Shin-Etsu (4063.T) and Sumitomo Electric (5802.T). Quality exposure to structural demand.

Cyclical oligopolists (the middle): SUMCO (3436.T), Siltronic (WAF.DE), GlobalWafers (6488.TWO), STMicro (STM), Infineon (IFX.DE). So far cyclical but can become structurally sound depending on AI capex.

Monopoly and moonshot single-bets (the small slice): Soitec (SOI.PA), AXT (AXTI), Navitas (NVTS), IQE (IQE.L), and the eyes-open Chinese plays SICC (688234.SH) and Innoscience (2577.HK).

I would like to conclude this post by saying that even thought I find a lot of interesting monopolies like IQE, SOI (holding), AXTI, one name has been on my watchlist for long: STMicro. Along with SOI, I think it is one of the names to own in this theme. They also have exposure to other business verticals.

But that’s a discussion for another day. Hope you enjoyed this post.

We do have a small favor to ask…

If you got value out of this article, we would really appreciate if you can spread the word amongst your friends, family or social media. We see it as a huge token of recognition from you. Thanks a lot for your time.

Any questions about this post or other things regarding stock market? Questions about some of our holdings? Let us hear them in the comments!

Leave a comment

Please do restack, share and comment as a show of your approval (or disapproval) of the content

Glossary

Parasitic capacitance: A capacitor is just two conductors with an insulator between them. It stores charge, and it takes time and energy to charge and discharge. Each transistor sits directly on the conductive wafer, an unwanted capacitor forms between the two. Every time the transistor flips between on and off, it has to charge or drain that unwanted capacitor first. That takes a moment and burns a little power. Multiply across billions of transistors switching billions of times a second and it drags down speed and wastes energy.

Epitaxy (epi): an ultra-pure crystal film grown one atomic layer at a time on top of a finished wafer, carrying the active device while the bulk wafer underneath provides support.

Buried oxide (BOX): the thin glass (silicon dioxide) insulating layer sandwiched inside an SOI wafer, sitting under the thin working silicon and above the thick support silicon.

Latch-up: a failure mode in silicon logic where an unintended current path switches on and can destroy the chip; epi layers help suppress it.

Indirect bandgap: the property of silicon’s crystal structure that stops it emitting usable light, which is why light-emitting jobs need compound materials like indium phosphide.

Subscribe now

Every tech hype cycle has the same problem: Cool, sexy product unable to clearly define how (or how much for that fact) they will earn money off it. Humanoid robotics aren’t the exceptions. Viral videos. Investor buzz. Visionary promises. Zero commercial traction... until now.

The first company in this space to go public with serious institutional backing, a real order book, and operational hours in live warehouses. Question it has to answer: “Can they scale fast enough, cheap enough and big enough?”

Agility robotics has the unenviable task of answering this billion dollar question. Its robot is called Digit. And it is the first US-listed pure-play humanoid robotics company in history. They already have humanoids working in warehouses of some notable names. It’s aiming to go public via SPAC mechanism. Ticker :

To read my full disclaimer please click here

TL;DR: Agility Robotics is the only humanoid robot company with documented commercial deployments. Revenue hasn’t been disclosed yet. But the strategic backing, commercial proof points, and first-mover advantage in this market makes this a genuinely interesting opportunity

1.Brief History

Agility Robotics was spun out of Oregon State University’s Dynamic Robotics Lab in late 2015 by Jonathan Hurst, Damion Shelton, and Mikhail Jones. The founding thesis: solve bipedal locomotion so robots can work in spaces designed for humans, aisles, ramps, stairs, where wheeled automation can’t go.

The early years were research-first. ATRIAS (the OSU prototype) became Cassie (2017), a legless biped sold to labs. An $8M Series A from Playground Global and DARPA grants kept the lights on. In 2019, they added a torso, arms, and sensor, Digit was born. Ford signed on as the first customer to test last-mile delivery. That use case fizzled, but it proved the hardware worked outside a lab.

The real pivot came in 2022. A $150M Series B led by DCVC and Amazon’s Industrial Innovation Fund shifted Agility’s focus squarely to warehouse logistics. By late 2023, Digit was running in Amazon fulfilment centres and a GXO/Spanx facility near Atlanta. In 2024, GXO signed the industry’s first humanoid RaaS contract and Digit earned ISO 13482 certification for cage-free operation.

2.Product Portfolio / Business Verticals

Digit is a full-size humanoid built for material handling, payloads up to 35 lbs, bipedal navigation through human-designed environments, no facility retrofitting required. Current deployments handle tote picking, conveyor loading/unloading, stacking, and manufacturing line-feeding. It works, but within constraints: battery life limits continuous duty cycles, early models had thermal throttling issues in warm warehouses (addressed in a 2025 cooling redesign), and all current deployments still require physical separation between Digit and human workers.

Digit v5 is the next generation, slated for 2026 commercial launch. The headline feature is cooperative safety, the ability to work alongside humans without cages or barriers. If it works as designed, this is the unlock for full-facility deployment. If it doesn’t ship on time or the safety certification stalls, the commercial pipeline compresses significantly. The entire $300M order book is for v5 units.

Agility Arc is the cloud platform that turns individual Digits into a managed fleet - task assignment, WMS/MES integration, remote monitoring. One operator can supervise dozens of units. Arc is what makes the RaaS model operationally viable; without it, each Digit would need dedicated human oversight that destroys the labour-cost savings.

Verticals today: Logistics/3PL (GXO, Amazon, Mercado Libre), automotive manufacturing (Toyota Motor Manufacturing Canada), and industrial manufacturing (Schaeffler). All material-handling use cases. The company talks about grocery, hospitality, and elder care as future TAM expansion, but those are roadmap slides, not revenue.

Share Glocal Investor

3.USP

A decade of locomotion research. Agility’s bipedal locomotion work began over a decade before the current wave of humanoid startups. The trajectory from DARPA-funded research through Cassie to Digit represents compounding iteration that cannot be replicated by throwing money at foundation models. The company holds 34 issued patents and has 60+ pending non-provisional applications.

Critical hardware systems, including actuators, modular end-effectors, the whole-body control platform, sensor architecture, long-life quick-charge batteries, and the system-wide safety architecture, are all proprietary and in-house. McKinsey, as cited in Agility's investor presentation, has identified precisely these categories as the primary drivers of cost and performance differentiation in humanoid robots.

Commercial deployments: Most important differentiator in the sector right now. Nine committed customer facilities. 65,000+ hours of real-world commercial operation. GXO Logistics signed the industry’s first formal commercial humanoid RaaS contract with Agility in 2024.

At GXO’s Flowery Branch, Georgia facility, Digit has moved more than 100,000 totes at 98% accuracy. Schaeffler has 8 active deployments. Toyota Motor Manufacturing Canada has a signed RaaS agreement for its Woodstock, Ontario plant. Mercado Libre is live. Amazon has run multiple pilots and is expected to be an active customer again with Digit v5.

No other humanoid company can show this. Everyone else is still in demonstrations and controlled environments.

The data flywheel: Because Digit is already operating at scale across multiple real customer environments, it is generating proprietary physical AI training data that competitors without commercial deployments cannot match. Each deployment generates operational data that improves the embodied AI, which unlocks new workflows, which enables more deployments, which generates more data. This is the same compounding dynamic that gave early cloud players an advantage: the more you deploy, the better you get, and the harder you are to displace.

Strategic backing: Amazon, NVIDIA, SoftBank, Foxconn, Schaeffler are vested into Agility’s success not just as a shareholder. Amazon needs warehouse automation. NVIDIA wants its Isaac robotics and Halos safety stacks deployed commercially, Agility is the first company to adopt Halos. Foxconn is anchoring the $200M PIPE, signalling manufacturing partnership intent. Schaeffler is both an investor and a paying customer. These relationships create integration depth that a new entrant cannot replicate quickly.

RoboFab: Agility’s manufacturing facility in Salem, Oregon is designed for 10,000 units of annual production capacity. Approximately 75% of Digit’s parts are domestically sourced, which is a meaningful competitive positioning in the current US industrial policy environment.

The Customer Acceleration Program (CAP). This is Agility’s structured onboarding pipeline. A prospect goes from Proof of Tech (0-2 months, testing in Agility’s facility) to Proof of Concept (2-3 months, on-site pilot) to RaaS Pilot (4-6 months, commercial evaluation) before converting to an ongoing service relationship. There are currently 30+ potential customers in active pipeline discussions or in the CAP program. This is a repeatable commercial motion, not ad-hoc deal-making.

4.Corporate Governance

100% of existing Agility shareholders are rolling their equity into the combined public company. That is meaningful: the insiders and early investors are not cashing out. They are betting alongside new public investors.

The leadership team was restructured deliberately in 2024 with the Peggy Johnson appointment. Johnson brings enterprise sales, government relations, and large-organization management experience, exactly what you need when transitioning from startup to scaled commercial operation. The founding duo (Hurst and Shelton) remain in operational and strategic roles, preserving technical continuity.

The SPAC structure itself introduces governance considerations worth flagging. SPACs typically mean fewer disclosures at deal announcement than a traditional IPO. That said, Michael Klein's Churchill Capital franchise’s recent track record is hard to dismiss: . Both have delivered significant returns post-merger under the Churchill banner

5.Financial Analyses

Fundamentals

Doing financial analyses on a company that has no revenue is difficult. Comparing their 2024 and 2025 figures : SG&A has more than doubled; R&D has increased by ~40%. Apart from CapEx, practically every cost contributor has increased. Signs are clear that the commercial organizations are taking shape.

Positive read on this: Cash position. At expected cash position of ~$575M post SPAC, this cash is sufficient to survive 3 years at 2026 run rate. This breathing room provides ample space for the company to commercialize itself and secure meaningful revenue / profitability. Worth noting that Digit v5 is expected to be commercially rolled out in 2026. So meaningful revenue is expected, speaking of which…

Unit economics

We had to set this section up separately because Agility’s business model relies a lot on unit economics. Before that they have 2 offerings to customers:

1. RaaS (Robot as a Service) - Agility rents Digits to the customers for a fixed annual price of ~$100K which includes maintenance. Since they are replacing a full time employee, payback is essentially immediate. ROI of 2x

   

2. Ownership - Agility sells Digits to the customers for ~$400K that includes a fixed upfront cost and a recurring subscription cost. Payback is projected to be in ~1 year. Over 5 years, customers can expect a labor cost benefit equivalent to ~$1.1M

In both models, agility makes margins through scaling up. Higher the volume, lower the unit cost and higher the margin. At launch, product margins are estimated at approximately 40% (RaaS) and 50% (Ownership). The chart in their investor presentation captures the impact of unit economics clearly

Over the long term, the company aims to reduce the cost of a single Digit to ~$50K at 10,000-unit annual production

The company has announced that it has an order book worth ~$300M. This represents an order book of 1,000 Digit v5 units on a 3-year RaaS contract at $100K/year per unit.

As BOM costs fall toward ~$50K at scale, margins expand significantly. But that cost reduction accrues to Agility, not the customer. The RaaS price stays at $100K/year unless competitive pressure forces it down.

Valuations & Future outlook

Traditional valuation frameworks do not fully apply until revenue is disclosed in the S-4. At $10 per share price, agility ( ) is valued at $2.5B. Even if you triple the price (its almost 2x now), Agility is priced at a significant discount to every humanoid robotics private-market comparable.

• Figure AI raised at a $39B valuation (community/media estimates).

• 1X and Unitree command multi-billion valuations in private markets.

• Tesla’s optimus valuations are hidden under the multiple businesses it runs

Agility is, by this measure, the cheapest entry point into pure-play humanoid robotics anywhere, public or private, by a wide margin.

Right now agility is “pre-revenue SPAC name” trying to one day be “commercial robotics platform with visible ARR.” The full re-rating opportunity is when and whether it achieves this goal. As per the filing, the company has orders for ~1000 units and 30+ customers in the pipeline. Let’s assume, liberally, that they fill their order book with another 1000 digits from those customers in pipeline. So sales pipeline considered is with 2000 digits and 30+ customers for 2027.

Subscribe now

Let’s consider possible scenarios:

1. Bull: 80% conversion of the sales pipeline. RoboFab runs at 1,500+ units/year without supply chain disruption. US reshoring policy accelerates the labour ROI case. No material competitive disruption from Tesla or Figure in the 2026-2027 window.

2. Base: 50% conversion of the sales pipeline. RoboFab runs at 1,000 units/year. v5 ships on time but early deployment friction slows expansion velocity. Figure AI establishes credible automotive wins, narrowing Agility's competitive lead. BOM cost reduction lags the illustrative curve.

3. Bear: 20% conversion of the sales pipeline. RoboFab underused. Post-SPAC underperformance pressures talent retention. Figure AI and Tesla begins external deployments at any scale, introducing pricing pressure anxiety.

   

As we type, the is trading at ~$20. So as you can see, there is no way the stock can make returns based on traditional valuation metrics unless meaningful expansion of their pipeline & manufacturing capability. Having said that, in current market conditions, such companies aren’t traded under the valuation lens. A notable Churchill Capital SPAC, , is a case in point.

6.Competition Analysis

Humanoid space is highly competitive with some of the elite names plying their trade. Each of the players have unique approach, technology solutions and manufacturing strategy. They do have one thing in common, every single one of them: Near non-existent commercial proof

Figure AI is the most immediate commercial threat. Figure 03 is operating at BMW’s South Carolina facility with the Helix AI system enabling verbal task instruction. Figure is better funded (raised at $39B), has automotive manufacturing exposure, and is moving fast. Agility is winning for now. But the gap on commercial traction is narrowing.

Tesla Optimus is the long-term existential question. Tesla has $20B in capex allocated in 2026 and is converting Fremont production lines to Optimus manufacturing. Musk has guided to a $20,000-$30,000 unit price at scale. If Tesla achieves something close to that, the RaaS model’s economics come under pressure from the ownership alternative. No other player in this space has Tesla’s manufacturing depth, supply chain, or capital.

Boston Dynamics Atlas is Hyundai-owned, targets 30,000 units per year by 2028, and exceeds Digit in raw physical capability (50 kg lift, 7.5 foot reach). But Atlas is more general-purpose, more expensive, and does not have Agility’s commercial RaaS track record.

Unitree (China) is relevant primarily as a price disruptor. Chinese government support of approximately $138B in robotics investment over 20 years (per International Federation of Robotics) means Chinese competitors will likely undercut on price.

1X (Norway, OpenAI-backed) is focused on home use, not industrial logistics. Not a direct threat in Agility’s core market.

Where Agility stands: The only company with formally contracted commercial humanoid RaaS deployments, documented tote counts, NRTL safety certification in a live production line, and 30+ prospects in an active sales pipeline. The question is how much of this is durable and converts into commercial sales.

Share

7.Bullish View (Opportunities)

Vast TAM: Digit v5’s cooperative safety unlocks deployment everywhere humans work, expanding TAM from logistics and manufacturing into grocery, hospitality, elder care, and home.

Employment crisis: US manufacturing job openings are projected at 1.9M unfilled by 2033. There is a real incentive for the customers to use the humanoid products.

Value for customers: The 2x customer ROI at near-immediate payback usually shows up on the customers’ bottom line.

Healthy cash position: The $574M post-close balance sheet covers approximately 3 years of runway at current burn.

Cheapest in an overvalued space: At $2.5B entry against Figure AI's $39B private valuation, Agility is the cheapest pure-play humanoid exposure available anywhere.

8.Bearish View (Risks)

Revenue blind spot: Revenue is undisclosed and the S-4 has not been filed. We are investing on operational proof, not financials. The picture could look very different once audited numbers are out.

Single customer concentration: The $300M order is from one unnamed customer, milestone-dependent, and explicitly “not a measure of current revenue.”

Amazon on pause: The most strategically important customer and investor is not currently active, waiting for Digit v5. If that relationship does not reactivate, the commercial story narrows considerably.

Burn acceleration: Opex steps up to an estimated $175M+ in 2026, the sharpest single-year increase in the company’s history. Revenue needs to materialise fast to avoid another capital raise and dilution.

SPAC overhang: Warrant dilution, redemption risk, and a historically poor deSPAC track record weigh on near-term sentiment regardless of fundamentals.

Tesla: With a sub-$30k Optimus price target and manufacturing capital that dwarfs Agility’s entire balance sheet, Tesla is the shadow over every competitor’s unit economics in this space.

Objective Opinion

My days of evaluating made me learn an important communication paradigm about tech opportunities. A great tech doesn’t mean a great company, a great company doesn’t mean a great stock. Objectively speaking:

Are humanoids great tech? Undoubtedly. Western world have a structural population problem where finding sufficient people to fill in for the menial repetitive task is challenging. Even if you manage to get one, it is difficult to manage them. Humanoid solve real problems & the space is surely getting there.

Is Agility a great company? I would say so. As far as pre-revenue companies go, they do have an impressive partner list and a product that has meaningful miles behind it in real world application and use cases.

Is CCXI a great stock? This is where the thesis breaks for me. It’s quite honestly not a great stock simply due to its risks mentioned before. In the future, they might surely become one. With facts known right now, they aren’t there yet.

Things I am watching:

• S-4 filing: First audited financials and real revenue disclosure. The most important document for this investment.

• Digit v5 commercial launch (planned 2026): Does cooperative safety work in live environments?

• Amazon re-activation

• Pipeline conversion: How many of the 30+ prospects convert to commercial deployments?

• Lock-up expiry (180 days post-close): Watch for insider selling pressure.

• ISO 25785-1 certification progress: Regulatory milestone for unguarded humanoid deployment.

Conclusion

Verdict: Speculative buy.
It could be an entry at the ground floor of generational opportunity or typical peak bull market moonshot company that disappears before the ink dries

Agility Robotics is not the most exciting name in humanoid robotics. It does not have Musk’s distribution, Figure’s war chest, or Atlas’s physical capability. What it has: paying customers, the industry’s only formally contracted commercial RaaS deployments, a real manufacturing facility, and a unit economics model that makes sense for both sides.

Let’ s be clear the risks are real. Growing losses. Undisclosed revenue. SPAC mechanics. Tesla’s long-term price pressure. v5 shipping on time. Our own valuation section said there’s no way this can return money soon. These are not footnote concerns.

But so is the opportunity. Labour shortage tailwinds are structural, not cyclical. Strategic partnerships with Amazon, NVIDIA, SoftBank, and Foxconn. $574M in post-close cash. A 2.5x illustrative customer ROI. A $2.5B entry valuation against a $39B private-market competition. And a pipeline of 30+ prospects heading into the most important product launch in the company’s history.

This is not a buy based on valuation. It's a buy based on narrative, timing, and the possibility that Digit v5 converts the pipeline into real ARR. Size accordingly. One simply has to look at , the other Churchill Capital SPAC, to see how powerful the narrative & timing can be.

We do have a small favor to ask…

If you got value out of this article, we would really appreciate if you can spread the word amongst your friends, family or social media. We see it as a huge token of recognition from you. Thanks a lot for your time.

Any questions about this post or other things regarding stock market? Questions about some of our holdings? Let us hear them in the comments!

Leave a comment

Please do restack, share and comment as a show of your approval (or disapproval) of the content

How can it escape this discount? Prove to the market that this lost revenue is multiplied not just replaced. Countdown has been on since April 2019. To the company’s credit, they have done some good things since then: Setting up automotive business, adding muscle to CPU business, expanding edge AI. After initial surge, stock has returned a grand 25% in the past 5 yrs.

Qualcomm then had two things happen to it recently that genuinely changes the landscape of the stock and the way it can trigger a serious rerating: Physical AI and data centre revenue. Lets see how exactly..

Subscribe now

To read my full disclaimer please click here

TL;DR: Company with exposure to automotive, data centre, and IoT, are scaling faster than the market credits. Leadership in edge AI necessary for robotics. PE Multiple of 19x is too cheap for what this company is becoming.

1.Brief History

QUALity COMMunication was founded in 1985 in San Diego by Irwin Jacobs and six colleagues. Its early bet on CDMA (a cellular transmission standard) turned it into the essential infrastructure of mobile communications. By the 3G and 4G eras, Qualcomm held foundational patents that every smartphone manufacturer had to license, and it built the Snapdragon chip line to dominate premium Android.

The 5G transition accelerated both businesses. Today it is navigating its third major pivot: from mobile-first to a compute platform spanning cars, data centres, and intelligent devices. They still very much are pioneering the communication too through their work in 6G.

2.Product Portfolio / Business Verticals

Qualcomm primarily organizes their vast products into 2 divisions

QCT: Qualcomm Communication Technology

All Qualcomm products ranging from mobile chipset to AI. If one were to sum up the product lineup, it would be: Handsets are from yesterday, Automotive is for stability & Predictability; Edge & IoT and Data centre are the multipliers for tomorrow.

Handsets: $27.6B annualised.

• The original engine, now in transition. Snapdragon series of mobile chips powers the Samsung Galaxy, Xiaomi and many more premium handsets. Market share in the above-$500 Android tier is dominant.

• Qualcomm's Snapdragon 8 Elite consistently leads on AI inference performance and camera processing.

Sales to Apple (or lack thereof) comes under this division. The steep fall is already part of guidance and is largely priced in

Automotive: $4.8B annualised.

• Automotive business are slow to ramp but creates very deep integration resulting in strong, stable revenue (and margins). After years of work and partnership, Snapdragon ride integrates ADAS, digital cockpit & connectivity onto a single SoC (system-on-chip). One computer to handle all mission critical vehicle operations.

• Snapdragon ride requires platform level integration which mean even higher switching costs. BMW’s 2026, Mercedes, GM, VW, Li Auto, and Leap Motor are all on the platform. The design-win pipeline was raised to $65 billion at Investor Day. 80% of the next four years’ revenue is already contracted.

The $65B pipeline currently under contract ships in 2027-2031. This is contracted backlog, not speculative revenue. Target: $10 billion by FY2029.

IoT and Edge AI: $6.8B annualised.

• AI PCs. The Snapdragon X Elite is the silicon inside Microsoft's Copilot+ PC programme, purpose-built for on-device AI inference via its Hexagon NPU (Neural Processing Unit: the chip block dedicated to running AI models locally, without a cloud connection)

• Industrial edge: smart factory sensors, networking chips, AR/XR headsets, and robotics platforms. These are bundled together in Qualcomm's reporting but represent distinct growth curves. Industrial automation sub-segment is already at commercial scale, providing a floor under the IoT revenue line even before AI PC volumes ramp.

  Share Glocal Investor

• Qualcomm's RB5 and QCS platforms power warehouse automation robots (deployed at Amazon and DHL facilities), smart cameras for defect detection in manufacturing line. NEURA Robotics and Qualcomm recently formed a long-term strategic collaboration to advance next-gen robotics & physical AI and to work jointly on “Brain + Nervous System” reference architectures.

Management targets $14B+ for the combined IoT segment by FY2029.

Data Centre: Huge Potential recently launched

The Dragonfly portfolio was launched in 2026. They are taking the industry challenges head on right from reducing CUDA dependence to offering memory solution to reduce HBM (Memory) dependence, they have gone all guns blazing.

• Qualcomm launched Dragonfly C1000 CPU, High Bandwidth Compute (HBC), Dragonfly AI300 inference accelerator, and connectivity products, together with custom silicon solutions, all engineered to maximize performance per watt and token throughput at lower total cost of ownership

• C1000 CPU: Has > 2 TB/s leading-edge PCIe Gen 7 connectivity, plus CXL connectivity. CXL is a high-speed interconnect standard allowing CPUs and accelerators to share memory efficiently, reducing latency in AI workloads. It enables higher CPU performance at lower power and cost. Commercial sampling is expected in 2028

• High Bandwidth Compute (HBC): Purpose-built near-memory computing architecture that bonds compute with highly-accelerated memory bandwidth in a 3D-stacked silicon solution to address AI’s fundamental data movement bottleneck. Aims to offer better TCO & energy efficiency than HBM. Designed to enable a 6x increase in bandwidth per watt versus HBM normalized at card level and 200x increase in capacity per watt versus SRAM normalized at rack-level. Commercial sampling is expected in mid 2027

• AI300 (Card and Rack) Third-generation, air- and direct-liquid-cooled rack-level AI inference platform. Enables industry-leading memory capacity and effective bandwidth enabling high-throughput, low-latency performance for LLM, LMM inference and agentic AI workloads. Expecting 4x-8x better performance-per-watt compared to existing GPU-based architectures on memory bandwidth per watt per card. Commercial sampling is expected in 2028

Hero of the recently concluded investor day. Revenue target: $5 billion by FY2027, $15 billion by FY2029.

QTL: Qualcomm Technology Licensing

Every 5G device manufactured globally, regardless of whose chip is inside, pays Qualcomm a royalty under this segment. It is the highest-margin, most durable business in the portfolio. The revenue guarantee is equivalent to a bond. 70%+ margins with zero incremental cost.

Qualcomm charges ~3.25% of the selling price of a 5G device, applied against a blended ASP cap. $1,000 flagship generates approximately $13 per unit in QTL revenue. With 600 million+ 5G handsets shipping annually and global 5G penetration still below 50% of the installed base, QTL has a structural tailwind simply from the 4G-to-5G upgrade cycle in India, Southeast Asia, and MENA playing out through 2028.

Share

3.USP

Their Moat can be bucketed into 4 tiers:

• Edge AI leadership: AI models are moving from clouds and data centers to the downstream devices in the form of drones, robots, cars etc. AI systems perceive, reason, and act in the physical world in real time. These systems cannot tolerate cloud latency. They need local inference, low power consumption, and connectivity. Edge AI chips offers all that. Decades of experience in edge AI inference chips in mobile phones becomes crucial in physical AI world. The Hexagon NPU in every Snapdragon has processed more on-device AI queries than any chip family outside NVIDIA’s data centers. It forms the base of dragonfly portfolio. TAM : $1.7T by 2030.

• Vast Patent portfolio: They have numerous essential patents without which it is impossible to ship a mobile phone. As long as 3G, 4G and 5G is in existence, QCOM makes money off of it through their ~140K patents. The fact they have announced partnership for 6G by 2029 suggests this is to last longer.

  

• Single architecture across the compute continuum. The Oryon CPU core in the Dragonfly C1000 server is the same architecture in the Snapdragon X Elite AI PC and Snapdragon Ride automotive platform. Qualcomm is not rebuilding from scratch for each market. Shared R&D across phone, car, robot, and server is a structural cost advantage competitors cannot easily replicate.

• The fabless model. Qualcomm designs chips, TSMC manufactures them. Capital intensity stays off Qualcomm’s books. That is why FCF (free cash flow, cash remaining after operations and investment) margins run at 29%.

4.Corporate Governance

CEO Cristiano Amon has run the company since 2021 and hit every major diversification target. Capital allocation is disciplined: $5.4 billion in buybacks in H1 FY2026 alone, $20 billion authorisation announced in Q2. Stable dividend. Net debt of $5.5 billion is manageable relative to $12.8 billion in annual FCF. No governance red flags.

5.Financial Analyses

Fundamentals

Revenue headlines has been almost flat growing at a ~6% CAGR since 2021. The composition underneath it is rotating fast, with automotive delivering the growth the handset segment is losing to Apple. What matters is whether the rotation completes before the Apple headwind peaks. The contracted backlog means FY2027 and FY2028 automotive revenues are already largely locked in regardless of what the macro does.

The R&D line is the other thing I’d call out. At 21% of revenues, it is the visible cost of building three revenue engines simultaneously. That investment shows up in the income statement. It is a drag today but a multiplier later.

Net margin of Sep '25 cratered to -27.7% and Mar '26 spiked to 69.5%. The Sep '25 drop is the $7.1B one-time tax provision from IP repatriation, concentrated in Q4 FY2025. The Mar '26 spike is the offsetting side. When a company does a major IP restructuring or repatriation, the initial charge typically creates deferred tax assets on the balance sheet. Those deferred tax assets get recognised in subsequent quarters as tax benefits.

Share Glocal Investor

Valuation

Read more

Share

To read my full disclaimer please click here

Stocks

I run a concentrated portfolio across EUR and USD-listed equities, spread across many themes I have written about here: semiconductors, AI infrastructure, robotics, digital platforms, and a few bets I haven’t spoken about yet and am willing to be patient on.

Read more

Subscribe now

Iran did not try to meet the US Navy’s carrier groups head-on. They used the oldest, cheapest & most effective weapon available in a narrow strait: sea mines. Seeded across the Strait’s shipping lanes, they denied something that trillion-dollar air defense networks, stealth aircraft, and precision cruise missiles all needed. A passable waterway. Clearing them requires getting close to the seafloor, systematically, with high-resolution sonar capable of distinguishing a mine from a rock at speed.

Enter Kraken Robotics, a company most people have never heard of, suddenly sitting at the center of one of the most urgent capability gaps in modern naval warfare.

To read my full disclaimer please click here

TL;DR: Kraken is a marine tech company with a differentiated position in seabed imagery and subsea batteries. Recent acquisition of Covelya Group triples revenue and broadens the tech stack further. Undeniable potential & priced for it.

1.Brief History

Karl Kenny founded Kraken in 2012 with one goal: Deliver sharp seabed images using advanced sonar technique at fraction of cost. Synthetic aperture sonar (SAS), a technique, stitches overlapping sound-wave returns to produce seabed images sharper than conventional sonar.

Share

They used to exist only in expensive military programs. Kraken managed to deliver it compact & affordable for smaller unmanned vehicles. The company listed on the TSX Venture Exchange as Kraken Sonar Inc. on February 24, 2015, renamed to Kraken Robotics Inc. in September 2017 as its product line broadened.

Greg Reid, former CFO, became President and CEO in January 2023. Kenny passed in February 2025. Reid now leads the company through its most consequential period: the ~$450M acquisition of Covelya Group, announced March 3, 2026, closing July 2.

2.Product Portfolio / Business Verticals

Under water, GPS does not work. Radio signals do not penetrate seawater. A submarine, a pipeline, or a sea mine sitting on the seafloor is essentially invisible unless you get a sensor close to it and use sound. Most of the things Kraken builds is to solve this constraint.

Kraken's standalone business runs two segments: Products and Services around its products.

Products

SAS: Synthetic Aperture Sonar: Flagship product of the company. Conventional sonar sends a single ping and reads the echo. SAS works like a panoramic camera. It keeps pinging signals and collects echoes as the device traverses in water. It renders multiple echoes to form cheap, sharp seabed image. SAS maps intricate details of undersea bed like fractures in pipelines, material composition, mines etc. Kraken made it compact and affordable for unmanned vehicles. It costs a fraction of what navies previously needed.

KATFISH - Towed SAS: That sonar sensor is integrated into KATFISH, a torpedo-shaped towed body pulled behind an uncrewed surface vessel at high speed. It’s a faster seabed version of erstwhile Google maps vehicle in roads. It sweeps large areas quickly for use cases like mine-hunting, seabed survey etc.

SeaPower: Pressure-tolerant lithium-ion battery systems. At deep sea depths, pressure is unimaginable (remember Titan sub implosion?) Normal batteries need a bulky steel housing to survive it. Kraken's seapower work without that, recovering weight and volume a vehicle can use for other payloads. SeaPower is sold not just to Kraken's customers but to competitors too, including Teledyne Marine, one of the largest names in ocean technology.

Services

Built via the 2025 acquisition of 3D at Depth Inc.: underwater LiDAR inspection, sub-bottom imaging, and offshore energy client work. Underwater LiDAR gives a precise 3D model of a specific structure: turbine foundation, deep sea well heads etc.

Read more

Subscribe now

The success (or lack thereof) of Google Earth in that exact form validated that it was, at that point in time, ahead of its time. Ultima…

Read more

To read my full disclaimer please click here

Subscribe now

First things first, what did the numbers say:

Q3 Results
• Revenue $41.5B vs Est. $35.…

Read more

Now we get to the fun part…

Read more

Read more

To read my full disclaimer please click here

The Platfor…

Read more

Vertiv is a Westerville, Ohio-based supplier…

Read more

To read my full disclaimer please click here

Read more

Winding the clock back by a decade, I was in Malaysia for vacation. As soon as I landed, wherever I saw it was that now famous bright fluorescent banners screaming “GRAB” and by when I came to airport entrance, I had the app install…

Read more

• Revenue decreased by 32% year-over-year.

• Adjusted EBIT declined by 68% year-over-year.

• Orders on hand were 26% lower compared to the previous year.

Results were announced on April 30th the stock doubled a month later, it is down today at the time of wr…

Read more

My satirical “How-to” guide based on how this was used to deadly effect. Here we go:

Shouldn’t the s…

Read more

Amongst all that there are 2 themes right now that is significantly

There are 2 major them…

Read more

As someone from an emerging market, I know the value of convenience. World around you is so chaotic that you just want convenience. Right from online delivery to quick commerce (order something - potatoes to phones get it in under 30 minutes). You don’t even …

Read more

Physical AI is not one stock. It is a stack : sensing, compute, network, precision mechanics, power, humanoids, assemblers, toolchains. Each layer has its own moat, its own economics, its own risk profile.

Humanoids are just one application within Physical AI gambit…

Read more