Most people have never heard of this Singaporean fund. Here is the link to check them out

Four of its top ten — $MKSI, $TXN, $SIMO, $MOD — are the names to watch because our estimated Average prices are still

Just in the last year, those were their returns.

The last four quarters: +46%, +24%, +18%, +47%. Compounded: +212% in twelve months, against 21% for the S&P. Woah?

Disclaimer: I used Stockdrifts.io to filter and find and look for these investors. Best part is that you can save them for future monitoring. Institutional product to be able to give this entire 13F dataset corpous to your Claude or Codex is coming soon.

I am running a promotion on the Ultra tier for 15% off
Code is “15-13F”

Who are these people?

Founded 2009 in Singapore by Matt Hu and John Wu. A long/short equity fund plus a VC arm, and their stated hunting ground is technology supply chains.

Bloomberg profiled them in April: no pod-shop model, Hu keeps a close grip on trading decisions, analysts work in “monastic cells”.

They cashed out in Q1, then shoved back in

The reason they outperformed so well is that they were quietly cashing out in Q1 2026. Reported long value was cut 59% in one quarter, while the positions they kept were up 18%.

They sold really hard in the downturn.

Then in Q2. They closed 18 positions — Teradyne, ON Holding, Lumentum, L3Harris, ASML, Broadcom, Microsoft, gone — opened 5, and rebuilt the book to $1.05B with 91% in the top ten.

Poker translation: they pulled most of their chips off the table in the Q1 selloff, then pushed back in with 18 hands. Concentrated As F—-.

The book is mostly AI-plumbing ETF

$MKSI 24.5%, $TXN 16.3%, $CRDO 16.0%, $HWM 8.0%, $TSM 7.5%, $SIMO 5.7%, $MOD 4.7%. Chemistry and drills for AI boards, analog chips, SSD controllers, data-center chillers.

$NVDA is 0.6% of this book. $AMD is 1.6%. The weight sits one or two layers below the GPU — where the shortages are.

Pam “they’re the same picture” — FengHe’s top 10 vs the AI supply chain

This meme is akin to the situation.

A year ago the same book was Astera Labs 17.5%, AMD 13.3%, Marvell 8.6%. They sold the first-derivative AI names and rotated down the stack.

Okay now that we have unpacked their strategy and Prowess as an investor, here are the four names worth your time

$MKSI — the chemistry and the drill, 24.5% of the book

MKS is their biggest bet, first bought in Q1 and topped up 27% in Q2 at an average $223. Vacuum gear for etch tools, PCB drilling gear, and the chemistry for AI boards.

The Q2 call was loud. Semi segment growing over 50% year-over-year in Q3, depth etch running well above that, AI chemistry now 15–20% of chemistry revenue from 5% in 2024. Chemistry equipment used to top out at $200M a year; this year is “significantly higher”, with visibility through 2027 and a new Guangzhou factory committed.

Just look at the Acceleration in China and South Korea’s revenue

Punchline: whoever wins the GPU race, the board still needs holes and chemistry. I like the acceleration in revenue and operating leverage

Just look at the acceleration on this revenue streams of Photonics and Vaccuum division

Definitely a good buy under 200, but also interesting to start a position here for a medium to long term hold

The drag is that factory — 50 to 80 bps of gross margin per quarter for “the next couple at least”, charged straight to COGS

$TXN — the boring toll booth with empty clean rooms, 16.3%

A brand-new 16% position in Texas Instruments. Analog is the least sexy corner of AI, and that’s the point.

Haviv Ilan called demand “very strong and broad”, spreading from industrial and data center into automotive, and TI’s data-center business is “doubling more or less year-to-date”. Competitors are quoting 52-week lead times. TI has clean rooms already built and capacity “good for the next 3 years”.

Super-nerd stat: TI’s own lead times moved up “maybe a couple of weeks”. When the neighbour’s queue is a year long and yours barely moved, you take their customers.

$SIMO — the traffic cop inside every SSD, 5.7%

Silicon Motion makes the controllers that run NAND flash. It doesn’t own the NAND, and the NAND makers are busy: they “increasingly rely on third-party controller while focusing their own resources on DRAM HBM solutions”. That’s the trade.

NAND is scarce and management expects it to stay scarce until 2028 when new fabs land. Meanwhile the enterprise mix exploded: Ferri and Boot Drive went from 4% of revenue to “near 30%” in a year. MonTitan enterprise SSDs are in production with 2 tier-1 customers, 5 more ramp in H2, and a 4nm PCIe Gen 6 controller tapes out in August with design wins at flash makers and cloud providers.

2026 revenue guided up more than 100%. Q3 guide $519–541M, operating margin 27.5–28.5%, gross margin already beating guidance at 50.2%. Naice. Stock is up 223% year-to-date and FengHe still opened it at $219.

$MOD — the AC guy for GPU racks, and the one they’re underwater on, 4.7%

Modine builds chillers and cooling for data centers. Data-center revenue grew 90% year-over-year, an 80%+ CAGR over three years, third straight quarter of record orders, and a long-term agreement with a large hyperscaler running out to 2029.

FengHe tripled the position in Q2 (+197%) at an average $215. Stock is $180 now — the only one of the four below their cost.

Why? Parts shortages cost the new sites downtime and margin, the head of the data-center business resigned “for unexpected personal reasons”, and the CEO is running it himself while the Gentherm spin-merger closes by year-end. Three real problems. But management says the shortages don’t touch the long-term agreements or this year’s targets.

CONCLUSION

Here is the fund link to check them out
FENG HE

Bull case:

• $MKSI: semi segment +50% YoY in Q3, visibility through 2027, biggest position of a book that did 3x in a year.

• $TXN: competitors at 52-week lead times, TI with 3 years of built capacity — share gains without capex.

• $SIMO: revenue doubling in 2026, NAND scarcity to 2028, 7 tier-1 enterprise SSD customers by year-end.

• $MOD: 80%+ data-center CAGR, three record order quarters, a hyperscaler deal to 2029, 16% below FengHe’s cost.

Bear case:

• $MKSI: 50–80 bps of gross margin a quarter burned on Guangzhou for at least two more quarters.

• $TXN: $233B and already up 76% this year; the share gains reverse when competitors’ fabs catch up.

• $SIMO: up 223% YTD; NAND supply normalizing early kills the mix story.

• $MOD: no permanent data-center boss, plants running below plan, a spin-merger to digest.

Risks:

1. 13F lag. These positions were filed 38 days after quarter-end and are 65 days old as I write. FengHe closed 18 of 31 names last quarter.

2. Concentration cuts both ways. A 24.5% position means one bad MKSI print moves the whole book.

3. All four are AI-capex derivatives. If hyperscaler capex guides down, they fall together.

I am not invested in $MKSI, $TXN, $SIMO and $MOD, but I do keep them on my watchlist in case there is carnage in the markets

If you wish you can read the original thesis here. They are releasing their Q2 Earnings tomorrow.

But then it gave it all back — 75% off the $17.43 high, trading currently at $5.65

And now my DMs all ask the same thing: is the thesis dead?

I sat down with the management Rhone Resch (CSO) on a recorded 42-minute call to walk through all of it. Here’s the full honest picture — the good, the hedged, and the dangerous risks

It is the only Solar business that reports Positive EBITDA and Positive Net income (±100M of EBITDA in 2025 and guided for $90-100M of Net income in 2026) without 45X credits

There are a lot of Policy related unknowns that will need answering

Here is a refresher. The company has been experiencing a massive growth

Q1 2026: $142.8M revenue, up a massive 177% YoY. Gross profit $47.8M — a 33.5% margin!! Net income $28.4M.

One quarter. A year earlier, the same quarter was a $3.7M loss.

Have a look again.

Market cap: $240M. That’s ±2.4x current P/E, for a company that’s growing massively. Value-stock pricing on a growth-stock income statement.

Q1 operating cash flow was +$33.4M. Customers have prepaid $133M in deposits for future modules.

So why 2.4x Only? Clearly there are big risks. I will explore further, but here they are in short

1. AD/CVD circumvention on Ethiopia. That Ethiopia’s duty-free status doesn’t survive

2. §337 Patent issue that FSLR sued all of the industry naming TOYO as an infringer. TE/T1 just recently spent 135M acquiring Singapore-based Evervolt Green Energy Holding for Topcon Patents.

3. ⁠Texas funding gap. ±$310M funding gap exists for the HJT facility that they are trying to resurrect in the US

4. Section 232, how much benefits can they claim?
   

What management told me on the call

Four things stood out. I’m quoting precisely, because precision is the whole point here.

1. On the 45X tax credits - up-to-$0.07/watt subsidy for US-made modules. They are generating credits at the Houston plant. They have booked nothing and sold nothing. A third-party legal opinion is in hand for 2025, another in progress for 2026. When I pushed — do the credits come for sure? — the answer was:

“We’re not going to say for sure... we only recognize them on the P&L once we actually either sold them or have a verification from the IRS.”

That’s the conservative answer, and I respect it. But note what it is not: booked money. Buyers now want the credits wrapped in insurance before they pay. An actual executed credit sale would almost certainly be a stock mover event. Completely not priced in for 2026. Although the slides have 45X through 2030 pasted everywhere.

2. On dilution. The $357M Texas cell fab needs funding, and the June raise — $50M at $11, with a full warrant at $13.20 attached — cost the stock 38.8% in a single day. Management’s words on the call:
   
   “Anything equity or equity-linked is not a viable source of financing at this point. The plan is 45X monetization, non-dilutive debt, and operating cash”

Read that carefully. It means “we can’t issue at these prices.” It does not mean “we promise never to.” The plan holds — if the cash keeps flowing.

The management is cognizant that the equity is super undervalued right now and using it to raise cash would be short sightedness.

3. On equipment. The HJT cell tools are already secured. China now restricts solar manufacturing equipment exports to the US, per management. So competitors announcing shiny new cell fabs still have to shop at a store that stopped serving Americans. $TOYO already checked out. You can’t build a fab without the oven.
   

4. They chose to build Cell plant with HJT technology because it is about 3-5% more efficient requiring less land and providing more 45X as well as electrons over it’s lifetime

INSIGHT 1: The real risk date

Forget the patent case for a minute. The existential question is antidumping.

Quick mechanism. The US keeps duties of roughly 271% AD / 124% CVD on Chinese solar cells. $TOYO’s play: US polysilicon → cells made in Ethiopia (hydro power, duty-free) → modules assembled in Texas. Legal, clever, very profitable.

Commerce’s job is to ask one question: Should Ethiopian cells be US duty free?

In May, eight US manufacturers — First Solar and Qcells among them — petitioned Commerce, naming Toyo Solar Manufacturing in Ethiopia directly.

The allegation: The cells are made from Chinese wafers.

On July 17, Commerce opened a formal circumvention inquiry. Ethiopian solar imports to the US went from basically zero to $277M in under a year — the growth we cheered is the petitioners’ Exhibit

Here is the report on federal register

Dates that matter: preliminary determination **December 10, 2026**. Final expected around May 2027. And duties can reach back to the July 17 initiation date — modules landed since then carry contingent liability.

But keep in mind , In 2023 Commerce ran this exact play in Cambodia, Malaysia, Thailand and Vietnam. Found circumvention in all four. But it let exporters certify their way out — if the Solar cells were not made from Chinese-produced wafers. Even wafers made outside China from Chinese polysilicon counted as clean.

So the entire $TOYO debate compresses into one supply-chain question:

Where are the wafers born?

Non-Chinese wafers → $TOYO certifies out, the machine keeps printing, and December becomes a non-event the market is pricing as a funeral. Chinese wafers → the US arb stops until they re-source.

That’s the question you should listen for on the August 19 earnings call.

---

INSIGHT 2: Section 232 pulled an Uno reverse

Everyone feared the tariff hammer. It landed August 6 — and it mostly helps.

This meme is akin to the situation.

The proclamation, effective December 4, sets minimum import prices: $21/kg polysilicon, $100/kg wafers, $0.22/W cells, $0.38/W modules.

That $0.38 floor hits competitors who import modules. $TOYO assembles modules in Texas — no floor on those, while everyone else’s landed cost goes up. A price umbrella, courtesy of the White House.

Better still: the proclamation grants duty-free import volumes “commensurate with investment” while companies build US manufacturing. On my call weeks before the proclamation dropped — management described exactly this offset being negotiated with Commerce. Then it appeared in the official text.

Their DC read was right, in public, with a timestamp. Small thing. Worth respecting. Wonder how much are the incentives since they committed on building in America

---

INSIGHT 3:The patent case is the slow train

First Solar’s Section 337 case names 47 respondents — basically the whole global TOPCon industry, $TOYO and VSUN included. Target date: November 15, 2027, plus a 60-day presidential review.

Three reasons this is not the December monster:

1. Section 337 awards no damages. Worst case is an import exclusion or a settlement royalty — a margin haircut, not a back-bill.

2. It only reaches imports. The Texas cell fab ramps Q1 2028, right when any exclusion order could bite. Cells made in Texas are out of the ITC’s reach.

3. The separate JinkoSolar suit against TOYO and VSUN is already dismissed after a settlement.

The honest downside: TOYO’s IR told me that they own 10 Topcon patents. However, I do not know in which juridisction and to what useful extent. T1 just paid $135M for a Trina TOPCon portfolio partly to have a seat at that table. Painful, survivable — and the HJT pivot (built on long-expired Panasonic-era patents) is the long-term exit from the entire TOPCon patent war.

---

CONCLUSION


Bull case:

1. Wafers certify clean, the AD/CVD threat dies in December, and 3.6x trailing earnings re-rates violently.

2. A 45X credit sale closes — the $357M fab gets funded with zero new shares.

3. The $0.38/W module floor lifts US prices while $TOYO assembles domestically underneath it.

Bear case:

1. Chinese wafers confirmed, affirmative prelim on December 10, retroactive deposits on everything shipped since July 17.

2. Cash flow stalls, the “no equity” stance breaks, and dilution returns at the lows.

3. A TOPCon royalty stacks on top

Risks (concrete, not theatre):

1. December 10 is binary

2. The quarterlies are unaudited and 25% of Q1 revenue was related-party (VSUN).

3. Institutions own very little and the float is tiny since Abalance owns 60% of the float and so moves are violent in both directions

I am invested in $TOYO. I sold most into the May run and cautiously building positions back again.

BOTTOM LINE: I have built half my position again around $5.6-5.8 and will be on the lookout for $5.2 if there is negative reaction

Nothing here is financial advice. No chasing, no FOMOing. This is an extremely risky stock!

I have peppered this article with short videos to help you understand clearly.

In this article I dove deeper with Samuel Dean , A molecular and cell biology professor from University of Warwick, UK to help me cover this industry

Lets get a little intro to the Drug discovery problem before we can understand what does he mean with AI native companies and processes

There is a lot of promise and hype in Pharma/healthcare and AI, we try to separate the marketing vs. reality together

PROBLEM: 10–15 years to get one drug from idea to pharmacy shelf. ~$2.6B per approved drug. with 99% of screened compound do not even reach humans

The whole system was built to say no.

Make thousands of compounds. Throw them at a target. Watch almost all of them do nothing, or kill the cell. Move the few survivors into animals. Then into humans. Watch most of those die too.

It’s the world’s most expensive bouncer: turns away 9,999 people to let one in — and charges full price for every rejection at the door.

Big Pharma built the only factory on Earth whose main product is no. Only Berghain comes close.

The number of possible small molecules sits around 10 to the power of 60 (Essentially infinite)— a combination lock so long that guessing one click per second outlasts the age of the universe.

And chemists were doing exactly that. Turning the dial by hand, a few thousand tries per program, hoping to hit the code.

AI models do it completely differently. They study the shape of chemical space, predict which corners of it stick to a target, and then write molecules that never existed in nature — tuned for blood-brain-barrier crossing, oral absorption and low toxicity all at the same time.

Allright lets jump into the important insights

INSIGHT 1
Molecules vs. Antibody drugs

Antibodies are another important class of drugs that go through the same process and are very important for certain types of diseases.

However, they are big molecules

And big means they can’t get everywhere.

Throw a great antibody at a tumour and it strips the outer layers — but it can’t reach the center. It’s a size problem, not a toxicity problem.

If you can’t get there, you cannot fix it

Small molecules are tiny and precise. They reach the center of the tumor.

Second point, the chemical space is effectively infinite — there are no running out of new molecules to design. Ever.

If you had to master one drug class, you’d pick small molecules.

Antibodies have their own advantages like having less toxicity or off targets effects. ABCL focuses on antibodies

Lets tackle it later

INSIGHT 2
No AI-discovered drug has FDA approval YET.

The furthest any have progressed is Phase 2. That’s a useful sobriety check against the drug discovery-side hype — the sector is still in “prove it,” not “proven.” phase

INSIGHT 3
The model is basically free now — testing is the bottleneck

The AlphaFold (AF) jump was architectural, not just “more data.”

AF1/2 leaned on training-set correlation (co-evolving amino acids, deep sequence alignments).

AF3 is diffusion-based — it reconstructs 3D structure from a randomized atom cloud the way a language model predicts the next token, and this keys into the actual mechanism of folding rather than memorized pattern-matching.

That’s why it works on inputs/proteins “it’s got no right to work on.” Things that aren’t in the training set at all. It’s a step change. Woah seriously what? This single unlock is arguably what re-rated the whole sector.

So every bro who told us AI can only regurgitate yesterday's knowledge can sit down.

Here is a video with the explanation

Pay Attention to Sam’s last sentence. AF3 is not limited to proteins, it can be extended to small molecules, nucleotides like RNA etc.

Now, what the ramifications of AF3 is that it allows you to have in Silico screens (Screening for hundreds of thousands of potential candidates on your laptop) and arrive at a suitable candidate quicker

INSIGHT 4

The winning architecture is a closed loop, not a smarter model.

Propose in Silico (Your laptop) → then test in an automated wet lab → feed results back → redesign.

Sam is explicit that this loop — tight, fast, ideally robotic and human-light — is what compounds an edge over time, more than any single AI model’s raw quality. He explicitly sees this trending toward humans being largely out of the loop within a few years.

These structural predictions are great. But you fundamentally need to test these predictions in a lab. And here is the Biggest insight in this article

”Let AI design, plan and conduct its own experiments, sort of like Tool calling” in order to reach the drug discovery goal you gave them”. Hear the insight directly below

I think that AI models can design new experimental protocols and receive feedback, achieving about 80-90% zero-shot success on expression protocols, indicating strong autonomous scientific reasoning.The same intuition is now being applied by many startups like Chai Discovery (Raised $400M) or Lila AI (Raised $550M)

INSIGHT 5

Toxicity/off-target effects — not efficacy — is the industry’s real gatekeeper. Drug dropout comes down to two questions:

Does it work, and is it toxic.

AI has made real progress on the first; almost none on the second. The ideal fix — screening a candidate in silico against every other protein in the body to catch unintended interactions — is “computationally hard” and still mostly aspirational, not solved.

This is the reason AI is limited to discovery for now. There are simulations, but its still a few years away. I am writing a part 2 on this for beneficiaries of this particular insight.

INSIGHT 6

A “design-plus-royalty” business model is emerging, borrowed from chip design. Get paid up front for design/service work, then take a profit-share/royalty if what you designed succeeds downstream (echoes how chip design houses get paid at tape-out, then earn more on yield). I think that Drug discovery process is exactly the same process as ASIC chip design. I will expand on this in PART 2

RISKS

• AI can design the molecule, but AI cannot make the FDA skip the part where you prove its safe and it works, - in actual humans over actual years.

• AI helps by raising the odds that the molecule survives the test by dramatically dropping the odds of dropout rates.

• We might be early in industry really appreciating these insights

Before we move on, special shoutout to Sam for helping me reach these insights. He has just started a substack and here is his first post.

Samuel

Twins that aren’t

If you screen for AI-driven antibody companies, two names come up next to each other again and again: Absci (ABSI) and AbCellera (ABCL). Both discover and develop therapeutic antibodies. Both put artificial intelligence at the centre of their story. Both went public during COVID, both partner with big pharma, both are now pushing their own drugs into th…

Read more

a month ago · Mechanism Matters

Okay so now that we understand the basic insights, lets cover the two stocks that are poised to benefit from all this. Unfortunately a lot of the value creation is happening in private with VC money, however there are some investing options available to us

ABSI ($1.6B Market cap)

They got a great first slide. Summarizing our insights

• Absci is the first validated platform to de novo design full length antibodies against ‘zero-prior’ epitopes with atomic accuracy

• I am particularly excited about their cure for Endometriosis. 10% of women worldwide face this disease and there is simply no cure. Bryan Johnson’s partner/founder Kate recently announced that she got it

INSILICO ($3696 Hong Kong based ±$2B Market cap)

• AI-first, closed-loop pharma — not a bolt-on: PandaOmics (AI target-finder) finds the disease target, generative AI designs the molecule, then an in-house automated robot lab tests and iterates. Full stack, no legacy.

• Real clinical signal, not noise — their pulmonary fibrosis drug reverses the disease indicator in Chinese trials, while the two existing approved drugs only slow it. Sam said that the odds of that being a false signal are “vanishingly low.”

  

• Big catalyst incoming — the larger US FDA trial readout is due ~summer 2026. The important thing to know is that Chinese phase 2 was already found to be successful. No reason why US FDA would fail.

• ~$2B market cap (HK-listed, ~572M shares) is arguably low if this scales. It is not hard to be a $50-$100B company.

  

• Especially because they are allowing renting their AI wet labs and will keep a revenue share. So any drug that works on their platform, will have to pay them royalties
  

• Founders delayed their stock unlocks by a year so summer 2027
  

• Backed by Eli Lilly. They already got ~$115M upfront payment with a chance to get to $2.75B

  

• Cayman-incorporated, non-Chinese CEO but Chinese leadership elsewhere, which may be why the US listing didn’t happen — still an open research thread.

So watch the numbers that count. Insilico has spun out 200-plus molecules per program, several already at IND, in 12 to 18 months from target to candidate — against the old industry clock of 4 to 6 years.

The age of expensive failure is closing, and the age of designed medicine is opening. That’s the trade.

Stay Tuned for Part 2

Ameresco is the on-site power integrator and it printed a record $1.8B of awards. $1.2B of that is data centers. Five projects, more than 1GW of behind-the-meter power, with hyperscalers and neoclouds in the deals!!

And buried in the project description is an important insight for $FLNC holders: every single one of those power stacks ships with a battery in it. Recips, turbines, fuel cells — and BESS, glued together as a microgrid!

Here is the direct quote from today’s call

“The solutions we are providing include a combination of reciprocating engines, gas turbines, fuel cells, battery energy storage systems, and integrated microgrids, designed to deliver the reliability required by today’s most demanding data center customers.”

They are clearly saying that there will be a combination of power behind the meter for a lot of reasons. Here is a quote again to understand why Batteries are important

“One is for the resiliency for when you're doing maintenance or upsets. The other side of that is really just to stabilize the load shift from the varying, especially with the AI load profile. It would be combination.”

And then lastly the important question is how long duration are their Data centre customers going for Batteries.

It’s 2 hours.

There was a lot of FUD that FLNC will not be able to service Data centre demands as they are not long duration.

Risks: $FLNC runs 11–13% gross margins with ASPs down 50% in two years — volume must keep outrunning price. And this market punishes timing slips: Vertiv grew 24%, raised guidance, missed revenue by 3% — and fell 17%

Since last year, Chinese models usage are on an upward march. Peak week: 46%. A year ago their share was 4.5%. And is currently higher than American models. Let that Sink In!!

Not because American companies turned pro-Beijing. But because the tokens are frontier levels and 60–90% cheaper

China’s newest export isn’t EVs. It isn’t solar panels or batteries. It’s intelligence — that is priced to kill.

And tomorrow an important company CXMT, China’s DRAM champion, lists in Shanghai after raising ¥66.6B — about $9.8B, the biggest semiconductor IPO the mainland has ever done. Everyone will cover it as a chip story.

Wrong lens.

Cheap tokens and cheap memory come off the same machine — one that’s been running for five years. It eats Chinese household wealth on one end and exports deflation out the other.

Lets unpack.

· · ·

INSIGHT 1: The quietest mega-crash in financial history

China’s property market peaked in Q3 2021.

Since then: $18–20T of household wealth has been erased. That’s the BIS number, not a doomer blog. Real prices are down 25% from the peak — below where they stood in 2006.

Two decades of housing gains. Gone.

And it’s not finding a floor. New home prices have now fallen 17 quarters in a row.

Property was the Chinese savings account. So the Chinese consumers have pretty much shut down making expenses

INSIGHT 2: Beijing’s answer was factories, not welfare checks

America’s answer to its housing bust was to reflate the consumer.

Beijing looked at the $19T hole and chose a different religion: build more supply.

Except the printer wasn’t pointed at households. It was pointed at industrial capacity.

The output shows up exactly where you’d expect. China ran a record $1.19T trade surplus in 2025 — while producing 30% of everything manufactured on Earth.

INSIGHT 3: You’ve already watched this movie. It was called solar.

Solar modules cost over $4/W in 2008. Today: about $0.25. Down 94%.

Now put that curve next to what DeepSeek did to frontier token prices.

Yep!!

INSIGHT 4: The newest export is intelligence — and it’s tariff-proof

Going back to the number in the open. 4.5% to 46% in one year

The mechanism is brutally simple: GLM-5.2 and Kimi-K3 land within spitting distance of the US frontier while running 60–90% cheaper.

If you are a believer of 24X7, Always on AI, this is an important change to pay attention to.

Chinese labs have a bigger share than the American counter parts now.

INSIGHT 5: CXMT watch it for interesting clues

Back to tomorrow’s IPO.

CXMT priced at ¥8.66 — an $85B valuation, the largest mainland chip listing ever. Business Times reckons it could end day one as the biggest listed company in all of China.

Will memory get the solar treatment? Honest answer: I don’t know. AI demand is a different animal

INSIGHT 6: No Lehman. Something slower and meaner

China’s capital account is closed, and so the losses are domestic, held by state banks, socialized in slow motion. A $19T wipeout with no margin call. Nothing like the 2008 Great financial crisis.

The real casualties are industrials, not financial. “As China exports its surplus, Korea, Germany and Japan pay the industrial price” — that’s a January news headline, not my phrasing. By March, the EU Parliament had commissioned a formal overcapacity study. By June, the Atlantic Council was writing that Europe “has had enough.”

Disinflation for consumers. Deflation for competitors. Trade wars for everyone.

What am I doing with this?

1. Respect the machine. Anything where the Chinese marginal producer sets the price — commodity chemicals, mass-market EVs, legacy solar — is structurally short. The competitor’s margin IS the export

2. Own the cost-curve consumers. Cheap tokens subsidize the application layer the same way cheap panels subsidized solar developers. Falling input prices re-rate whoever BUYS intelligence at scale

3. Watch what CXMT does with the war chest, not the day-one pop. Capacity announcements over the next year tell you if memory gets the solar treatment

4. Watch out for the Open source beneficiaries

5. Staying away from ORCL since, they bet on Open AI and I see value that accrued on the frontier labs are starting to shift

6. America is trying to inflate away its debt by exporting inflation while China is exporting it’s deflation. Truly the war of the titans!
   
   

US labs have no choice but to cut prices! This is why Fable was pretending to leave the Stable (Subscription) , but it could never do so


P.S.

• This article is a premise to 3 insightful articles that I will publish in a flurry. So stay tuned

• I don’t intend to buy any positions in CXMT

Perplexity, Claude, Chatgpt all fail at this task because they got so much data that is high noise.

I will be sending more industry analysis in the coming days using this agent. For our Enterprise clients, click through auditability on transcript quotes was important, so we built it.

We are rolling this out to our Enterprise customers, so if you would like a demo, please write/DM me

DISCLAIMER: Beyond this point, the email is completely agent generated. But is incredibly insightful on finding opportunities

QUERY: What do you think are the next emerging themes/niches and are guiding for massive growth but the stock’s still yet to see that in action. Focus on AI beneficiaries

Across a deep scan of small-to-mega-cap earnings transcripts, several distinct AI-beneficiary niches stand out where management teams are describing genuine demand, raising guidance, or building entirely new product categories — yet the stock hasn’t followed. Below, I group them by theme, cite exactly what executives said, and show the performance gap.

1. AI’s “Trust Layer” — Governance, Security & Recovery for Autonomous Agents

This is perhaps the cleanest new IT spending category that barely existed 12 months ago. As enterprises deploy AI agents, the conversation has pivoted from “what can AI do” to “can I trust, govern, and recover when it goes wrong.”

AvePoint (AVPT) — CEO Tianyi Jiang, FY2026 Q1 (May 8, 2026):

“The conversation has pivoted away from productivity and towards something far more important, enterprise trust in this new enormously powerful technology… as AI agents operate more autonomously across enterprise productivity apps, companies truly need a trust layer so that they can scale AI adoption without losing control of data security, privacy and compliance.” — Tianyi Jiang, CEO · AVPT FY2026 Q1

AvePoint posted its 12th straight quarter of double-digit organic net new ARR growth and was validated by Gartner as superior to Microsoft’s native Agent 365. The company launched specific products to monitor AI agent behavior (Agent Plus), define risk for agents, and auto-remediate issues.

N-able (NABL) — CEO John Pagliuca, FY2026 Q1 (May 8, 2026):

“As we look ahead to a world with agents owning more workloads for businesses, the possibility of agents making costly mistakes also rises. We see the need to effectively undo agent mistakes and restore operations through a clean prior state as a potential demand catalyst.” — John Pagliuca, CEO · NABL FY2026 Q1

N-able is positioning its security platform as the “control plane to govern and secure agents.” ARR hit $548M (11% growth), with 106% net revenue retention.

2. AI-Powered Agent Orchestration & Analytics Platforms

Beyond the trust layer, there’s an emerging category of platforms that coordinate AI agents, build AI apps from natural language, and run real production workloads — not just dashboards.

Domo (DOMO) — CEO Joshua James, FY2026 Q4 (March 10, 2026):

“The next wave of enterprise AI will be less about models and more about coordinating data decisions and workflows… our platform doesn’t stop at insight. It unifies data, provides AI-driven intelligence via our AI service layer and with Agent Catalyst enables agentic workflows in a single system.” — Joshua James, CEO · DOMO FY2026 Q4

Domo reported record quarterly billings ($111.2M, +8% YoY), highest gross retention in 3 years (88%), and described 15 production AI agent deployments across restaurant chains, home improvement retailers, pharma companies, financial services, industrial manufacturers, and compliance tech. AI now comes up on ~70% of customer calls. The stock? Dismal.

Unity (U) — CEO Matthew Bromberg, FY2026 Q1 (May 7, 2026):

“Unity is on an incredible trajectory… strategic revenue growth of 35% year-over-year… our Vector revenue in the first quarter of 2026 is 80% larger than 1 year ago, an astonishing result.” — Matthew Bromberg, CEO · U FY2026 Q1

Unity launched Unity AI (an integrated agent that writes code directly into Unity projects) and ingests image pixels, outputs production-ready game assets. 90% of game developers already use AI in workflows, new mobile apps are up 60% YoY, and Unity expects to become GAAP profitable by Q4 2026. Yet the stock has been hammered in the near term.

Amplitude (AMPL) — CEO Spenser Skates, FY2026 Q1 (May 7, 2026) confirmed “a significant increase” in AI agent adoption on the analytics platform, deployed “forward engineers” to help customers integrate AI.

⚠️ Micro-cap risk is significant — especially DOMO at $2.42. These declines may reflect solvency or business model concerns, not just mispricing.

3. AI Infrastructure With a $553 Billion Backlog — The Stock Doesn’t Reflect It

Oracle (ORCL) is the single largest gap between AI narrative and stock I found. On the FY2026 Q3 call (March 10, 2026), CEO Clay Magouyrk:

“Multicloud database revenue grew 531% year over year. AI infrastructure revenue grew 243% year over year. Both also have demand that exceeds supply… This is directly visible in our $553 billion RPO.” — Clay Magouyrk, CEO · ORCL FY2026 Q3

Oracle has secured 10+ gigawatts of power and data capacity over three years (>90% partner-funded), signed $29 billion in new contracts since the prior quarter alone using a bring-your-own-hardware model, and hit 32% AI infrastructure gross margins (above guidance). The company is “constantly raising our FY27 forecast.” The stock: down 6% YTD, down 11% over 1 year ($184.29).

Contrast with Cisco (CSCO) — AI infrastructure orders of $5.3B YTD (4.5x FY25), Acacia optics growing 200%+ YoY, campus refresh multiyear cycle underway. Stock: +57% YTD. The market is very discriminating here — Cisco’s AI story has been bought; Oracle’s hasn’t, despite arguably larger numbers.

4. AI-Driven Drug Discovery — Faster, Cheaper, Fewer Compounds

Recursion Pharmaceuticals (RXRX) is building an end-to-end AI-powered drug discovery engine. From the FY2026 Q1 call (May 6, 2026), R&D Chief Najat Khan:

“We are synthesizing 90% fewer compounds than the industry benchmark… about 330 compounds on average versus 2,500 to 5,000 compounds… while advancing development candidates roughly twice as fast.” — Najat Khan, R&D Chief · RXRX FY2026 Q1

Recursion has 50+ petabytes of proprietary multimodal data, over 10 high-dimensional disease biology maps (half in partnership with Roche/Genentech), and clinical enrollment running 30-60% faster with its AI ClinTech tools. Phase 1 assets are now dosing (REC-4539 in small cell lung cancer). Yet the stock is down 23% YTD, down 33% over 1 year ($3.23). This is a pre-revenue biotech carrying high binary risk, but the AI moat is real.

5. AI in Legal Tech / eDiscovery

CS Disco (LAW) applies AI to legal document review. CEO Eric Friedrichsen, FY2026 Q1 (May 6, 2026):

“Overall, I’m incredibly pleased about the revenue, 14% revenue growth for the business this quarter.” — Eric Friedrichsen, CEO · LAW FY2026 Q1

CFO Aaron Barfoot added that $100K+ customer adds grew 5% sequentially and described it as a “leading indicator” — matters come in, expand, and convert to platform revenue. The Auto Review AI product is bringing in new matters and converting some to managed services. Yet the stock is down 52% YTD ($3.47).

6. AI in Compliance-Heavy Payroll & HR Systems

Asure Software (ASUR) is a small-cap payroll and HR compliance platform where AI is a moat-widener, not a disruptor. CEO Patrick Goepel, FY2026 Q1 (April 30, 2026):

“We move approximately $20 billion annually on behalf of our clients… These are not functions that a generic AI layer can absorb. The regulatory complexity does not go away. In fact, it compounds.” — Patrick Goepel, CEO · ASUR FY2026 Q1

Revenue grew 23% YoY to $42.8M. Organic growth accelerated to 7% (from 3% in Q1 2025). Luna, the AI agent, saw interactions up ~50% QoQ, adopted by >15% of potential users with zero marketing. The AI layer is converting manual compliance workflows into proactive continuous systems, and the new AsureWorks managed service brings 2-3x the revenue of a payroll-only client. Stock: down 9% YTD ($8.27).

7. Edge AI Fiber & Distributed Compute Infrastructure (Early Stage)

Clearfield (CLFD) is a small-cap fiber company pivoting toward edge AI infrastructure. CEO Cheryl Beranek, FY2026 Q2 (May 6, 2026):

“Industry trends continue to support a shift toward compute closer to the end user, as low-latency AI applications require faster processing capabilities… This dynamic will drive the build-out of smaller distributed edge locations that function like compact data centers and require high-density fiber connectivity.” — Cheryl Beranek, CEO · CLFD FY2026 Q2

Backlog rose 39% sequentially (1.3 book-to-bill), the NOVA Platform for edge AI ships in H2, and BABA-compliant fiber extrusion is ready. The stock is up 37% YTD but only +5% over the past year — early recognition of the thesis.

Dycom (DY) — EVP Dan Peyovich on the FY2026 Q4 call (March 4, 2026) described a $20B+ TAM for AI-driven long-haul and middle-mile fiber construction:

“Is it more than $20 billion? We strongly believe that… today, we are getting more phone calls and seeing more opportunities than we saw even a quarter ago or, frankly, even a week ago.” — Dan Peyovich, EVP · DY FY2026 Q4

Dycom’s fiber-for-AI story is back-half weighted, ramping through 2027-2028. Stock is up 31% YTD (+97% 1-year) — partially discovered.

8. Inference Economics & The Token Multiplier

Multiple CEOs flagged the same inflection: agentic AI drives 5x-30x more tokens per task vs. simple chat. This means the inference economy is radically underbuilt.

Digital Realty (DLR) CTO Chris Sharp, FY2026 Q1 (April 23, 2026):

“As agents come to market, it’s a demand multiplier… 5x to 30x more tokens per task… that’s going to really drive another inflection point.” — Chris Sharp, CTO · DLR FY2026 Q1

MARA Holdings (MARA) CEO Frederick Thiel, FY2026 Q1 (May 11, 2026):

“We’re seeing thousand-fold increases in agentic token consumption when somebody moves from chat to using Cowork or Claude Code… Just talk to any CIO and ask what their token bills are lately.” — Frederick Thiel, CEO · MARA FY2026 Q1

DLR is up 21% YTD; MARA is up 43% YTD. The inference hosting/token factory thesis has partial recognition but is still early.

✅ Already Priced In: Themes Where Stocks Have Moved

For clear contrast, these AI beneficiaries have seen massive price action — the narrative is well-understood by the market:

Key Takeaways

• AI Trust & Governance is the freshest unmet theme. No company I found has a clearer story than AVPT and NABL — both explicitly building the security and control infrastructure for autonomous agents, both growing ARR double-digits, both down 20-60% YTD. The risk is that platform giants build native governance, but these companies are pre-integrated and validated.

• Oracle has the biggest narrative-to-price gap of any mega-cap. A $553B RPO, 531% database cloud growth, and $29B signed in three months at improving margins — yet the stock is down YoY. This is a puzzle the earnings calls repeatedly flag: “the market has yet to recognize” is a recurring management theme.

• The AI agent orchestration plays (DOMO, AMPL, U) are heavily beaten down but pivoting hard. Unity’s 35% strategic revenue growth and Vector AI scaling, Domo’s 15 production agent deployments, and Amplitude’s agent adoption surge are real. These names carry execution risk and micro/small-cap survivability concerns.

• AI Drug Discovery (RXRX) and AI Legal (LAW) are longer-duration bets. Both have real AI differentiation (50PB data, 90% fewer compounds; 14% revenue growth, expanding $100K+ customers) but the stocks need catalysts — clinical data readouts for RXRX, sustained revenue reacceleration for LAW.

• The inference token multiplier is a theme everyone is talking about but few are positioned for. DLR’s CTO on 5-30x tokens, MARA’s CEO on thousand-fold increases — the buildout of inference infrastructure (token factories, private AI clouds, edge compute) is a multi-year wave where the capital hasn’t fully deployed yet.

• The market is discriminating sharply. Memory (MU +260%), custom silicon (MRVL +247%), and copper/fiber/cooling (BE, DOCN, VRT) are well-recognized. The next wave of recognition seems most likely in AI governance/trust and in the companies building the software orchestration layer for agentic AI — but those are smaller, riskier names.

Disclosure: I have used excessive AI generated memeS for this article.

• • •

Everyone thinks the battery business is about batteries.

Cells. Chemistry. Gigawatt-hours. Who makes the cheapest box.

They’re looking at the wrong thing.

The actual money is NOT in making the battery. It’s in running it.

Charging when power is cheap, discharging when it’s expensive, and getting paid a third time just for being on standby. Same physical asset, many revenue streams

A well-run battery is a quant trading desk wearing a steel enclosure.

This is the model that quietly produced the best numbers in the entire energy sector last year. Not the panel makers. Not the cell makers. The owner-operators!!

INSIGHT 1: THE BOX IS THE COMMODITY. THE TRADING IS THE BUSINESS.

Battery costs are massively down and there is currently an oversupply in batteries due to China subsidizing this industry for a few years now.

Lets unpack what a merchant battery actually does in a day.

Midday, solar floods the grid. Power prices crater — sometimes negative. (Hello Germany!!) The battery buys. (Yes, sometimes it gets paid to charge.)

Evening, the sun sets, everyone turns on the AC and the EV charger, prices spikes multiple x’es. The battery sells.

That spread — buy at $10, sell at $40-$80 — is the entire game. The battery is long volatility. It doesn’t care if power is cheap or expensive. It cares that the price moves.

I have written about this insight extensively last year here.

And it stacks with other use cases. A battery earns from:

1. Energy arbitrage — buy low, sell high, every single day.

2. Ancillary services — frequency regulation and reserves, where you get paid just to be available to respond in milliseconds. Pure standby income.

3. Capacity payments — utilities pay you to exist so they don’t have to build a peaker plant.

The grid pays for all of it because it physically cannot keep the lights on without something that absorbs solar’s midday glut and releases it at 7pm.

Punchline: you’re not selling electricity. You’re selling optionality

Here is a great illustration

Read more

Demnang2 released a thoughtful piece, breaking down what will drive Edge AI’s success. It contains a key insight to build conviction in our pick.

Now I don’t want to share his work here for free, so please go check out the article here

Damnang’s Substack

Edge AI Investing Guide: Where Capital Goes After Cloud AI

For the past two years, almost all of the capital flowing into AI has gone to the data center. NVIDIA’s quarterly revenue exploded, and HBM and CoWoS owned the headlines every day…

Read more

4 months ago · 6 likes · 3 comments · Damnang

A lot of people will feed this article to their AI and reach similar conclusions.

I am gate keeping the insight/name for subs but will release the articles for everyone soon. Last time WOLF, IPWR, and MX and they ran so fast quickly that we all barely time to digest and position into them

Read more

For today, there is a stock that is completely misunderstood and hasn’t run much yet. You know how tightly I control the quality

Read more

When I picked up the phone and spoke to the IR team of TOYO, I realized something profound. Also realized why Leopold might have bought into TE.

Hours of researching the solar supply chain helped me connect the dots and I now know what the US govt. is incentivized to do.

This article will help you build conviction

So China has complete chokehold on Polysilcon.

You could try to produce Polysilicon elsewhere, but it is 3 times more expensive. Just look at Hemlock/GLW (US), Wacker(Germany) and OCI (Korea) the biggest non Chinese Polysilicon producers.

Like literal 3X

I pointed out in the previous article that Polysilicon (the raw material for Solar) is used and required for Semi conductor chips. 2% of the highest grade is used in semi conductors and 98% in Solar.

That’s the part that finally clicked for me

Solar and Semi are married at the hip

The economics of producing Polysilicon at home, doesn’t work, if you don’t have Solar buying up the 98% of the remaining produce . So in order to have American suppply chain for semi conductors, YOU have to have a business case for the 98% of its product which is solar.

TLDR:

Meaning, If you want to be long SEMI, You have to be long Solar.

I can already see what Trump administration will have to do. They got no choice but to pump American solar industry and continue incentivizing the build out, because they love their chips and want to bring it onshore

With this thesis, CSIQ, ARRY, NXT all warrant more closer look

I have written extensive deep dives on TE, if you haven’t read those yet. Here they are. As always DO NOT CHASE, when it is ignored again, swoop in.

P.S. This is why I wouldnt touch FSLR as they are the biggest US solar manufacturer but uses CdTe thin-film, not silicon. So if the policy goal is propping up the polysilicon→semi corridor, FSLR is outside the thesis

I was writing an in depth article on a Robotics bottleneck play.

Serenity unfortunately just now mentioned it and I am forced to release my half baked article.

If you have money in your IBKR account, you could get in on Sunday late night (Japan’s stock market open) and play along.

I started mentioning this name to my inner circle in March. I will start publishing my early ideas more often. I want you to consume them before the swarms of bees arrive. Maybe in a chat community.

I have a bigger article coming up on how inflation is here to stay and is going to be structural. OPEC talked about loosing 30% of their production capacity. And Cathie is delusional (Again) when she says tech is going to cause disinflation.

Lets begin

HARMONIC DRIVE (6324)

Here is the short thesis

- Every humanoid uses 10–40 strain-wave actuators per unit at $2k–$5k each → arguably the single highest-margin BOM line in the robot. There is literally no other western provider for this except Harmonic drive

A strain wave gear (aka harmonic drive) is the precision reducer that turns a fast, weak motor into the slow, high-torque, zero-backlash motion a robot joint needs.

Three parts: an elliptical wave generator spins inside a flexible steel cup (flexspline) whose teeth mesh, at two points, with a slightly-larger-toothed outer circular spline. Each motor revolution advances the output by only a few teeth → 50–160× reduction in one stage, sub arcminute backlash, very high torque density.

- Micron-level grinding and flexspline metallurgy are the moat.

Only real challenger: Leaderdrive / Green Harmonic (688017.SH), which already supplies Tesla Optimus and owns >60% of China domestic share. Nidec, Sumitomo (cycloidal alt.), and a handful of Korean startups (e.g. SBB Tech) chase.

- Closest analog: ASML in lithography — one Japanese firm holds the precision-machining know-how the rest of the industry needs to scale.

1. Why HDS holds the chokepoint

The thing isn’t really “harmonic drive” the product — it’s strain wave gearing, a category invented by C. Walton

Musser in 1955. That original patent is long expired; HDS’s moat today rests on three layered things:

- Tooth-profile IP. HDS owns the modern profiles that actually work — the “S-tooth” and “IH-tooth” geometries that get ~30% of teeth in continuous engagement (vs. far fewer in involute gears). That is what gives the unit zero backlash, sub-arcsecond repeatability, and high torsional stiffness in a pancake form factor. Patents extend to lubrication sealing, dual-stress separation (US 10,788,114 — Ishikawa 2020), and powder lubricant delivery.

- Process know-how. Flexsplines are thin-wall hardened steel that has to fatigue-cycle hundreds of millions of revolutions. Heat treat, grinding, and surface finish tolerances are deeply tacit — patents don’t tell you how to actually make them survive. This is the harder-to-copy layer.

- Reference designation. In semiconductor lithography, surgical robots, satellites, and high-end SCARA/six-axis arms, HDS is the de-facto spec. Once you’re in a customer’s CAD and qualification dossier, you’re in for a decade.

2. Can they be innovated out?

Partially yes, already happening — but only at the low/mid end:

- Chinese challengers (Suzhou Green Harmonic, Leader Drive) sell harmonic reducers at 40–60% of HDS price and have ~25% global share. Green Harmonic is bringing a 500k-unit/yr Suzhou plant online in 2026 specifically for humanoid demand.

- For commodity humanoid joints, “good enough” precision + cost will erode HDS’s unit share. Tesla’s strategy of dual-sourcing HDS + Green Harmonic is exactly this — qualify the Japanese gear for early bring-up, ramp on Chinese gear when volumes arrive.

- For sub-arcsec / aerospace / semicap / surgical, the manufacturing moat still holds. No credible substitute has emerged — RV/cycloidal (Nabtesco) is heavier and used at the base joints, planetary roller screws are linear-only,and magnetic gears haven’t matured.

So: HDS can be innovated around at low precision but probably not replaced at the high end inside a 5-year horizon.

The real risk is mix shift — they own a smaller piece of a much larger pie.

3. Does Tesla buy from them?

Yes, but not exclusively, and there is no publicly disclosed direct purchase agreement. Industry supply-chain teardowns (China Merchants, 36Kr, futunn) consistently list HDS alongside Suzhou Green Harmonic as Optimus harmonic-reducer suppliers. Tesla Optimus uses 14 rotary actuators, each = frameless torque motor + harmonic drive + sensors. Tesla’s known posture is dual-source; the $685M Sanhua order was for thermal/linear actuators, not harmonic gears.

HDS itself has not disclosed Tesla as a named customer in any IR document I can find — they discuss “robot manufacturers in Japan/China/Europe” generically. So treat the Tesla angle as strongly inferred from supply-chain reporting, not company-confirmed.

RISKS:

No insider buys in the last year. I am scratching my head on this one. Stockdrifts has Japanese and Korean stock insider graph.

—————————————————————————————————————————————————————

A few months ago, I talked about Status quo and the fundamental shift that is required to realize 30-50X increase in Power density within AI data centres.

Let me do a quick recap

Problem:

• A single NVIDIA B200 or AMD MI300X consumes 0.9kW – 1.4 kW. Dont even ask me about Ruben.
• A “pod” of 8 such accelerators + CPUs approaches 12–15 kW
• An AI training cluster is hundreds of pods – multi-megawatt

At 12 V distribution you would need kilo-ampere bus-bars thicker than a wrist; power loss alone would exceed the chips’ TDP

• On top, Every 1 cm² of copper busbar you delete saves you north of $20/kg and a small argument with the procurement team.

Hear about it directly from the Horse’s mouth

NVIDIA’s own whitepaper claims ~45% less copper in an 800 V plant.

So First principles first, whats actually happening?

Here is a nice infographics to follow along the transition. Power distribution is changing from AC into DC microgrids.

Optics is forcing this transition quickly. And now that the retail knows what happened to Optics stocks, they are coming swinging for this sector too.

Here is what you need to understand about the Power Semis

Voltage is the cheat code. Crank it up early and keep it up for as long as possible. Then step it down ONLY inches from silicon, and you can unlock power dense, greener, cheaper AI factories.

If you need a refresher, go read this 6 months old post first.
On this post, VICR did 5X since this post.

If you are confused about SiC, vs GaN etc. Here is a great infographics that shows where does each of them thrive in the Data center value chain

Now that the market is interested in this deeply, lets go through the 3 small cap names that might do a few multi baggers for us. I will assess them and rank them.

IPWR: ($45M MC. Riskiest with the highest upside )

The 800V DC data center transition is creating structural NEW demand for solid-state circuit breakers. B-TRAN is uniquely positioned to serve this. CEO bought $250K personally at $2.75 (the 52-week low).

- Lead customer qualification Q4 2026, revenue ramp 2027. Markets might sniff it out if it works. They are eyeing the $200M sales pipeline
- They raised recently and will enough cash until the end of year atleast.

All excellent points from their deck

Under‑Appreciated Technical & Strategic Angles

IPWR’s B‑TRAN architecture targets a part of power semis that many investors (and some comps like NVTS) barely touch.

• B‑TRAN is a fundamentally different device: a true bidirectional bipolar transistor vs. unidirectional FETs and IGBTs.

• B‑TRAN: one double‑sided device replaces four devices (two switches + two diodes) in a conventional bidirectional leg.

  

• Economic value is in bidirectional, high‑current protection and switching, not just generic conversion (where NVTS focuses).

• Design‑win path and Stellantis work suggest “system‑level” attach, with long product lives once designed in.

• Capital‑light model with foundry partners lets them scale on mainstream silicon, not exotic substrates.

• Patents: 52 U.S. + 47 foreign issued, 76 pending (as of 12/31/2025) on B‑TRAN.

  

NVTS’s GaN is optimized for high‑frequency switching DC‑DC and AC‑DC; in protector/contactor/SSCB roles its conduction losses and need for extra devices are typically worse. Module‑level density and thermals

Fewer die + much lower conduction loss ⇒: Smaller heat sinks and bus bars.

Implication vs NVTS:

Navitas competes mainly in conversion (chargers, inverters, some motor drives) with unidirectional FETs; IPWR is attacking bidirectional protection and switching functions where a true bidirectional bipolar device has a structural advantage and can displace multi‑device SiC / IGBT stacks.

NVTS has a strong GaN IP portfolio, but it’s all FET‑based. To attack IPWR head‑on in these exact bidirectional switch topologies, they’d likely need new IP and process development.

Key Disconnect and Summary

1. Binary pre-revenue bet, but the risk/reward is massively asymmetric. Lead customer qualification Q4 2026, revenue ramp 2027. Markets might sniff it out

2. Risk: Pre-revenue. Illiquid. Technology adoption could slip. If B-TRAN doesn’t get qualified by year-end, cash runway becomes a concern.

WOLF: ($1.26B)

SiC is becoming the bottleneck material for the AI power stack. Structural capacity constraint on Silicon Carbide wafers. Wolfspeed owns the only 200mm wafer fab on the planet. They are lining up for 300mm wafers and do it in Ameria. Supply chain risk might make this a strategic priority

The Alpha: Market cap (~$1.25B) roughly equals cash ($1.3B). You’re getting the world’s only 200mm SiC fab, 2,300+ patents, and 300mm SiC technology leadership essentially for free. Renesas’s $2B+ investment implies $55-65/share — nearly 2x current price. AI data center revenue is growing +50% QoQ.

Analysts still have a consensus “Reduce” at $14.33 target — 48% below current price. They are stale and will be forced to upgrade.

Emerged from bankruptcy with debt at $2B vs. 6B. Big progress

Key Disconnect: The market is stuck in “bankruptcy hangover” mode while the business is pivoting to AI at breakneck speed. SiC is the ONLY material that works for 800V HVDC (physics moat). Mohawk Valley at ~30-35% utilization means massive operating leverage ahead — every 10% utilization gain drops straight to the bottom line.

Also, their fabs are American.

Risk: Competition on 6-inch substrates (though not on 200mm/300mm).

Watch: May 5 Q3 FY26 earnings (AI revenue trajectory + utilization update). Design win announcements for 10kV SiC MOSFET. 300mm customer qualification milestones.

MX Magnachip: ($155M)

• Cash ($103.8M) = 66% of market cap

• 55 new products launched in 2025 vs 4 in 2024 — a 14x acceleration

• They are actually buying back instead of diluting which most small caps do

• Byreforge LLC filed a 13D (activist intent) for 8.5%

The Alpha: Market prices MX at 0.4x book as a dying display company. Reality: it’s a focused pure-play power semi company with an activist-loaded shareholder base and insider buying at every level. CEO bought 30K shares at $2.84. CFO bought 10K at $2.87. A director bought 60K shares in March

If you simply glance at the fundamentals, they don’t exactly inspire confidence. But, what’s unknown is that they decided to do a full pivot into Power semis. Sold their display business to concentrate fully.

Their outstanding shares are reducing. This is a business in a turnaround stage.

Key Disconnect: The new product pipeline (40V/60V MV MOSFETs for AI servers, IGBTs for solar/ESS, Hyundai Mobis automotive JV) is invisible to the market because headline revenue is still declining from the display exit. Q1 2026 guidance shows gross margins improving from 9.3% to 14-16% sequentially — the inflection is starting NOW.

Risk: Execution risk is real. Legacy pricing pressure. Power semi revenue must ramp fast enough to offset display runoff. Competition from much larger players (Infineon, ON Semi).

Watch: Q1 earnings April 28 — this is 5 days away and the volume surge today (+22.7% on 10-15x normal volume) suggests someone knows something.

These KPI metrics do not show any improvements and hence, Analysts are clearly not pricing in the chance of growth in their business

MOSFET line (0.7mΩ Rdson, 397A, 25% faster switching, PDFN56 package) for sync rectification in the final power stage. That’s where they compete and compete reasonably well against the likes of Infineon and Alpha & Omega at that voltage class

Closing Thoughts:

I Asked Claude Code to determine the Upside Case. Do what you will with it. They are all in different Risk categories to be honest

In my last message on 16 Mar , I talked about how I am raising cash in my folio since a few weeks and that bounces were to be sold. This bounce seems no other, especially because Trump yesterday said nothing new in his speech.

Trump TACO’d 3 times and markets simply ignored until 2 trading days ago. And now this reaction is getting awashed too. Remember that bullish news + bearish action = bearish

This rally has been so good so far, but On a dead cat or oversold bounce, the QQQ is going to rebound aggressively to overbought conditions, stop there, and then crash back down to earth.

So we gotta stay vigilant and watch the reaction today and tomorrow, SPY at 610-615 is gonna be very juicy if there is a last leg downwards.


My entire strategy is to find misunderstood names that are overly negative and get a good entry point. Timing the macro markets perfectly is not where the real money is made, but rather buying when everybody is scared in strategic SPY and QQQ levels like 610-620. Thats when I get in heavy with the cash I raised.

This has allowed me to do well as well as stay calm in these moments of volatility. My opinion is that the markets don’t bottom when Oil is still making all time Highs. Lets see

That being said, here are some of my favorite names to play right now. I usually talk about small caps, however, in downturns, you wanna get into famous big names so you can hold with conviction and for the strong bounce.

RDDT

It is getting lumped into a software basket. Too much negativity for a product that has real network effects and real moats. Yes there is some scare around META’s lawsuit and being liable to creating addictive software, but it is worth to look at

It’s showing early signs of Operating leverage. Expenses are not lifting off as fast as the profits are. This is a sign of a great business. Ofcourse the margins and cash flow is seeing a lift as well

I am super pleased to see the uplift in ARPU in the US as well as the global average revenue. Seems like they have been able to grow their Daily active users as well as the revenue they generate from them

Stockdrifts is working on a feature to identify if growth is accelerating or decelerating. Market is punishing deceleration, so this feature is important. In the meantime here is a snapshot of how growth is slightly accelerating in the US and Internationaly

Buy Zone: $90-$110 is a great accumulation zone

MU

All the news about DRAM prices dropping is peak stupidity.
90% of DRAM is sold as contracts and the contract prices is what actually matters.

This is what reputable research house are talking about. 60% QoQ increase in 2026. HOLD MY BEER!!

There are plenty of X posts talking about how turboquant is a year old innovation and how it is not going to relieve the pressure on tight Memory supply

This correction is a gift. Just look at how this revenue is actually accelerating. Not growing, but accelerating.

Another reason why MU is a better bet than SK Hynix and Samsung because MU has better access to North American helium-linked supply chains

U.S. helium logistics can be more secure during the Iran and US war

Helium prices have roughly doubled since the Iran war began, according to Fitch Ratings. Severe helium disruption is actually bullish for premium memory pricing And that usually helps Micron compared to its Asian competitors

Ofcourse, MU has done 3X suddenly from the last base, so you gotta be careful about your entry points. I am eyeing the 260-290 level. Fingers crossed.

Buy Zone: $260-290

Rest articles covers 2 more names I am excited about

Read more

What Happened & Why the Stock is Down 20%?

For paid subscribers, I have my analysis of T1 and TOYO’s earnings and the arguments about the Ethiopian TOYO’s strategies. Here is a snapshot, TOYO is making a lot of EBITDA vs TE

Read more

This is a conversation space exclusively for subscribers—kind of like a group chat or live hangout. I’ll post questions and updates that come my way, and you can jump into the discussion.

Join chat

How to get started

1. Get the Substack app by clicking this link or the button below. New chat threads won’t be sent sent via email, so turn on push notifications so you don’t miss conversation as it happens. You can also access chat on the web.

Get app

2. Open the app and tap the Chat icon. It looks like two bubbles in the bottom bar, and you’ll see a row for my chat inside.

3. That’s it! Jump into my thread to say hi, and if you have any issues, check out Substack’s FAQ.

They hold a lot of memory patents and is crucial for HBM, MRDIMM memory stack. All the big ones SK, MU, Samsung as well as Google will have to pay them substantial amounts. Litigations going on for over 15 years. Will reach a conclusion this year. District court already ruled in favour with $1B. Should be more actually

There might be partnership deal with SK Hynix coming very very soon. SK Hynix 5 year partnership deal is expiring. The space is too valuable for SK Hynix to NOT renew.

But leaving the litigations aside, their product revenue is ramping up massively. Here is a snapshot from Earnings from a few weeks ago.

There are every indication that Q1 2026 might be better as well.

Here is some commentary from the CEO. I find that very very interesting

“Sales volume for Lightning, Netlist’s overclocked and low latency DDR5 line of products, is ramping up nicely in the system integrator market segment. Adding to this, we’ve completed qualifications at a global server OEM, and they, in turn, are currently in testing with many of their end customers.”

— Chuck Hong, Q4 2025 Earnings Call (prepared remarks)

So their products are coming along well and they are not just relying on litigation and IP fees.

Also,

“A ruling in Netlist’s favor would direct U.S. Customs and Border Protection to stop Samsung memory products that infringe these patents from entering the U.S.”

— Chuck Hong, Q4 2025 Earnings Call (ITC discussion)

Secondly, do you think for a such an important product like memory, Samsung would diddy around and NOT settle? They have already won this case 3 times in district court. ITC is just going to force a fast resolution.

Also, I thought it was interesting to point out that cost of borrowing is at 20%. And now suddenly overnight 8.5M shares become available.

I expect manipulation or some kind of funny business soon. So if it comes to $1-1.2, I might be looking to averaging upwards.

I was the ONLY analyst covering NLST on X, but I can assure that there are plenty of cult retailers on this name over the years.

I am raising some money in my account to profit from the opportunity IF it arrives due to a prolonged war.

I dont think the market’s really realized the risks/chances that Fed might not cut or might actually hike. Market’s been expecting Fed’s control by Trump and this might not materialize in mid terms.

In any case, all you can do is get into misunderstood stocks, at a good entry point and wait for your thesis to play out.

In the next post, I want to share some exciting ways I have been able to build alpha pipelines. Will send the next post soon

Spoiler Alert: $TE will do just fine!!

Module assembly capacity has surged to ~65 GW annually, HOWEVER, the supply chain remains imbalanced with only ~8.3 GW of wafer capacity and ~3.2 GW of c-Si cell capacity domestically

Okay while we scratch our heads over that, there is an S&J’s interview with TE’s VP this Thursday. Definitely don’t miss

Okay, lets look at the supply chain that exists. 5 steps to get to that sweet sweet Solar Panel

Supply chain capacities across the world

This chart allows you to appreciate the mismatch and how the dream of full vertical integration is still far far away.

STAGE 1: POLYSILICON

Combined US nameplate: ~50-55K MT/yr. Actual production ~33K MT. If all of that was converted into Solar cells it would be equivalent to ~13-16 GW of solar modules.

Keep in mind that Polysilicon is also required for semi conductors. Albeit the grade gotta be much purer than solar. 98% of Polysilicon is used for Solar

The real issue is that the 45X Credit: $3/kg only and the gap between Western polysilicon and Chinese is around 12$. So that really is a big gap. Pure, US supply chain is going to come at a short term cost. Not to mention there is simply no capacity for Polysilicon in the US

I think that TE, TOYO and eventually Tesla might do the most minimum possible to satisfy FEOC. They are eating glass or paying 3 times the price to get US Polysilicon.

Atleast the Good news is that we can see Corning promising to increase the production massively. These infrastructure companies do not simply increase capacity. They do it when there is massive shortage.

It is also an indication that the demand is through the roof for Polysilicon if GLW is planning to exponentially increase their Solar revenue.

Lastly, I spun up Claude code Python to calculate the difference. When you source from Non China Polysilicon, it suddenly becomes 1/3 your total costs.

I think there is a lot you can do to mitigate this. For eg. Credits might help and some more financial sorcery.

Lets see if this really comes through in 2027-2028. However, it did happen that a lot of Chinese supply shut down and Polysilicon prices bounced up ~30-50% up.

Also P.S. cell-level cost is deceptively low; a company’s success ultimately depends on the quality of its cell manufacturing process. Because a better manufacturing process means higher efficiency for the panels. And even a little bit more efficiency over its lifetime of a few decades can compound quickly.

Have a look at how Solar cells are produced.

I am sure Elon will delete a LOT of steps and rethink the whole industry from fundamental point of view, however the reality of things are still going to be that having American Solar supply chain is not a short term target.

I am planning to give away all of my Solar watchlist to Annual subscribers. My favorite small caps or best multibagger opportunities with liquidity in the space.


I have been sitting on it for far too long. Will also release interesting watchlist in other themes as well.

If you have been looking to invest in alternative grid scale Batteries and never got around to EOSE, here is a really interesting stock <$1B market cap

Before I talk about this stock, lets set the stage right. After the disastrous earnings of FLNC with suppressed margins and having to move around Battery supply chains through 3 different US facilities in order to chase them 45X credits, it’s time to look at something new

Here is CATL’s Founder Robin Zeng saying that Sodium Batteries can displace 50% of the LFP battery market last year

They officially launched it in production 2 weeks ago

There is a way to get exposure to the Western/American Sodium Batteries Via

I wanted to get this report quickly. But in the subsequent deep dives, I will attach a video from my discussions with Battery expert Jordan G from The Limiting Factor Youtube Channel

“Sodium Ion Batteries are better than Lithium Ion LFP batteries in every way. Cheaper, easier to scale, more abundantly available, less risk of fire”

But then we also discussed how LFP batteries enjoy economies of scale and by the time other alternative forms scales LFP batteries would have moved the goal post


I talked about how Javascript is a an inefficient framework, but it STILL took over the world because it was a first mover and early and had network effects.

With that said, lets jump into snapshot.

Technically they are at a perfect point to get in, if the fundamentals are right

Consolidation and a Retest with volume

Here is what I like about them:

• Revenue backlog of ~1GW

• Ownership model rather than . Assets predicted to generated $40M EBITDA per year now and growing

• They are building AI data centre Battery infrastructure with Peak Energy (Bringing Sodium Ion Batteries to production in the west

• Exclusive rights to bring Sodium Battteries from Peak in Japan and Australia

• Private investors believe in it enough to raise 140M convertible

• The deal was oversubscribed that why upsized from $125M to $140M

• Not a dilution Fest.

  
  

  

  

  

  
  
  
  

Lets unpack the Convertible debt terms here

Lets look at fully diluted stocks

These are my estimates for the worst case dilution from this convertible debt raising.

The capped call is the critical detail. Here’s how dilution actually plays out at different stock prices:

Stock Price: Below $5.18

Dilution Impact: Zero dilution

What Happens: Notes stay as debt. No conversion. You just pay 5.25% interest.

────────────────────────────────────────

Stock Price: $5.18 to $8.12

Dilution Impact: Effectively zero dilution

What Happens: Notes are in-the-money BUT the capped call offsets the dilution. Company can settle in cash, stock, or combo.

────────────────────────────────────────

Stock Price: Above $8.12

Dilution Impact: Incremental dilution begins

What Happens: Only on the portion above $8.12. Example: at $10/share, dilution is only on the $10 - $8.12 = $1.88 spread per share, not the full

conversion.

────────────────────────────────────────

Stock Price: Way above $8.12 (e.g., $15+)

Dilution Impact: Full ~27M share dilution

What Happens: But at that point, the stock has nearly 4x’d — dilution is a “good problem to have.”

Key insight: If the stock stays between $4 and $8.12, there is essentially NO dilution from these notes. The dilution only matters if the stock

rips past $8.12, and at that point shareholders have already made 100%+ gains.

I created a thread from the last earnings call. Wanted to get this article out before market open. As I build more conviction on this name, I will share my deep dives into their tech.


DO NOT GO IN WITH SIZING. It’s Only a spec position for me

1/8 Energy Vault (NRGV) pivoted from pure EPC to asset owner, and Q3 shows the Asset Vault story getting real.

• Asset Vault platform formally launched this quarter

• First two Texas and California storage IPP assets now operating

• Strategy targets top-quartile returns at “privileged” interconnect points

2/8 Management frames Q3 2025 as “one of the most pivotal” in company history, with execution catching up to strategy.

• Four projects now operating or in construction totaling 340 MW

• These initial assets expected to deliver >40 M EBITDA annually

• 10–15 year contracted revenue tails as projects come online

3/8 The OIC preferred deal now funds a sizable self-reinforcing growth engine where NRGV builds, owns, and services its own assets.

• 300 M preferred equity from OIC closes, launching Asset Vault Fund 1

• Fund 1 enables >1.1 B CapEx across 1.5 GW storage projects

• NRGV expects 100–150 M cash flow back from project margins and services

4/8 Asset origination is ramping fast, with NRGV leaning into both grid storage and emerging AI power demand.

• Acquired 150 MW SOSA interconnect site near Houston from Savion

• Stoney Creek 125 MW project in Australia backed by 14-year offtake

• Collaboration deepening with Crusoe on AI data center power solutions

5/8 Commercial momentum is visible in backlog and pipeline, even with tariff noise and macro uncertainty in the U.S. market.

• Revenue backlog at 920 M, up 112% year-to-date

• Backlog has more than 4x’d since Q3 2024

• Development pipeline sits around 2.1 B, or 8.7 GWh advanced projects

6/8 On the call, management stressed diversification and agility as buffers against U.S. tariffs and policy volatility.

• Only about 10% of backlog exposed to U.S. tariff swings

• Australia represented more than half of Q3 revenue

• Asset Vault pipeline curated for “upper-tier” IRR and bankable offtake

7/8 Financially, Q3 shows the early earnings power of the new model, even on a small revenue base.

• Q3 revenue 33.3 M vs 1.2 M a year ago, up ~27x

• Q3 gross margin 27%, year-to-date gross margin ~32.6%

• Adjusted EBITDA loss narrows to 6 M from 14.7 M loss last year

I have huge respect for Daniel, S&J, and DVB and value their insights on TE massively. Cherish them as fellow investors. However, I think there are a few angles they might be overlooking. Let’s unpack them

Here Daniel mentioned that having whole domestic US supply chain is important.

TOYO’s been alleged to be trade loop hole extractor

DeepValueBagger is assuming that TOYO isn’t US production.

However, Even as seasoned/knowledgable these investors are on the Solar topic, here are my assessment and rebuttal on it

INSIGHT 1: TE literally doing the same as TOYO. Shipping their US polysilicon to Asia, converting it into cells and shipping it back

First lets start with what TE/T1 is actually doing with their OWN cells

- T1 Energy CEO Daniel Barcelo in Nov 2025 explicitly stated:

“Until we have our solar cell plant up and running, we currently are sending our polysilicon to Southeast Asia in order to make wafers and cells and then bring those cells back to the United States.”

Here is a timestamped video for when he said that.

So TE/T1 is literally doing what TOYO is doing and getting a lot of shit for it. Since TOYO has now shifted to Ethiopia, travel time there is still lesser than the South east partners of TE/T1

In the show it’s talked about how there are literally 1 or 2 GW of Solar cell capacity in the entire US. Just TE’s G1 capacity is 5GW of modules.

Recall how Solar modules are finally produced

Also, Remember that Tree Oaks collaboration told us that TE/T1 is aiming to be 70% US supply chain by end of 2027. That’s still 2 years away.

Conclusion: Even the American Solar TOP DAWG is doing exactly what TOYO is doing and aiming to be only 70% American by the end of 2027

I wrote 3 TE/T1 deep dives back in start of October. I have spent countless hours diving deep into the industry

INSIGHT 2:

Read more