By the Numbers

Revenue came in at $437M, up 157% year-over-year. The company cleared the high end of its own guidance range and came in just ahead of consensus estimates.

The first item on my checklist was margin. My concern had been that a heavier optical product mix could structurally compress the blended gross margin. What actually happened: Non-GAAP gross margin held at 68% — not just above the 64% threshold, but well above it. The margin compression concern was wrong at the assumption stage. Operating margin came in at 49%.

The second item — next-quarter guidance — landed between $465M and $475M, above the lower bound I’d flagged.

Beyond the checklist: FY27 annual revenue growth guidance was raised from “>+50% YoY” to “>+80% YoY.” Optical alone is now guided to exceed $600M for the year, with all three product lines — DSP, SiPho PIC, and ZeroFlap — each individually crossing $100M.

What This Company Does

For readers less familiar with the name: Credo’s core business is AEC — Active Electrical Cable. These are short-distance copper interconnects linking servers, switches, and GPUs inside data centers, with signal-processing chips embedded at both ends to correct for signal loss. In an AI cluster running tens of thousands of GPUs simultaneously, a single cable failure can halt an entire training run. Reliability and power efficiency are the product. Credo holds roughly 88% of this market, which accounts for over 70% of its revenue.

The company is now expanding into optics. On May 27, Credo closed its acquisition of DustPhotonics — internalizing silicon photonics PIC, a key optical component. That’s the DustPhotonics item on my checklist. The $750M cash outflow hits next quarter’s balance sheet.

Strong Earnings, Cold Reception

Why did the stock pull back despite a clean set of numbers?

Two things converged. First, revenue came in just barely below Wall Street consensus, and the stock had already more than doubled from my March entry. At that level of appreciation, even a solid pass tends to trigger profit-taking. Cumulative insider selling in the months around earnings added to the near-term ceiling.

Second — and more decisive — the company guided its first half conservatively. Management called for mid-single-digit sequential growth per quarter through H1, while the market had expected a more aggressive ramp from the opening whistle.

My read: this is valuation normalization, not fundamental deterioration. In my last sector check, I wrote about companies whose destination had moved farther away even as the stock doubled — CRDO was the lead example. Now the direction has flipped. The stock pulled back, but the destination moved farther again. The conclusion is the same either way. So what was actually behind that conservative tone?

Behind the Conservative Tone

I went through the conference call carefully.

The CFO explicitly framed H1 as mid-single-digit sequential growth and H2 as an aggressive ramp. This wasn’t management being evasive — the CFO was describing the actual shape of the production curve. Three supply bottlenecks explain that shape.

First, optical product validation. The CEO flagged significant supply chain tightness and the importance of securing capacity commitments across key product families. Optical modules require sequential validation across lasers, PICs, drivers, and DSPs — yield stabilization takes one to two quarters per component. Until yields normalize, shipments are intentionally constrained.

Second, DustPhotonics integration. The acquisition closed May 27, and the integration cycle is only beginning. H1 is a work-before-synergy period; meaningful incremental revenue contribution comes in the back half once the integration settles.

Third, the 200G transition. The CEO stated directly that 200G per lane would contribute only lightly within FY27. This is an industry-wide variable, not something Credo controls.

But the most interesting moment in the call came precisely on the 200G question. When asked about the delay risk, the CEO explained that large customers are already designing with two physical ports per GPU — making an 8-lane 100G configuration a practical alternative to 4-lane 200G. A slower 200G transition doesn’t reduce Credo’s revenue exposure; it shifts the configuration. The risk I’d flagged in my prior sector check is effectively neutralized.

One more note: the CFO remarked that the company grew more than twice as fast as expected from the start of FY25. That’s relevant context for reading the FY27 “+80%” guide — it may be a starting point, not a ceiling.

My conclusion: H1 isn’t a dry field. It’s a planting and watering period. The harvest is designed for the back half, and management described that design in their own words.

How the Coordinates Shifted

Read more

In my last article on $FLNC following Q2 FY26 earnings, I wrote this: “If the first hyperscaler order converts at a meaningful scale, FY27 revenue estimates could move higher.” That trigger arrived faster than expected — and with a much bigger name attached.

Today, Siemens released an official press release. Fluence’s Smartstack has been named a core component of the NVIDIA DSX Vera Rubin NVL72 reference architecture. A joint announcement signed by Siemens, NVIDIA, and Fluence.

What happened today

The facts as stated in the press release.

Siemens has developed a reference design that translates NVIDIA’s AI factory vision into a deployable electrical, power, and controls architecture. The standard unit is sized at 136MW total facility capacity with a 100MW IT load. It covers the full path from utility connection (34.5kV) to the rack interface, built on repeatable, scalable modules aligned with NVIDIA DSX Vera Rubin deployment units.

Within this architecture, Fluence holds the BESS role. Siemens handles power controls, nVent handles liquid cooling, and Fluence handles battery energy storage — a three-way standard. Fluence CGO Jeff Monday stated in the press release that Smartstack is central to this architecture, with voltage and frequency ride-through, black start, grid demand response, and AI load smoothing cited as core functions.

The line between a marketing partnership and a technical standard comes down to one question: did NVIDIA explicitly name a specific supplier in a reference design?

Today’s press release answers yes.

What this means for valuation

The coordinates in my May 7 earnings note were built on the $5.6B backlog and the MSA signings with two major hyperscalers. Trigger 1 at the time was the first hyperscaler order conversion — something I expected to confirm at Q3 earnings in August.

Read more

At the time, the company’s market cap was around $44M. Its Bitcoin holdings were worth roughly $62M. The market was effectively valuing the core semiconductor business at negative $18M — two decades of IoT chip-making, priced below zero. I wrote then: “If clearing the Bitcoin position was the first door to open, the second is H2 revenue actually materializing.”

Today (May 28), the first door opened.

Read more

Summer is arriving in Florence — the sun has been relentless, and between the heat and the caffeine from my morning pour-over, sleep isn’t coming easy tonight. So I’m back at the desk, reviewing coordinates.

Back in March, I identified optical communications as one of this year’s defining themes and published a deep-dive research piece. The market caught on quickly. Prices ran — some names blew past their peak targets — and I found myself harvesting faster than expected. My last published update was the March 27th letter, “A Stricter Look at Optical Communications Companies.”

After that, optical went quiet. Meanwhile, memory took the spotlight — and for a clear reason: earnings were coming in far ahead of expectations, and analyst price target upgrades started piling up. We flagged that move a month early.

But here’s the thing. Within the AI sector, there’s another theme where earnings are tracking significantly above expectations. It’s optical communications.

I’d stopped publishing updates after that last letter. With quarterly results now in from most of the names I track, I revisited the coordinates. What I found was worth sharing.

Read more

I first mentioned $MOS in Trump Goes to Beijing: Finding Fields That Grow Either Way, published in early May. Korean subscribers will know the stock was on my radar since a March agriculture sector piece — but for Substack readers, the Beijing piece is where this thesis begins. That was about eleven weeks ago. Since then, sulfur prices hit an all-time high, the US-Iran ceasefire talks reached what looks like a final stage, and Q1 earnings came out. This note walks through how the logic has evolved — honestly — from that first mention to now.

Read more

$SHLS. I first flagged this name in my deep-dive on the EIA’s 200-page energy outlook, and followed up with a full tracker three weeks ago, right after Q1 earnings dropped.

At the time, I wrote this: the seed had sprouted, and the question was whether the second field would deliver a real harvest. I held the valuation reference points unchanged — with gross margin at 29.6%, I said the midpoint wouldn’t move meaningfully until margin recovery showed up in actual numbers. That confirmation is penciled in for August.

Three weeks have passed. $SHLS pulled back, then came right back to $9.50. And in the meantime, two things have shifted. I want to share both.

First — the headwinds are clearing, one by one

When I first identified this company, the thesis was straightforward. SHLS sat at the most congested bottleneck in large-scale solar project EBOS, backlog was at an all-time high, and the stock was still depressed from a market misread. Strong fundamentals, underappreciated positioning — that was the idea.

What I didn’t fully anticipate was how hard the financials would get squeezed in the interim. Q1 gross margin came in at 29.6%, down 540bps year-over-year. Three culprits: higher short-term raw material procurement costs, temporary labor inefficiencies from the Tennessee facility consolidation, and tariff-related cost pressure.

Here’s where each of those stands now.

Facility consolidation costs — removed. On May 19th, the Mega Facility in Portland, Tennessee held its official grand opening. 638,000 square feet, consolidating three existing Tennessee sites into one. The CFO noted that roughly one-third of Q1’s margin pressure came from facility transition costs. With the Mega Facility now live, those costs start dropping out — partially in Q2, fully in Q3. This isn’t just a new factory. It’s a fixed-cost consolidation, a production automation platform, and the manufacturing base for future BESS and data center products — all in one investment. One headwind, physically removed.

Tariff risk — additional escalation taken off the table. Following the Trump-Xi summit in Busan last October, US tariffs on Chinese imports settled at around 47%, with a formal truce extended through November 2026. The Beijing meetings this month reconfirmed that both sides intend to hold that level. SHLS manufactures in the US, so finished-goods tariffs were never the issue — but some raw material and component sourcing carries exposure. The risk of further escalation has been substantially reduced.

Raw material procurement costs — still unconfirmed. This is the one remaining variable. Whether spot procurement rates have normalized won’t be visible until Q2 results in August. That’s the real confirmation point for margin recovery.

Two of three headwinds cleared. One remains. The picture is resolving faster than I expected.

Second — the same eye sees a wider field

When I first saw the SHLS bottleneck in solar EBOS, the core insight was this.

EBOS represents just 2.5% of total utility-scale solar project cost. But nothing moves until EBOS is installed — it’s the last component in, and the last to be delivered. When that bottleneck backs up, the entire project stops. SHLS built its BLA (Big Lead Assembly) system over more than a decade to solve exactly this: move complex field wiring into the factory, eliminate the need for licensed electricians on-site, slash installation labor. Over 80GW of contracted, under-construction, and operating projects is the proof.

Now I’m applying that same eye to BESS and data centers.

Read more

Since the last tracker, $GCTS has been generating a lot of conversation. That’s a good sign — and also a signal that some things need to be set straight. There are new facts to confirm, and some inaccurate numbers making the rounds. Today I’ll address both.

First: What the amended filing changed

Shortly after the last tracker, Alphachips filed a correction to their supply contract disclosure. The contract name changed.

“Semiconductor contract manufacturing” → “Custom semiconductor volume production”

The difference matters. Contract manufacturing means the customer controls design and production instructions while the manufacturer simply runs the factory. Custom volume production means GCTS hands over its chip design entirely, and Alphachips produces a dedicated chip based on that design — built for commercial delivery, not test samples or small-batch prototypes.

Read more

I’ve been watching this company since last autumn. $GCTS never quite met expectations — always a step behind — so I kept my estimates conservative and waited. Then last night’s Q1 numbers came in. I was genuinely surprised. In a good way.

1. What came out

Q1 2026 results dropped after the close on May 12th.

Revenue came in at $1.92M, up 287% year-over-year. 5G chipset shipments grew 58% quarter-over-quarter, from 1,900 units in Q4 to 3,000 units. EPS was −$0.15, a slight miss versus consensus of −$0.13.

In my previous note, I said 300+ units would confirm the ramp-up story was on track. The number that came back was 3,000. Ten times my threshold — and 300 was itself a bottom-up estimate, not a conservative guess pulled from thin air.

On EPS: the miss is noise. Management flagged that Q1 service revenue included a one-time licensing recognition, and that going forward, service revenue will step down as the business shifts toward product shipments. The unit count, not the EPS, is the right lens for this company.

One more thing alongside the results: the licensing agreement with one of the world’s largest satellite operators — signed in January — was expanded. A reference platform agreement was added on May 7th. First shipments are expected in H2 2026.

Read more

$FLNC. While the other names from our April 6th BESS deep dive — Sanil Electric and Samsung SDI — kept climbing, this one had us holding our breath. Today, it made its comeback.

April was rough. UBS slashed its target from $22 to $8 and slapped a Sell rating on the stock, warning that EV factories converting to BESS production would flood the market with supply. The stock dropped 9% in a single day. Citi followed, cutting its target from $20 to $15. The stock entered today’s print carrying a YTD loss of 41% — and is now trading up 30% in after-hours at $17.25.

I held through all of it, and shared that view with readers. Today I want to explain why.

What happened today

Fluence Energy (Nasdaq: FLNC) reported its Q2 FY2026 results. The fiscal second quarter covers January through March.

Revenue of approximately $465M, up 7.7% year-over-year. Adjusted Gross Margin of 11.1%, up from 10.4% in the same quarter last year. Adjusted EBITDA of –$9.4M, a $21M improvement from –$30.4M a year ago. Backlog at $5.6B, a new record. Annual Recurring Revenue (ARR) at $157M. Total liquidity maintained at approximately $900M.

Two strategic events stand out. First, Fluence signed Master Supply Agreements with two major hyperscalers, with the first order expected to convert in Q3. Second, year-to-date order intake reached approximately $2.0B — more than double the same period last year.

The direction of these numbers is clear.

One item needs context. Approximately $80M of revenue slipped from Q2 into Q3, driven by temporary port-related shipping delays. Inventory on the balance sheet grew sharply quarter-over-quarter from $455M to $764M, which corroborates this. The delayed shipments were received in April, and IR materials confirm no further delays in Q3. With this in mind, the Q2 revenue figure understates the underlying business momentum.

Who is Fluence

For those less familiar: Fluence Energy is a battery energy storage system (BESS) company co-founded by Siemens and AES. It designs and delivers energy storage solutions, layering on operational services and AI-powered asset optimization software (Nispera). With over 7.4GW deployed across nearly 50 markets, Fluence ranks second in the U.S. market by installed capacity — sitting right behind Tesla Megapack. It operates as an integrated solution provider: any battery cell from any supplier, packaged into a customer-ready system — and after delivery, it continues generating revenue through services and software.

Technology and product updates

Three product developments from this quarter are worth noting.

The first is Smartstack’s commercial debut. Fluence’s newest product platform entered live operation this quarter. It ships at 7.5MWh but is upgradable as requirements change — a modular architecture that accepts battery cells from five different suppliers. The split design, which separates critical control systems from battery packs, reduces maintenance downtime and enables AI-driven edge computing for advanced controls. How quickly the reference base builds from here will directly shape future order competitiveness.

Read more

Yesterday after the close, I said I’d be watching margins closely in today’s earnings. Q1 was likely to be the quarter where tariff cost pressure was most concentrated, and the key question was how management would talk about the timing of margin recovery. The numbers are out.

Why I first flagged this company

Back in March, when I first shared SHLS, the thesis wasn’t a simple undervaluation call. SHLS sat at the most severe bottleneck in the Balance of System layer of large-scale solar projects — record backlog, solid fundamentals, but a stock still recovering from a market misread. The idea was straightforward: the bottleneck is real, the market doesn’t know it yet.

About 40 days have passed. GEV moved from $877 to $1,090 (+24%). SHLS moved from $6 to $9.50 (+58%). The thesis played out.

And while that was happening, a new field was quietly opening.

The new field — BESS and data centers

SHLS owns the DC wiring bottleneck in solar. But at the investor seminar in March, management wasn’t talking about solar.

BESS is evolving from a simple ride-through technology into UPS infrastructure for data centers and a grid-firming resource — and at the core DC-coupled bottleneck of that architecture sits SHLS’s DC-DC Recombiner. GPU load fluctuations causing voltage instability, high-speed power delivery demands — these are creating new demand, and SHLS is the only scaled company with over a decade of high-voltage DC customization experience.

Management put numbers to it directly. Per 1GW data center: 400–600 Shoals Recombiners, $42–72M revenue opportunity. The US data center market is estimated at 10–15GW per year. Management stated the opportunity is “significantly up” compared to seven months ago.

This is roughly half the reason the market is pricing SHLS at $9.50. The solar bottleneck thesis was right, and now a new TAM in BESS and data centers is beginning to stack on top of it.

What happened today

Revenue $140.6M, Adjusted EBITDA $21.1M — both at or above the top of guidance. Backlog and awarded orders hit a company record of $758M, with $628M scheduled for shipment within the next four quarters.

Then came the margin I was watching.

Adjusted gross margin: 29.6%. Down 540bps year-over-year from 35.0%. The short-term raw material sourcing cost burden from tariffs was visible, as expected. Somewhat disappointing.

Inventory on the balance sheet surged sharply, which I’m reading as pre-positioning ahead of the Mega Plant ramp. Management guided for this inventory to be absorbed through Q2–Q3 as operating cash flow recovers, and maintained full-year 2026 OCF guidance of $65–85M. Per the Shoals IR presentation, management’s view is that “the U.S. market appears to be extremely resilient, and our capacity expansion could not have come at a better time in our history.”

What changed in the valuation

At the time of my initial analysis on March 26, 2025, SHLS was trading at $6.00 — sitting 14% below my fair value midpoint of $7.00, with an upside range of +17% to +83% toward the upper heat zone of $11.00. The bottleneck was intact. The market wasn’t aware yet.

Today, with the stock at $9.50, those same coordinates — fair value midpoint $7.00, upper heat zone $11.00 — remain unchanged. What moved is the stock, not the destination.

The stock has already passed the fair value midpoint and now sits 29% above it. The remaining distance to the upper heat zone is just +16%. The upside range has flipped: the lower bound is now negative. The current price is already discounting a significant portion of the 2027 Bull scenario.

After reviewing the earnings and the financials, I wasn’t able to move either coordinate. The underlying cash flow this company is generating right now simply doesn’t support it. At 29.6% gross margin, even a revenue beat can’t shift the intrinsic value. A meaningful move in the fair value midpoint becomes possible only after margin recovery is confirmed — likely Q2 at the earliest.

The upper heat zone of $11.00 requires both Bull-case earnings and peak multiples simultaneously. It’s not a coordinate you arrive at simply because results improve.

What to watch

Priority 1 — August 2026, Q2 gross margin

This single data point separates the Base and Bear scenarios. If margin recovers above 31%, the “Q1 was the trough” narrative is complete and the fair value midpoint begins to move. If not, the opposite. Management’s outlined path remains plausible.

Priority 2 — June 2026, ITC Voltage final ruling

If the import restriction on Voltage is confirmed, SHLS gains stronger pricing power and market share defense. This doesn’t directly move the midpoint — it raises the probability of the Bull scenario.

Priority 3 — H2 2026, BESS backlog accumulation

Currently around $67M. Once quarterly bookings of $20–30M push the total past $100M, the data center TAM becomes visible in the numbers. At that point, the market will begin to redefine SHLS — not as a solar BOS company, but as a solar + BESS infrastructure company. That redefinition is when the coordinate moves.

The bottleneck in solar was what I saw first. That view played out and the stock followed. A new field opened in the meantime. The stock has already run ahead to that new field.

The seed has sprouted. Now it’s time to see whether the second field actually delivers a harvest. That confirmation may be two to three months away. Whether to trim or let it grow is yours to decide.

Ciao 🌱

On April 22, we looked closely at three memory names. (→ [Deep Dive] It All Comes Down to Memory: The Three Walls GPU Is Waiting On)

“Relative to the scale of earnings being generated right now, the price level still reads as undervalued.”

Two weeks have passed. All three companies reported record quarterly results. Prices moved. But what matters more is this: while the stock prices rose, the earnings outlook rose faster.

In other words — the price went up, but the remaining upside didn’t shrink. For some names, it widened. Let’s confirm the current coordinates, even if slightly delayed.

Why This Earnings Season Mattered

Memory companies carry a persistent label: “cycle stocks.” Up when times are good, down when they’re not — a business where investors always apply a discount because “peak earnings don’t last.”

But in this earnings season, two companies said the same thing independently: the current demand environment is structurally different from past cycles.

SK Hynix’s IR materials explained why directly. AI has moved beyond large-model training into what they call “agentic AI” — systems that reason and act in real time across diverse service environments. As AI agents generate more data, the memory required to process and store it expands across the entire DRAM and NAND product spectrum.

Why is this cycle different? Past cycles were driven by smartphones and PCs — markets sensitive to economic conditions, where inventory buildup killed orders quickly. Today’s buyers are hyperscalers building AI infrastructure. They respond to competitive dynamics, not economic cycles. Missing a quarter of investment means falling behind rivals — so demand has a stickiness that didn’t exist before. Put simply: the more sophisticated AI becomes, the more memory it needs — in type and in volume simultaneously.

Samsung’s IR was even more direct. Supply fulfillment rates relative to demand are at historic lows. Unlike previous years, customers are pre-booking 2027 demand now, out of concern about supply constraints. Even with current bookings alone, the supply gap in 2027 is expected to be wider than 2026.

Why is supply so slow to respond? Memory fabs take a minimum of two years from investment decision to first output. Even if capacity expansion is announced today, it won’t hit supply until late 2027. The key valuation variable isn’t how strong demand is — it’s how long demand stays ahead of supply. Both companies’ IR materials sent an official signal that this period may be longer than the market had assumed.

The numbers confirm it. SK Hynix posted an EBITDA margin of 79% this quarter. Samsung’s consolidated EBITDA margin jumped from 22% to 51%. These aren’t just big profit numbers — they demonstrate the quality of earnings. When cycle stocks sustain margins at this level, it’s worth asking how much of a discount still applies.

This earnings season showed in hard numbers that the gap between supply and demand may be longer and steeper than expected. Keep this context in mind as we check each name’s coordinates.

SK Hynix (KRX: 000660)

What happened

Q1 2026 results were reported on April 23. Revenue: KRW 52.6T. Operating profit: KRW 37.6T. Both quarterly records — and the first time in history quarterly revenue has crossed KRW 50T.

The numbers were strong. But what surprised me more was the substance. DRAM ASP rose mid-60% quarter-over-quarter. NAND ASP rose mid-70% quarter-over-quarter. Not just HBM — every product category repriced simultaneously, and by far more than consensus expected. Operating margin: 72%. EBITDA margin: 79%. Both quarterly records.

The balance sheet transformation is equally notable. Cash has grown nearly 4x in one year — from KRW 14.3T to KRW 54.3T. Debt fell from KRW 23.3T to KRW 19.3T. Net cash surged to KRW 35T. The company has set a target of reaching KRW 100T in net cash.

One more development: discussions around long-term supply agreements (LTA) with major customers have been officially acknowledged. If contracts are finalized, volumes and prices for the coming years would be locked in advance — a meaningful shift in how the market should think about the “cycle stock” designation.

What changed in the valuation

The target price revision this time is significant. My valuation approach estimates SK Hynix’s value based on the total earnings it can generate going forward. As of April 22, that estimate was running at a certain level — the market knew prices were rising, but actual contracted unit prices per customer aren’t disclosed until earnings. When the results came in, those contracted prices were substantially above consensus estimates. The magnitude of future earnings changed — and the target price reflects that.

SK Hynix — Valuation Update

Read more