DK Street Journal

A category built from three companies that make glass, lenses and display chemicals is showing a tidy-looking 18.6% pullback over the past month. But averaging these three names together erases the actual news: Corning Incorporated (GLW), the dominant U.S. maker of optical fiber and specialty glass for telecom networks, AI data centers and smartphone screens, crashed 46% from a June peak in a single earnings reaction; LightPath Technologies (LPTH), a micro-cap maker of infrared lenses and optical assemblies for defense and thermal-imaging customers, sold off after diluting shareholders even as its order book surged; and Universal Display Corporation (OLED), which licenses OLED (organic light-emitting diode) materials and collects royalties from Samsung Display and LG Display, has simply kept shrinking for months, with its stock never leaving bear territory to begin with.

Corning: real growth, a guidance miss, and a sector-wide reflex. Corning's Optical Communications segment — the part of the business selling fiber and connectors into AI data centers — grew sales 32% year-over-year in the second quarter of 2026, with segment net income up 77% to $438 million, and the company has signed multi-year fiber agreements with Meta, Apple, Nvidia and Amazon, according to Yahoo Finance's earnings recap. That's the business case, and it's real, not manufactured. But the stock fell more than 15% on July 28 not because results were bad — they beat — but because third-quarter guidance of $4.9–5.0 billion in core sales merely met, rather than exceeded, expectations that had been bid up alongside every other AI-optics supplier; CNBC reported the selloff also hit Coherent, Lumentum, Marvell and AXT, and Benzinga tied it to fresh doubts about hyperscaler AI capex after Alphabet's quarter showed capital spending doubling while free cash flow went negative. Corning traded at roughly 87–106x trailing earnings and 52–59x forward earnings before the drop; even after a 46% retreat from its peak, the multiple remains rich against its own history. This is a sector-wide AI-optics de-rating, not a gradual, company-specific band cut — the internal trend rating only marked the downgrade a day after the price had already collapsed.

LightPath: business accelerating, stock diluted. LightPath's fiscal third-quarter revenue (ended March 2026) rose 109% year-over-year to $19.1 million with backlog up 196% to $110.6 million, the company reported — a genuine acceleration in infrared-optics demand. The stock's ~30% July slide traces instead to a $100 million stock offering priced at $14 that diluted holders to shore up a thin $11.5 million cash balance, plus a broader semiconductor selloff in late July, per reporting on the offering and the sector pullback. With no meaningful trailing earnings, price-to-sales near 12.7x is the relevant yardstick — a premium that reflects growth expectations, funded now by more shares outstanding.

Universal Display: the business really is shrinking. Unlike the other two, OLED's stock decline matches its fundamentals. Second-quarter 2026 revenue fell 11.6% year-over-year to $152 million, material sales dropped 26%, and royalty revenue fell 26% in the first quarter, according to Gurufocus's earnings recap and an SEC filing. Management narrowed full-year guidance to the low end of $630–670 million, citing weak consumer-electronics visibility, per Biggo Finance, and would not commit to a 2027 date for commercializing blue phosphorescent emitters — the technology underpinning its entire long-term growth case. Revenue is concentrated with Samsung Display and LG Display, exposing it to Asian trade and demand swings, per Biggo. The stock has held a persistent bear rating since at least March, never breaking down because it never rallied — trailing earnings multiples near 17–19x are cheap only against OLED's own history as a premium growth licensor, not against a business that's still contracting.

The setup

Where it stands — Corning's AI-fiber growth is real but priced richly; LightPath's demand is real but diluted; OLED's decline is real and ongoing. Would confirm — Corning's Optical Communications revenue growth stays above 25% y/y and LightPath's backlog keeps converting to revenue without further share issuance. Would invalidate — Universal Display's royalty revenue stabilizes or grows sequentially, or Corning's Q3 core sales miss the $4.9–5.0 billion guided range. Watch next — Corning's Q3 2026 earnings report and any dated 2027 blue-emitter commercialization announcement from Universal Display. Valuation — GLW ~52-59x forward earnings versus its own multi-year range; LPTH ~12.7x sales; OLED ~17-19x trailing earnings, near a multi-year low.

What actually happened

Four companies that turn cheap natural gas into fertilizer and industrial chemicals — CF Industries, Nutrien, Methanex and LSB Industries — have posted a combined average stock gain of roughly 40% over the past year, and the pitch is simple: they buy the same gas that has crushed Appalachian and Haynesville drillers, so a weak Henry Hub price should widen their margins rather than shrink their revenue. But the underlying data show that story mostly describes one company. The rest of the group's gains trace to a geopolitical price spike in nitrogen fertilizer that has already reversed, and in one case has already shown up as a real earnings hit.

The spike is documented: urea prices jumped roughly 80%, from about $400 a ton to over $850, between February and April 2026 after the Strait of Hormuz conflict threatened Middle East supply routes. Prices then fell back to about $453 a ton by June as Iran resumed exports and a US-Iran deal eased the blockade. Ammonia and UAN, a liquid nitrogen fertilizer, followed the same pattern, falling further in July as the crisis premium unwound. That round trip lines up almost exactly with the stocks: three of the four names — Nutrien, Methanex and LSB — are down over the trailing 90 days (-8.2%, -13.5% and -26.3% respectively) even though a partial bounce leaves their 30-day returns positive. Trend-following price bands as of July 31 confirm the split: Nutrien, Methanex and LSB all carry bearish signals, the same label carried by every Appalachian and Haynesville gas driller sampled alongside them. Only CF Industries, which makes ammonia and urea at low-cost US gas-fed plants, holds a genuinely bullish signal, strengthening into a strong-uptrend reading in late July.

Company by company

CF Industries is the one name where the numbers agree with the price. Its enterprise value trades at 6.05 times EBITDA, 27% below its own 10-year median of 8.25 times, and its trailing price-to-earnings ratio has compressed to 9.5x from roughly 14x in May even as the stock held near highs — meaning earnings grew faster than the share price (EV/EBITDA data). The company reached a final investment decision on its roughly $4 billion Blue Point low-carbon ammonia project with partners JERA and Mitsui, underwriting most of the capex through offtake rather than the company's own balance sheet, and received a permitting boost from federal regulators. About $1.7 billion remains on its buyback authorization. Verdict: CONFIRMS.

Nutrien, the world's largest potash producer and a major crop-nutrient retailer whose 2,000-plus retail locations diversify it beyond pure nitrogen, actually improved its business in the first quarter — nitrogen segment earnings rose to $482 million and potash volumes set a record — even as its stock fell from about $75 to $63 in the same window and its trailing P/E compressed to 12.4x from 15.7x (Nutrien Q1 2026 results). That is a business improving while the tape got worse, a genuine dislocation, but only one quarter of confirming data exists so far. Verdict: INCONCLUSIVE.

Methanex, the world's largest methanol producer, closed a $2.05 billion acquisition of OCI Global's methanol business in mid-2025, adding Gulf Coast plants funded with cash, new shares and assumed debt (acquisition announcement). But methanol prices have stayed roughly flat through July, held down by ample Gulf Coast supply from Methanex's own restarted Geismar 3 plant and weak Chinese demand for methanol-to-olefins conversion (pricing data) — a supply glut, not the structurally short market the thesis assumes. Its trailing P/E is not meaningful near breakeven earnings. Verdict: CONTRADICTS.

LSB Industries, a smaller Oklahoma-based ammonia and nitrogen producer pivoting toward low-carbon products, is up 40.8% over the past year but its most recent quarter shows real damage: gross margin collapsed to 6.8% from 21.1% in the prior quarter, and net income swung to a $6.2 million loss from a $19.7 million profit, coincident with the nitrogen price crash. Verdict: CONTRADICTS.

The technical read

CF's bands strengthened to a strong uptrend on July 28, the only member of the group to genuinely diverge from the bearish signals carried by Nutrien, Methanex, LSB and every Appalachian/Haynesville driller sampled alongside them that same week — meaning three of the four "gas-buyer" chemical stocks currently carry the identical bearish label as the gas sellers they were supposed to be mirroring.

The setup

Where it stands — CF Industries alone shows fundamentals and valuation supporting its gains; Nutrien, Methanex and LSB do not currently confirm the cohort's feedstock-margin thesis. Would confirm — Nutrien's Q2 2026 nitrogen EBITDA holds near or above $482 million while its stock re-rates toward its improving fundamentals. Would invalidate — Urea or ammonia prices fail to stabilize above their June lows (~$453/t urea, ~$962/t ammonia) into the fourth quarter. Watch next — Nutrien's Q2 2026 earnings release and India's next urea tender award, both expected within weeks. Valuation — CF trades at 6.05x forward EV/EBITDA versus its own 8.25x 10-year median; Methanex's P/E is not meaningful near breakeven earnings.

Seven companies that supply the physical building blocks under modern electronics — laser chips, chip-patterning templates, power transistors and the raw semiconductor wafers beneath all of it — sold off together in July. But the group's roughly 15% average monthly decline, and its eye-popping year-long average gain, obscure four unrelated business stories rather than one industry cooling off in sync.

A stretched advance, not a bubble that's popped. AXT, a maker of indium-phosphide and gallium-arsenide substrate wafers whose China-based subsidiary carries export-licensing risk, posted a second-quarter profit of $0.19 a share versus $0.07 expected and guided next quarter's revenue up roughly 40%, on record indium-phosphide demand tied to AI data-center optics. It has locked in multi-year supply deals with Coherent and Lumentum worth over $100 million in customer advances and deposits — dated commercial commitments, not design-win press releases. Yet AXT trades near 48 times trailing sales after a roughly 28-fold twelve-month run, and its trend flipped from bullish to bearish on both the 7-day and 90-day view the week after that earnings beat — a sign the stock has run further than even strong fundamentals justify. nLIGHT, which builds high-power lasers for industrial cutting and, increasingly, defense directed-energy weapons, tells a similar story: revenue grew 55% year over year with a record $110 million defense backlog, and it was the only member of the group to hold its bullish trend through all of July. But at roughly 12 times sales, much of that pivot is already priced in.

Genuine deterioration, not just multiple compression. Photronics, which makes photomasks — the templates foundries use to pattern chips — fell 36% in a single May session after disclosing sequentially declining revenue and margin compression; a subsequent securities lawsuit alleges the company concealed a bottleneck in its high-end mask pipeline while publicly touting strong order patterns. Its stock has sat in the weakest trend band since mid-July. Wolfspeed, the only fully vertically integrated 200-millimeter silicon-carbide power-chip maker, emerged from Chapter 11 bankruptcy last September with $4.6 billion of debt erased, but its Mohawk Valley fab utilization ramp has repeatedly lagged management's own targets, gross margin remains negative, and Chinese competitors have crashed wafer pricing industry-wide. Wolfspeed was the cohort's worst 30-day decliner, and at roughly 3.4 times sales it now trades near what the company's own recent guidance implies is fair value for a still loss-making name — a de-rating that looks earned rather than excessive.

Growing businesses, falling stocks. IPG Photonics, which sells industrial fiber lasers for cutting and welding, and Vishay Intertechnology, which makes discrete diodes, power transistors and passive components for autos and industrial electronics, both show order books strengthening even as their shares fell hard. IPG guided second-quarter revenue higher after Q1 revenue grew 17% with book-to-bill above one and North American sales up 27%, yet trades at just 4.4 times trailing sales. Vishay beat its own guidance with 17% revenue growth and both automotive and industrial segments expanding, at a modest 2.4 times sales, with no company-specific bad news found. Both fell alongside a broader 24% drop in the Philadelphia Semiconductor Index from its June high — evidence the July slide here is sector rotation, not a verdict on either business. Vishay reports results August 5, the swing fact for whether that read holds.

The binary case. POET Technologies, a pre-revenue Toronto photonics company building an optical-interposer chip for AI data-center interconnects, saw Marvell cancel a roughly $5 million production order in April after a confidentiality breach, triggering a 46% one-day crash, its CFO's retirement, and a $400 million dilutive stock sale. It has since signed a new deal with Lumilens carrying an initial $50 million order that could grow to $500 million. Trading near 654 times trailing sales, POET's valuation still assumes a large revenue ramp that has yet to materialize.

The setup

Where it stands — Five of seven names fell in July for at least three unrelated reasons: real order loss (PLAB, WOLF), sector rotation despite growth (IPGP, VSH), and profit-taking on a stretched advance (AXTI, LASR). Would confirm — Vishay's August 5 print holding above its own $875-905M guidance would support the sector-rotation, not-order-deterioration read for IPGP/VSH. Would invalidate — A miss or cut to Vishay's Q2 guidance, or a Photronics mask-pipeline update showing no further slippage, would undercut this four-way split. Watch next — Vishay reports Q2 2026 earnings August 5, 2026; Wolfspeed's next quarterly Mohawk Valley utilization update is the following catalyst. Valuation — Range spans Vishay's 2.4x trailing sales to POET's 654x; IPGP at 4.4x and Wolfspeed at 3.4x sit near their own recent anchors, AXTI's 48x and POET's multiple remain the outliers.

Seven chipmakers that supply the optical transceivers, signal-conditioning and timing components inside AI data centers have given back roughly a fifth of their value over the past month — and for most of them, the pullback arrived in the same weeks they reported accelerating bookings and raised revenue guidance. That gap between what the businesses are reporting and what their stocks are doing is the story.

All seven names — Applied Optoelectronics (AAOI), which makes fiber-optic transceivers for hyperscale data centers; indie Semiconductor (INDI), which makes automotive radar and driver-assistance chips; MACOM (MTSI), which makes RF, microwave and optical-driver chips for data centers, telecom and defense; MaxLinear (MXL), which makes broadband and data-center connectivity chips including PAM4 signal processors; Qualcomm (QCOM), the mobile-chip giant now pushing into data-center CPUs; SiTime (SITM), which makes precision MEMS timing chips that synchronize AI server clusters; and Semtech (SMTC), which makes signal-processing chips including active copper cables for AI racks — fell over the trailing 30 days, from SiTime's -11.1% to MaxLinear's -28.2%. Six of the seven were cut from a strong uptrend to a milder one on their 30-day trend signal at the same time; indie Semiconductor was the exception, already sitting in the milder category after climbing out of a downtrend earlier this year. The synchrony argues this is a group-wide re-rating, not one or two names breaking.

The trigger was sector-wide, not company-specific. The pullback coincided with a broad chip selloff: the semiconductor index shed more than 20% from its June 2026 peak within three weeks, with Micron alone losing 13% in a single session, after Meta disclosed plans to resell surplus AI computing capacity — a move that revived doubts about the AI industry's assumption of perpetual chip scarcity and dragged the whole group lower regardless of individual results.

The businesses, mostly, kept improving. MACOM's fiscal second-quarter book-to-bill ratio hit 1.5x, the largest quarterly bookings total in company history, and management raised its data-center revenue growth guidance from 35-40% to over 60% while guiding gross margin toward 60%. MaxLinear's quarterly revenue rose 55% year over year, with its data-center optical segment up 145% on a new signal-processor ramp, and it raised its own full-year outlook again — yet the stock still fell on the print because it had already run up 22.6% beforehand. SiTime's data-center timing revenue grew 158% year over year as its new precision oscillator targets a market it estimates at $1.5 billion by 2030. Applied Optoelectronics booked its first volume order for next-generation 1.6-terabit transceivers, worth more than $200 million, from a major hyperscale customer.

Not every name confirms cleanly. Applied Optoelectronics is still loss-making at a 29% gross margin, far below peer Credo's 68%, and has been raising equity aggressively into the rally — a $500 million at-the-market program followed by a further $600 million offering, meaning shareholders are being diluted even as the stock multiplied. Semtech disclosed that its active-copper-cable product line, once expected to reach a $50 million revenue floor, will now come in below that with no ramp this fiscal year due to a rack-design change at customers — a genuine, company-specific setback layered on top of the sector selloff.

Qualcomm is diverging in a different way: its handset business faces a roughly 50% drop in Apple modem revenue between the September and December quarters as Apple's own chips ramp, but CEO Cristiano Amon says the company has "kind of replaced Apple with the data center," targeting $5 billion in data-center chip revenue by 2027 with Meta as a named future customer. At roughly 20 times trailing earnings and 4.4 times sales, Qualcomm remains the group's cheapest member. Indie Semiconductor is diverging upward for a reason that has nothing to do with AI interconnect: its $7.4 billion automotive radar and sensor backlog, including a $25 million radar production order tied to two carmakers, is an ADAS story, not a data-center one — it remains unprofitable with no documented AI-server exposure.

Verdict: on the business, the evidence mostly CONTRADICTS a fundamental-deterioration story — bookings, guidance and data-center revenue accelerated at MACOM, MaxLinear and SiTime even as the stocks fell, pointing to a market-wide multiple reset rather than cohort-wide operating trouble. On valuation, the picture is INCONCLUSIVE: Applied Optoelectronics and MACOM entered the pullback trading at 22-31 times sales and over 100 times earnings, leaving room for further de-rating, while MaxLinear's average analyst price target of roughly $68 sits close to its current price and Qualcomm's multiple looks unstretched by comparison.

The setup

Where it stands — Six of seven names cooled from a strong to a milder uptrend on the 30-day signal simultaneously, after a year of gains ranging from -14% (INDI) to +339% (MXL). Would confirm — A second straight month of cohort-wide declines alongside a guidance cut or book-to-bill deceleration at MACOM, MaxLinear or SiTime. Would invalidate — Forward orders and guidance keep rising at MACOM, MaxLinear and SiTime while the stocks stabilize or recover within the next quarter. Watch next — MACOM, MaxLinear and SiTime's next quarterly reports (expected roughly September-November 2026) for book-to-bill and data-center revenue trends. Valuation — AAOI traded near 31x sales and MTSI near 22x sales/137x earnings pre-pullback versus QCOM's 4.4x sales/20x earnings, its own multi-year low.

Four networking suppliers, one AI story, four different years

Arista Networks, Cisco Systems, Hewlett Packard Enterprise and Extreme Networks all sell the switches, routers and networking software that move data inside the giant computing centers being built for artificial intelligence — plumbing that has become one of the most closely watched slices of the AI buildout. Over the past year the group has gained an average of 83%, and on the surface that looks like a single, uniform AI trade. It isn't. Hewlett Packard Enterprise, a server-and-storage vendor that closed its $14 billion acquisition of network-equipment maker Juniper in July 2025, is up 136% over 12 months — nearly double the return of Arista Networks, the specialist that makes the high-speed Ethernet switches used to wire together clusters of AI chips, which is up just 68%, the weakest of the four.

The leader isn't who the hype suggests

Over the past 90 days the gap widens further: HPE gained roughly 68% and Extreme Networks, a much smaller campus- and edge-networking supplier, gained about 35%, while Cisco added 26% and Arista rose only 4%. Arista's own first-quarter print in May told the story — revenue grew 35% year-over-year and management raised full-year guidance, including its AI-networking revenue target to $3.5 billion, yet the stock fell 12.6% on the news as gross margin slipped to 62.4% from 63.4% the prior quarter on supply-chain-driven component trade-offs, with purchase commitments swelling to $8.9 billion. Two hyperscale customers, Meta and Microsoft, made up 42% of Arista's 2025 revenue — a concentration risk — and Nvidia's own Spectrum-X networking silicon has already overtaken Arista as the top vendor by data-center Ethernet shipments. Against a roughly 40-45x forward earnings multiple, the richest in the group, Arista's business is growing solidly but the stock is pricing in more than the numbers currently deliver — CONTRADICTS the idea that Arista is the cohort's engine.

Hewlett Packard Enterprise is the opposite case. Its Networking segment, absorbing Juniper, grew 148-151% year-over-year in recent quarters and AI systems backlog reached $5.9-6.3 billion, while the stock started this run from roughly 10-13x forward earnings and about 1.0x sales — genuinely cheap for a hyperscale-leveraged hardware supplier. That combination — accelerating segment growth from a low valuation base — CONFIRMS the price move as business-driven, even after a 100-session run in the strongest trend band the data tracks. The offset is debt: the Juniper deal pushed gross debt above $18 billion and net debt-to-EBITDA to about 2.8x, and HPE is now projected to hit only 2.45x by 2027, missing its own 2x deleveraging target, with a Justice Department settlement forcing it to divest its Instant On wireless business.

Extreme Networks, the smallest name here, doesn't really belong in an "AI-networking" story at all — its growth (net income up 206% year-over-year, five straight quarters of double-digit revenue gains, SaaS annual recurring revenue up 29%) is a campus-and-edge recurring-revenue turnaround with what the company itself describes as only secondary AI exposure. On a price-to-sales basis near 2.4x (P/E is not meaningful given its thin earnings base), the re-rating looks proportionate to the fundamentals — a separate, smaller CONFIRMS.

Cisco: multiple compression, not weaker orders

Cisco is the one name whose trend band was actually cut — from a strong uptrend to a milder one on July 20 — and it's the case the hypothesis got right. The stock is down about 11% from its June 4 high after running up 37% following a May earnings report in which Cisco raised its fiscal 2026 AI-infrastructure order guidance to $9 billion from $5 billion, with data-center switching orders up 40% year-over-year. No negative order data has emerged since. Coverage of the pullback frames it as investors weighing valuation against the rally, and Cisco's forward P/E has cooled to about 24-26x from a peak near 28x — still well above its pre-rally 17x "value" anchor. That's INCONCLUSIVE-to-CONFIRMS on the business (orders still accelerating) but the tape is repricing the multiple downward regardless — a genuine divergence between the numbers and the chart.

The technical layer

HPE, Arista and Extreme currently sit in the strongest trend band the data tracks; only Cisco has been demoted, and only on valuation, not fundamentals. Beneath all four, Broadcom's newly shipping Tomahawk 6 switch chip and Nvidia's own Ethernet-leaning ConnectX and BlueField hardware are squeezing the merchant-silicon layer that Arista itself depends on — a margin risk for the whole category, not just the laggard.

The setup

Where it stands — HPE and Extreme are driving the group's 12-month gain; Arista lags on margin compression; Cisco's de-rating tracks valuation, not orders. Would confirm — HPE's net debt-to-EBITDA falls toward its 2x target and Networking segment growth holds above 100% for another quarter. Would invalidate — Cisco discloses a downward revision to its $9 billion AI-hyperscaler order guide at its next earnings call. Watch next — Extreme Networks reports fiscal Q4 2026 results on August 5, 2026, against guidance of $330-335 million revenue. Valuation — Arista trades near 40-45x forward earnings versus Cisco's 24-26x and HPE's pre-rally 10-13x base, the widest spread in the group.

Two businesses, one label

A group of 17 energy stocks tied to the AI data-center buildout looks calm from a distance, up 15% over the past year and another 2.6% in the last month. But that average blends two businesses moving in opposite directions. Companies that own pipelines and water-handling systems and collect a fixed fee regardless of gas prices keep climbing. Companies that pull gas out of the Appalachian and Haynesville shales and sell it at whatever the market pays have been falling for three months. The split traces to a genuine collapse in the price of gas itself, not just a shift in investor mood.

The producers: a real price shock

EQT Corporation, the largest natural-gas producer in Appalachia's Marcellus and Utica shale, Range Resources, another large Appalachian gas and natural-gas-liquids producer, and Antero Resources, an Appalachian producer that also sells liquids through its own midstream system, all broke from a mild uptrend to a hard downtrend between late June and early July; EQT is down 9.9% over 90 days. Expand Energy, the Haynesville-focused producer formed from last year's Chesapeake-Southwestern merger, has sat in a downtrend continuously since April 14, a 109-session stretch that overlaps with CEO Nick Dell'Osso's February departure and an unresolved leadership search. Comstock Resources, a Haynesville shale gas producer, is down 23% over 90 days and 40% over six months, the worst of the group, after Q2 revenue fell and net income dropped year-over-year despite flat production, even as it sold a stake in its gathering unit to cut leverage. Coterra Energy, formerly a standalone Appalachian and Permian producer, merged into Devon Energy in May and no longer trades separately, leaving five starred producers instead of six.

The driver is Henry Hub, the US gas benchmark: spot prices averaged $7.72/MMBtu in January 2026 on an extreme cold snap, then were revised down to a $3.67 full-year forecast as storage rebuilt to 5% above its five-year average. EQT's realized price fell to $2.89/MMBtu in the second quarter even as it generated $330 million of free cash flow and raised production guidance. Notably, Appalachian basis actually strengthened, not widened, through mid-2026, meaning this is a national gas-price story, not a regional discount problem.

Fundamentals split the producers further

EQT's trailing price-to-earnings actually rose, from 11.3x to 12.3x, while its stock fell — earnings decayed faster than the share price, which CONFIRMS genuine deterioration rather than a cheap dislocation. Range and Antero tell a different story: Antero's trailing P/E compressed from 19x to 10x on only a modest price decline, and both remain up double digits over six months despite their bearish short-term bands — valuation here looks INCONCLUSIVE-to-cheapening, not a confirmed breakdown. Expand Energy, at 6.7x trailing earnings, is the extreme case: the user's own research explicitly calls it "dislocated," trading near 4x forward EBITDA on a low-to-mid-teens free-cash-flow yield, arguing the CEO overhang has hit the price but not the roughly 70% of core Haynesville inventory it controls. Comstock is flagged in the same research as the highest-torque and highest-leverage name in the group — the biggest winner if a forecast 2028 gas deficit arrives, the biggest loser if it doesn't.

The infrastructure side: fee-based and still compounding

Targa Resources, the dominant natural-gas processor in the Permian Basin, has held an uptrend for 232 straight sessions since mid-December and trades near 28x trailing earnings on record Adjusted EBITDA. Select Water Solutions, a produced-water management provider to shale drillers, has held its own uptrend for 217 sessions and is up 55% over six months — but its P/E has expanded from 82x to 94x, the richest multiple in the group, a stretched advance riding continued volume growth rather than a support level. Energy Transfer and Kinder Morgan, both major pipeline and gas-gathering operators, look cheaper by comparison at roughly 16x and 20x trailing earnings; Energy Transfer has signed roughly 900 million cubic feet per day of Oracle-linked gas-supply agreements and a 20-year deal to supply Meta's Hyperion data-center campus, and Kinder Morgan's $20 billion project backlog is about 60% linked to power and data-center demand — signed contracts with named counterparties, not just letters of intent. Williams Companies, which owns the Transco pipeline system, trades at the richest multiple of the three, near 34x, yet its stock is down 5% over 90 days despite remaining nominally in an uptrend — an early crack worth watching in an otherwise unbroken infrastructure rally.

The tape running ahead of the numbers

RPC, an oilfield-services company that provides pressure pumping and other completion services, was downgraded to a downtrend in the final week of July as frac-spread counts fell to 196 crews, even though its own Q2 results beat estimates with margins expanding 250 basis points as it shifted toward higher-return service lines. That gap between a bearish band and improving numbers suggests the oilfield-services tape may be pricing in more pessimism than the underlying business currently shows.

The setup

Where it stands — Pipeline and water names hold multi-month uptrends on signed data-center gas contracts; five of six starred producers sit in downtrends on a real Henry Hub price collapse. Would confirm — EQT and Expand Energy's realized gas prices stay below $3.50/MMBtu through Q3 2026 while trailing earnings keep falling faster than share prices. Would invalidate — Henry Hub's 2027 strip rises back above $4.00/MMBtu and Range or Antero's trend bands flip back to an uptrend for 20+ consecutive sessions. Watch next — Expand Energy's permanent CEO announcement and Q3 2026 earnings from EQT, Range and Antero in late October/early November. Valuation — Expand Energy trades near 4x forward EBITDA versus Energy Transfer's 8.5x and Select Water's 94x trailing P/E, its own three-year high.

A database that groups software companies by what they sell just tagged three "identity management" stocks as a hot trade, with a combined one-year gain near 50%. Pull the three names apart, though, and the average mostly describes one company licensing operating-system software to carmakers — not a re-rating of the login-security business.

The three names are BlackBerry Ltd. (BB), the former smartphone maker that now earns most of its revenue licensing the QNX operating system used in cars and factory robots, alongside a shrinking corporate-messaging security unit; Okta Inc. (OKTA), which sells cloud software that verifies employee and, increasingly, AI-software-agent logins into corporate networks; and SailPoint Inc. (SAIL), which re-listed on the stock market in February 2025 and sells software that decides which employees, contractors and now automated AI agents are allowed to touch which company systems.

The averages hide three different stories. BlackBerry is up 135.5% over twelve months and Okta 49.2%, while SailPoint is down 18.7% over the same span — the group's headline "+50%" figure is a simple average dragged up almost entirely by BlackBerry. The one-month figures invert the picture: BlackBerry alone fell 33.6% in the past 30 days after its 265%-in-three-months QNX-driven run, while Okta and SailPoint were roughly flat to higher.

BlackBerry's move is not an identity story. Its fiscal first-quarter results showed QNX revenue up 26% and Secure Communications up 24%, with adjusted EBITDA more than doubling, and shares jumped on the beat — but the growth is automotive and robotics licensing, not corporate login software, following the company's 2025 sale of its Cylance security unit to Arctic Wolf. At 89.9x trailing earnings and 9.1x trailing sales, one analyst already calls the stock roughly 34% overvalued. Verdict: business and tape agree on direction, but the business has nothing to do with identity software, and valuation looks stretched — CONTRADICTS the category thesis.

Okta's rally is real but concentrated in a single earnings day. Roughly 84 of its 90-day gain traces to a single May 29 session, when shares jumped 30% after a report showing remaining performance obligations up 16%, current RPO up 12%, and net revenue retention rising to 107%. New AI-agent-related products made up about 25% of new bookings with a 40% average deal-size uplift — a rare case of machine-identity demand showing up as disclosed revenue rather than talk, per Okta's own filings and a bullish Forbes writeup. But the multiple has already moved: price-to-sales expanded from 4.6x in May to 8.0x now, and the stock trades above the roughly $120 analyst consensus price target. Verdict: CONFIRMS an AI-driven demand story, but valuation already prices much of it in.

SailPoint shows the clearest gap between the business and the stock. Annual recurring revenue grew 26% and its cloud-subscription (SaaS) ARR grew 36%, with net revenue retention of 113% — better than Okta's — yet shares fell 21% on the day it reported those numbers because management gave cautious profit guidance. The stock trades at 6.7x sales with a $7.5 billion market value, well below its roughly $12.8 billion valuation at its February 2025 relisting despite the intervening growth.

The technicals confirm the fragility, not the strength. SailPoint's price-trend signal flipped from a sustained downtrend to an uptrend only in the final two trading sessions of available data, after 16 straight sessions the other way — too fresh to call durable. BlackBerry's uptrend signal never reacted to its 34% one-month drop, meaning the trend indicator is lagging actual price action. On competition, Microsoft is raising list prices for its rival Entra ID identity product this July, and CyberArk remains priced at a premium to peers — pricing pressure on identity vendors is not yet visible in 2026.

The setup

Where it stands — Okta's AI-agent bookings are a disclosed revenue driver already reflected in a stretched multiple; SailPoint's stock lags improving fundamentals; BlackBerry's gain is unrelated automotive software. Would confirm — SailPoint's uptrend signal holding for several more weeks and margin guidance improving at its next quarterly report. Would invalidate — SailPoint's trend flip reversing within days, or Okta's AI-agent bookings share falling back below 25% of new bookings next quarter. Watch next — SailPoint's fiscal second-quarter FY27 report, expected around September 2026, for ARR growth and margin guidance. Valuation — OKTA at 8.0x forward sales versus 4.6x three months ago; SAIL at 6.7x trailing sales versus a $12.8B re-listing valuation; BB at 9.1x trailing sales, called ~34% overvalued by one analyst.

Fastly, which sells the content-delivery and edge-computing network that keeps websites loading fast and fending off attacks, jumped 25% in July. Elastic, maker of the Elasticsearch search software that companies use to hunt through logs and, increasingly, feed AI applications, rose 13%. Five9, which runs the call-center software agents use to field customer calls, climbed 17%. Q2 Holdings, whose software powers the online and mobile banking apps of regional banks and credit unions, gained 16%. GitLab, the platform where software teams store code and now attach an AI coding assistant called Duo, added 8%. Only Lightspeed, which sells point-of-sale and e-commerce software to independent retailers and restaurants, fell — down 6% — even though its own numbers improved.

That split matters because these six sit at the small end of software, the segment investors spent two years marking down as most vulnerable to both AI-driven disruption and takeover bids. The July rally raises the question of whether the group is genuinely re-rating on improving fundamentals or simply enjoying an earnings-season bounce that has, in a few cases, already run ahead of the businesses underneath it.

Where the business supports the tape. Elastic's most recent quarter showed net revenue retention holding at 112% and Elastic Cloud revenue up 22%, with the company reporting 470-plus customers now running AI workloads on its platform and acquiring Jina AI to deepen its vector-search technology. Its price-to-sales multiple still sits at 3.74x, the cheapest in the group. Fastly's revenue growth is accelerating — security revenue up 47%, its newer compute segment up 67% — and no evidence of a major customer loss surfaced in its latest results; its current 4.99x sales multiple is still below the 8.1x peak it hit in May, so the July move has not fully closed that gap. Q2 Holdings raised its full-year subscription guidance after a record bookings quarter and a 28.6% adjusted EBITDA margin — genuine improvement, though its trailing price-to-earnings ratio of roughly 51x on 10-11% revenue growth is still what the desk's own prior notes call stretched.

Where the tape has run ahead. GitLab's first-quarter results beat estimates and disclosed roughly $20 million in Duo AI consumption run-rate, a rare hard number behind the AI-attach story — but full-year guidance decelerated to 16-17% growth, and about 20% of annual recurring revenue remains under seat-contraction pressure from customer layoffs. Its forward sales multiple has jumped from 4.0x in May to 5.54x, a re-rating that has outpaced the slower guide, not confirmed it — a gap widened further by unconfirmed takeover chatter involving Datadog that neither company has verified. Five9's AI-specific revenue grew 68% year over year and now makes up 13% of subscription revenue, management said, a genuine counter to the fear that AI voice agents would gut its business — but its own sales multiple has nearly doubled, from 1.12x to 1.77x, since May, meaning much of that reassurance is already priced in.

The outlier. Lightspeed's fundamentals moved the opposite direction from its stock. The company completed a strategic review, stayed public, upsized its buyback, turned EBITDA positive and grew its core retail and hospitality segments 24% — yet shares fell 6% in July, the one clean divergence in the group and the cheapest stock on a sales-multiple basis at 1.15x.

The price data show why the technical picture is less settled than the month's returns suggest: Elastic, GitLab and Q2 Holdings all flipped from bearish to bullish trend readings only on the final trading day of July, with no multi-week confirmation yet, and Q2 Holdings whipsawed back into a strong-bearish reading as recently as June. Only Five9 and Fastly have held bullish trends since June.

The setup

Where it stands — Five of six names rallied on July earnings, but three flipped bullish only on the month's last trading day. Would confirm — Elastic, GitLab and Q2 Holdings hold bullish trend readings for three-plus consecutive weeks into September. Would invalidate — Those three revert to bearish trend readings before mid-August, repeating Q2 Holdings' June whipsaw. Watch next — GitLab's fiscal second-quarter report, expected September 2026, for Duo AI run-rate growth beyond $20 million. Valuation — GitLab trades at 5.54x forward sales versus 4.0x three months ago; Elastic trades at 3.74x, the group's low.

Bitcoin and Ether, the two largest cryptocurrencies, have essentially stopped falling over the past month after a year in which each lost more than 40% of its value. Spot exchange-traded funds (ETFs) that let ordinary investors buy the coins through a brokerage account are, for the first time since spring, taking in more cash than they're losing. That would normally be good news for the handful of publicly traded companies whose entire business is holding or trading these coins. It isn't. MicroStrategy (MSTR), the software company turned Bitcoin holding vehicle now branded Strategy, Bitmine Immersion Technologies (BMNR), a small Las Vegas firm that has become the largest corporate holder of Ether, and Coinbase (COIN), the largest U.S. crypto exchange, have all kept falling even as the coins they're built on found a floor.

The coins: a real, if young, stabilization. Bitcoin's price trend was upgraded from a steep downtrend to a milder one on July 28 and has held there since; Ether's equivalent upgrade came nine days earlier, on July 20, and has lasted about two weeks. Both trends had been deeply negative since mid-June and, on a 365-day view, had been strongly positive a year ago — confirming the damage is now largely in the past rather than still unfolding. The evidence behind Ether's turn is more solid: spot Ether ETFs pulled in $365 million in July, their best month of 2026, after two months that shed more than $1 billion combined. Bitcoin's ETFs, by contrast, managed just $205–221 million in July — the smallest monthly haul since the funds launched in January 2024 — after $2.4 billion and $4.5 billion left in May and June. Bitcoin held on exchanges, where coins are ready to be sold, sits at 6.6% of supply, the lowest since 2017, consistent with coins moving into longer-term storage. Still, major on-chain research firms including CryptoQuant, Glassnode and Benjamin Cowen still peg the fourth quarter of 2026, not now, as the likeliest cycle bottom, with more than 8 million Bitcoin still sitting at a paper loss. Meanwhile, several public Bitcoin miners have been selling down treasuries to fund a pivot into AI data centers, an active supply overhang that hasn't fully cleared.

The proxies: a company story, not a coin story. Strategy's trend remained in its steepest downtrend through the end of July, even as Bitcoin's eased. The company now holds roughly 843,775 Bitcoin at an average cost of $66,385 each — above the roughly $63,000 spot price, meaning its treasury book is underwater in aggregate for the first time this cycle. Strategy also carries more than $1.5 billion a year in preferred-stock dividend obligations and only about 18 months of cash to cover them; in May it sold Bitcoin to help fund those payments, its first sale since 2022. Its shares trade at 0.68 times the value of its Bitcoin holdings — a discount, down from a 2.6-to-3.0-times premium in late 2024 — but the forced-seller dynamic around the dividend is a company overhang that would persist even in a coin recovery. Bitmine's diluted share count grew 69% in two quarters (325.7 million to 551.8 million) as it issued stock to keep buying Ether, and it books multibillion-dollar accounting losses tied to swings in its Ether holdings; its stock traded at 0.82 times book value, also a discount, even after bouncing 20% in the past month. Coinbase, the one name here that actually runs an operating business rather than just holding coins, reported second-quarter revenue down 19% year over year and missed estimates on both trading commissions and subscription services; its stock fell 11.6% over 30 days even as the coins it trades stabilized. Its subscription line — record $20 billion in customer stablecoin balances and rising custody and exchange market share — held up better than trading revenue, giving it a partial fundamentals floor trading revenue alone doesn't provide, but at 64 times trailing earnings against a shrinking top line, the stock's de-rating looks tied to its own numbers, not the coin tape.

Verdicts. For Bitcoin and Ether, the business case (ETF demand, exchange supply) is INCONCLUSIVE for Bitcoin and CONFIRMS for Ether, and neither is a settled bottom by on-chain researchers' own account. For MicroStrategy and Bitmine, the sub-1.0x valuation to their crypto holdings suggests a possible dislocation, but dilution and dividend obligations mean the de-rating partly CONTRADICTS a pure coin-recovery story — these are increasingly company problems layered on top of coin exposure. For Coinbase, the 64x earnings multiple against declining revenue looks like a JUSTIFIED de-rating tied to its own results, not a mispriced coin proxy.

The setup

Where it stands — Bitcoin and Ether have held milder downtrends for one and two weeks respectively; MicroStrategy, Bitmine and Coinbase remain in steep downtrends and keep falling. Would confirm — Bitcoin spot-ETF monthly net flows exceed $1 billion again and MicroStrategy's mNAV holds above 0.68x for multiple weeks. Would invalidate — Bitcoin or Ether trend downgrades back to a steep downtrend, or exchange balances start rising again. Watch next — August spot-ETF flow data and Strategy's next quarterly disclosure of BTC holdings and cash runway. Valuation — MSTR trades at 0.68x its Bitcoin holdings (was 2.6–3.0x in late 2024); COIN at 64x trailing earnings on declining revenue.

Eight companies that build the physical infrastructure of the AI boom — servers, storage systems and the contract manufacturing that assembles them for cloud providers — have seen their shares rise an average of 81% over the past twelve months as hyperscale cloud companies keep buying racks of AI hardware. The move is broad: seven of the eight are higher over that stretch, from Jabil's 41% gain to Dell's 205%. But the group is starting to split. Two of the contract manufacturers, Celestica and Flex, just had their price trends cut from strong to milder bullish readings, a shift that coincided with a broader market worry about whether the cloud giants funding this buildout can keep spending at the current pace — even though both companies' own results kept improving.

The eight, and what they actually sell

Dell Technologies, which sells PCs and increasingly builds AI-optimized servers for enterprises and cloud providers, has been the standout: shares up 205% and its uptrend unbroken since March 31, the longest bullish streak in the group. Super Micro Computer, which builds AI server racks and remains under scrutiny after two straight years of adverse findings on its internal financial controls, is the lone decliner, down 52% even as its order book improves. NetApp, which sells data-storage systems including the all-flash arrays used in AI data centers, is up 71%. Celestica, a contract manufacturer that builds networking and computing hardware for hyperscale customers, has gained 66%. Jabil, a diversified contract manufacturer whose Intelligent Infrastructure unit builds AI data-center hardware, is up 41%. Flex, a contract manufacturer spinning off its power-and-cooling infrastructure business, has gained 128%. Sanmina, a contract manufacturer that bought ZT Systems' AI server-integration business, is up 60%. Penguin Solutions, a small AI cluster integrator formerly known as Smart Global Holdings, has gained 123%.

The fundamentals mostly agree with the tape

Dell exited its fiscal first quarter with a record $51.3 billion AI-server backlog after booking $24.4 billion in new orders, and its server division's operating margin expanded to 10.5% even as rising memory-chip costs squeeze the business, the company told investors. Jabil raised its full-year AI-related revenue guidance to about $13.6 billion, up from $9 billion a year earlier, with its Intelligent Infrastructure segment growing 52% year over year. Sanmina's newly acquired ZT Systems unit is now contributing $1.1 billion a quarter, while its legacy business grew 17% on its own. NetApp posted record all-flash storage revenue of $4.2 billion for the fiscal year, up 11%, with roughly 500 AI-related storage wins in one quarter. Penguin Solutions' quarterly sales rose 48% year over year to a record $479 million.

Super Micro is the case that separates business from tape most clearly: it disclosed preliminary results showing more than $60 billion in new orders and doubled its gross-margin guidance to 15%-17%, a report strong enough to lift Dell and HPE shares in sympathy — yet its own stock stays down 52% over the year because of an unresolved SEC subpoena and two straight years of adverse internal-controls findings that predate the recovery. Celestica shows the opposite split: it raised full-year revenue guidance to $20.5 billion from $19 billion and lifted earnings guidance, yet shares fell 35% from their June high the same week, a decline commentators tied to Alphabet's negative free cash flow and $84.75 billion equity raise rattling confidence in hyperscaler capital spending broadly. Flex's downgrade landed the same window, even as it announced a spin-off of its faster-growing power-and-cooling unit.

Valuation: much of the re-rating is already spent

Trailing price-to-earnings multiples among the contract manufacturers now sit well above historical norms — Celestica near 41x, Flex near 55x, Sanmina near 50x, Penguin near 89x — while Dell, at roughly 32x trailing and about 18x forward earnings against 27% projected profit growth, remains the cheapest name leveraged to AI-server demand. Super Micro is cheapest overall at roughly 16x trailing earnings, but that discount reflects unresolved legal risk, not undervalued fundamentals.

Business fundamentals CONFIRM the price advance for six of the eight — Dell, Celestica, Jabil, Flex, Sanmina and NetApp show accelerating AI-linked revenue or backlog. Super Micro's fundamentals point up too, but governance risk keeps its stock down, a genuine divergence between business and tape. Valuation is closer to CONTRADICTS for the contract manufacturers: most have re-rated to multiples their own recent history calls largely exhausted, leaving less room for further gains even if server demand keeps growing.

On the tape, Dell's uptrend has held without interruption since March 31 — the longest streak in the group — while Celestica and Flex were both cut from strong to milder bullish readings in late June and mid-July, coincident with a broad AI-momentum pullback rather than their own results. Super Micro's trend has oscillated between mildly bullish and mildly bearish since early July, tracking its unresolved legal overhang rather than its improving order book.

The setup

Where it stands — Dell holds the group's longest uptrend on record AI backlog; Celestica and Flex were downgraded despite raised guidance, on hyperscaler-capex worry. Would confirm — Celestica and Flex resume gaining alongside Dell if Q3 hyperscaler capex commentary reaffirms spending plans intact. Would invalidate — A hyperscaler cuts 2027 capex guidance, or Dell's ISG operating margin falls below its current 10.5%. Watch next — Dell's next ISG margin print and Super Micro's SEC subpoena resolution, both due within the next two quarters. Valuation — Dell trades near 18x forward earnings versus a 27% profit-growth outlook; Celestica near 41x trailing, a re-rate its own coverage calls largely complete.