CPI Cools Hard, Semis Snap Back — and IBM’s Still the Worst Day Since 1961
Premarket briefing — Tuesday, July 14, 2026 · 7:10 AM ET · brought to you by SignalDeck.live.
Situational Awareness
📊 June CPI came in well below expectations — the day’s dominant macro catalyst. Headline -0.4% MoM (largest single-month decline since April 2020, vs. -0.1% expected); annual rate 3.5% vs. 3.8% expected. Core CPI flat MoM, 2.6% YoY vs. ~2.8-2.9% expected. Gasoline-price relief was the main driver. July Fed hike odds collapsed from 42% (Monday) to 17% on the print; September hike odds still ~60%. Nasdaq-100 futures +1% on the release.
💾 Semis/memory are snapping back hard — essentially yesterday’s selloff in reverse. SK Hynix ADR +9.0%, SanDisk +6.8%, Micron +5.1%, Western Digital +5.9%, Seagate +5.1%, KLAC +6.4%, Applied Materials +6.5%, Lam Research +5.8%, Marvell +5.4% — nearly the entire complex up 5-9% premarket. Drivers: the CPI-driven risk-on tone, plus a genuine V-shaped overnight reversal in Asia (KOSPI and the Nikkei 225 both staged sharp intraday reversals after Monday’s rout). This corrects an earlier version of this report, which wrongly attributed yesterday’s move and today’s bounce to a same-week Micron earnings beat — MU’s actual last print was 2026-06-24, three weeks earlier, not overnight. Yesterday’s real driver was SK Hynix’s volatile Nasdaq ADR debut plus fresh concern that HBM production yields are improving faster than expected (a threat to the memory-scarcity thesis); today’s bounce is the CPI/Asia-reversal story above, not a fresh Micron catalyst.
🇮🇷 Iran/Hormuz escalation continues, and just got sharper. Trump reinstated the Iranian port blockade AND is now demanding a 20% fee on all other cargo transiting the Strait — an escalation beyond Monday’s announcement. Crude remains elevated. This is the live countervailing risk-off force against the CPI-driven risk-on tone.
📉 IBM is still the standalone story on its own name. Preliminary Q2 revenue $17.2B (+1%, missed $17.86B consensus); non-GAAP EPS $2.93 (+5%, missed $3.01 consensus). Stock closed near -22.8% Monday, worst single session since 1961. Full analysis below.
Top Calls
TSEM +19.3% · news_driven · High — Tower Semiconductor announced a $3B capacity expansion in Japan (METI-backed, ~$1B in government grants) targeting AI/data-center demand, alongside a raised 2028 outlook ($3.6B revenue / $1.2B net profit guide vs. ~$3.17B street). A real, dated, company-specific catalyst — not sector beta.
⚠ CLSK +18.8% · unconfirmed catalyst · N/A — CleanSpark screened as a top gainer; sources found are inconsistent and none confirm a same-day catalyst (one references early-July Bitcoin production/hashrate records already a couple weeks old, another flags a bearish earnings outlook). Flagging, not recommending.
No fresh-EP names today — Pass 1 (earnings-window screener) was empty, the fourth straight session with zero fresh-earnings movers (consistent with the mid-July lull ahead of Q2 bank earnings this week). No governance-blacklist names surfaced.
The Setups
TSEM
The catalyst is a $3B, government-backed capacity expansion in Japan — a dual-track buildout of 300mm silicon photonics, SiGe, and advanced-packaging capacity, explicitly aimed at AI and data-center customers, alongside a 2028 outlook raise above current street estimates. Why it matters: this is a specialty-foundry name making a multi-year, subsidized capital commitment specifically to AI-adjacent demand — a structural bet, not a one-quarter print. What the price reflects: +19.3% on a real, company-specific announcement, well outside yesterday’s sector-wide semis move.
CLSK (flagged, not a call)
Screened as a top premarket gainer, but the sourcing is inconsistent: one report cites early-July Bitcoin production/hashrate records that are already stale relative to today; another flags a bearish earnings outlook alongside strong operational metrics. No same-day catalyst was confirmed. Per our own discipline: verify before acting.
Bottom line (today’s screener): TSEM is today’s one real single-name catalyst; everything else in the semis/memory complex is a broad, CPI- and Asia-reversal-driven sector rebound, not idiosyncratic news.
Second-Degree: Global Markets → US Read-Through
Added per operator request — this becomes a standing part of how this report treats cross-market and geopolitical moves, not a one-off.
The Asia rebound is technical, not fundamental — the underlying risks are still live
The Nikkei 225 and KOSPI both closed up ~0.7% today in a “V-shaped” reversal, and US semis/memory are riding that lead. But the mechanism matters: this was confirmed bargain-hunting after a historically oversold plunge (KOSPI -8.95% Monday, its 7th circuit breaker of the year, described in Korean coverage as “historic oversold territory”), backed by real institutional buying (₩1.62T net purchases by mid-morning) — not a resolution of what caused Monday’s crash. Monday’s KOSPI selloff was driven by two named factors: escalating US-Iran war risk, AND worries about a peak in memory-semiconductor pricing (the same HBM-yield-improvement fear flagged above). Neither of those two drivers actually resolved today — Iran/Hormuz tension is worse, not better (see below), and no new information on memory pricing emerged. This is a technical snapback trading on oversold conditions, not new facts. Treat today’s semis/memory strength as fragile until proven otherwise — the same two triggers that caused Monday’s plunge are both still live and could reassert.
Oil is a bigger, sharper move than the CPI print suggests, and it cuts against the CPI’s own dovish signal
Brent crude is near $85/bbl — up ~3.8% today alone, on top of a 9.6% surge yesterday (roughly +13-14% in two sessions), the highest level in a month, as US-Iran hostilities entered a third consecutive day. Strait of Hormuz transits fell more than 50% week-over-week. Direct sector read-throughs: energy E&P/integrated (direct beneficiary — already an active catalyst from Monday), airlines (jet fuel in North America has spiked 95% since the war began — a severe, quantified cost headwind, not a vague one), and petroleum-feedstock-dependent chemicals/plastics names. A structural vulnerability worth flagging: US strategic petroleum reserves have reportedly been depleted from earlier drawdowns this year, leaving less buffer against a further supply shock than the market had in March/April.
The sharper point, and the reason this belongs in Situational Awareness, not just a footnote: today’s celebrated CPI cool-down was driven substantially by falling gasoline prices in June — and the report itself notes that relief was “tied to easing geopolitical tension before the weekend’s re-escalation.” Oil has since risen ~13-14% in two days. The disinflationary signal the market is pricing off today’s print is already partially stale relative to what’s happening in the oil market right now. If Brent holds anywhere near $85 into July, that shows up in the next CPI print as a headwind, working directly against today’s rate-hike-odds relief. The Fed-policy read from this morning’s data is thus more fragile than the “hike odds collapsed from 42% to 17%” headline implies on its own.
Special Report: IBM Q2 2026 — Segment Breakdown
Three reported segments, all preliminary (sub-segment dollar detail isn’t disclosed until 7/22):
Software +5% YoY reported — the only segment that grew this quarter.
Consulting flat reported, +1% at constant currency.
Infrastructure -7% YoY — a mainframe (Z systems) shortfall and the attached software stack (Transaction Processing specifically called out weak). CEO Arvind Krishna’s stated reason: clients shifted late-June capex toward servers/storage/memory ahead of expected price increases tied to supply constraints — a pull-forward into hardware, at software and infrastructure’s expense this quarter.
FY2026 guidance was maintained, not cut: Consulting is guided to accelerate to low-to-mid-single-digit growth; Infrastructure to be down low-single-digits (~0.5pt consolidated drag).
The wrinkle that changes the read: how much of “+5%” is Confluent, not organic growth
IBM closed its $11.59B all-cash acquisition of Confluent (streaming data platform) on 2026-03-17 — so Q2 is the first full quarter of Confluent consolidated. A pre-earnings BofA estimate (a third-party estimate, not an IBM disclosure) put Confluent’s contribution at roughly $340M/quarter and ~5 points of Software segment growth. IBM’s own Q1 2026 actual had Software at +11% reported (+8% cc) — meaning the reported growth rate roughly halved quarter-over-quarter, in the same quarter the acquisition tailwind got bigger. That combination only reconciles one way: organic software growth decelerated hard, plausibly to something close to flat. This is an informed inference from IBM’s own reported figures plus a third-party estimate — not a confirmed IBM disclosure. The organic/inorganic split should get clarified on the 7/22 call.
Special Section: Investible Software Companies — Ranked Read-Throughs
Mapped to the specific IBM sub-segments still demonstrably growing (Red Hat: low-teens in Q1; Automation: +10% in Q1), not just superficial “also software” matches:
Nutanix (NTNX) — the cleanest Red Hat comp: hybrid-cloud platform software sold to the same infrastructure buyer, riding the same VMware/Broadcom-displacement dynamic underpinning Red Hat’s midteens FY guide.
Snowflake (SNOW) — shares the modern-data-stack budget line Confluent feeds; streaming pipelines terminate in analytical platforms like Snowflake. Caveat: Confluent’s own standalone Q2 growth isn’t disclosed yet, so this is inference on inference.
MongoDB (MDB) — operational database layer on the same data-infrastructure wallet, heavy deployment overlap with Kafka/Confluent architectures.
GitLab (GTLB) — the platform-engineering buyer is the HashiCorp buyer (Terraform, Vault). HashiCorp and Confluent are both acquired and not investible — that’s precisely why IBM bought them — making GTLB the closest public proxy for the infra-automation budget line.
Datadog (DDOG) — flagged as a partial mapping only: observability spend is a cloud-consumption proxy, not a direct comp to any IBM sub-segment.
Rejected as superficial mappings — worth naming explicitly, not just omitting: - UiPath, Pegasystems, Appian — name-match trap. IBM’s “Automation” is infrastructure automation (Terraform/Vault, orchestration) sold to platform teams; RPA/BPM is business-process automation sold to a different buyer entirely. No real read-through. - Palantir, C3.ai — application-layer AI with demand drivers (government, forward-deployed engagements) disjoint from every IBM sub-segment. - Hyperscalers (MSFT/AMZN/GOOGL) — a hybrid-cloud read-through is a rounding error on their consolidated numbers.
The other side of the pull-forward
Krishna’s own explanation — customers pulling server/storage/memory purchases forward ahead of expected price increases — is demand-side corroboration of the memory-tightness thesis, directionally supportive of Micron and storage-adjacent names (WDC, STX). Correction: an earlier version of this section cited a Micron “AH beat this same week” as a second, independent corroborating data point — that was wrong (see Situational Awareness correction above; Micron’s real print was 6/24, three weeks earlier). Removing that leg. Standard pull-forward caveats still apply (it borrows from future quarters), and there’s now a genuine bearish counter-data-point worth weighing: reports circulating 7/13 that HBM production yields are improving faster than expected, which is exactly the kind of development that would erode the scarcity premium this whole thesis depends on.
Overall verdict
This print makes the software-comp basket less interesting near-term, not more — with one carve-out. The organic-deceleration wrinkle means IBM’s “resilient software” headline validates nothing (it’s substantially acquisition-driven), and IBM’s own explanation for the miss is that enterprise wallets rotated away from software toward hardware — a budget-cannibalization headwind that isn’t IBM-idiosyncratic and applies to the whole basket above for the next couple of quarters. The carve-out is NTNX: the Red Hat midteens FY guide was maintained, and VMware-displacement is a structural driver, not a budget-cyclical one — the one name here where this print arguably nets positive. Everything else is a watch-list with a dated catalyst: the 7/22 call, where IBM should disclose the organic/inorganic software split. If organic Software prints materially positive there, this basket re-rates from “avoid the sympathy drag” to “buy the overhang.” Until then, the -23% is IBM repricing its own narrative — not an invitation to buy its neighbors.
Second-Degree: Does This Print Support Neoclouds and Chips?
Added 6:45 AM ET — a follow-on question worth its own section rather than a footnote.
The tempting trade after a print like this is symmetry: IBM says infrastructure demand is hot, therefore buy everything downstream — memory, GPUs, connectivity, and the GPU-rental landlords. The honest answer is the symmetry doesn’t hold. One leg is well supported; IBM adds a small brick to it. The other leg was already cracking two weeks before IBM reported, and IBM’s own commentary arguably makes it worse, not better.
Chips: the case holds — but be clear about what’s carrying it
Krishna’s capex-shift comment is about traditional enterprise on-prem buyers (mainframe shops, Power server accounts, storage-array customers) — not hyperscalers. What it tells you: a third buyer population that usually sits out these cycles is now panic-buying hardware on supply-constraint fears. That’s real corroboration of tightness, landing most directly on commodity memory (Micron, SanDisk) — but it’s a footnote, not the thesis.
The thesis is the hyperscaler data, and it’s moving the wrong way for bears. This account’s own memory-cycle framework (The AI Memory Trap, 6/25) named the #1 top-of-cycle signal as hyperscaler capex growth stepping down while the dollar level stays high. The opposite is happening: the Big 4 guided ~$725B combined 2026 capex vs. ~$410B in 2025 (itself up from ~$226B in 2024) — every one of the four raised the bar (Amazon $125B→$200B, Google $91B→$175-185B, Meta $72B→$115-135B, Microsoft $90B→$110-120B). Signal #1 has not fired — it’s receding, not approaching.
The scarcity is physical: data-center GPU lead times 36-52 weeks; SK Hynix AND Micron’s entire 2026 HBM output sold out; HBM3E lead times 20-26 weeks. NVDA did $75.2B in data-center revenue in a single quarter; AMD’s $5.8B grew 57% — faster than Nvidia’s own rate — with Goldman taking AMD’s target to $640 from $450 (7/5) on agentic-AI CPU demand.
By the framework’s own taxonomy: Camp A (Micron, SanDisk) — best supported right now (sold-out HBM gives earnings visibility most cyclicals never get), but still the names whose earnings evaporate when pricing turns; ride with the cycle clock in view. Camp B (Marvell, Astera Labs, Credo) — capex re-acceleration is exactly what high-multiple connectivity needs, since deceleration (not present here) is their kill mechanism. Broadcom/Arista remain the risk-adjusted core regardless.
One early flag, honestly labeled sub-actionable: Q3 DRAM contract prices are still projected +13-18% QoQ, NAND +10-15% — but at least one source calls this a real slowdown in the rate of increase versus earlier this year. The framework’s threshold is prices stop rising — that hasn’t happened. But the second derivative of pricing just went from steepening to flattening — the first grain of sand, worth a monthly check, not a trade yet.
A second flag, added on correction (7/14): reports circulating 7/13 attribute part of Monday’s semis selloff (MU -4.3% to -5.8%, alongside NVDA/AMD/MRVL/WDC/AMAT) to a fear that HBM production yields are improving faster than anticipated — the exact mechanism that would erode the scarcity premium behind “sold-out 2026 HBM.” This directly tensions with the sold-out-capacity data point above. Both can be true at once (sold out today, yields improving fast enough to worry about tomorrow) — but it’s a second, independent crack in the same wall as the pricing-rate deceleration, not a one-off. Weight it accordingly: still a flag, not yet a reason to act.
Net: chips yes, at roughly pre-IBM conviction, plus a small enterprise-demand kicker. IBM is corroboration, not cause.
Neoclouds: the link doesn’t hold — and the group has a live, independent problem
Two separate reasons the answer is no. First, the inference from IBM breaks at the first joint: IBM’s buyers are enterprises purchasing their own servers/storage/memory — a substitute for renting neocloud GPU capacity, not a complement. Read honestly, IBM’s print cuts mildly against CoreWeave/Nebius, not for them: marginal enterprise AI dollars going into owned infrastructure are dollars not going into rented capacity. IBM’s own Red Hat/watsonx-on-OpenShift-AI strategy is a bet on enterprises running hybrid AI themselves — not a data point for the rental layer.
Second, and more important: the neoclouds don’t need a weak inference from IBM, because they took a direct structural hit on 7/1/2026 — two weeks before this print, entirely unrelated to it. Meta announced “Meta Compute,” commercializing its internal AI infrastructure and renting raw GPU capacity by the hour — CoreWeave’s and Nebius’s own business model, backed by a company with $115-135B of 2026 capex. CRWV fell ~14%, NBIS ~17% in a session, and the market was right to react: Nebius carries a $27B Meta contract, CoreWeave a $21B one — the group’s largest customer just announced it’s becoming their competitor. This account’s own framework named CoreWeave “the canary in the coal mine” for this cycle. The canary didn’t die, but it just watched its biggest tenant start building next door.
The bull rebuttal is real: CoreWeave still shows a $99.4B backlog and 3.5GW contracted power; Nebius targets 800MW-1GW. As long as GPU scarcity stays this extreme, anyone with live capacity can fill it. But a backlog is only as good as its counterparties’ need to keep renting, and the largest one just told you it intends to sell capacity too.
Net: neoclouds no. Not because demand is weak — it isn’t — but because the IBM read-through is nonexistent-to-mildly-negative, and the group’s structural risk (customer concentration curdling into customer-as-competitor) went from theoretical to dated-and-priced on 7/1.
Rank order
Chips, clearly first. Hyperscaler capex acceleration is intact and strengthening, physical scarcity is verified across GPUs and HBM, and IBM adds a third buyer segment on top. Broadcom/Arista as the risk-adjusted core, Camp B (Marvell/Astera/Credo) as the leveraged play on acceleration continuing, Camp A (Micron) as the highest-beta expression — watch the DRAM/NAND price-increase-rate every month.
Neoclouds, a distant second — effectively a pass. The hoped-for symmetry from IBM’s print doesn’t exist, and the group absorbed an independent structural blow (Meta Compute) landing hardest on exactly the name this framework already flagged as most fragile. Want the AI-infrastructure trade? Own the shovels, not the leveraged landlord renting shovels to a customer who just opened a hardware store.
One consistency note: nothing here contradicts The AI Memory Trap. That piece’s kill conditions are second-derivative signals, and today’s data says the second derivative is still positive — the framework, applied honestly, is bullish here and now on the suppliers while keeping the same tripwires armed.
Bottom line: IBM’s own Q2 numbers argue against the “software held up” narrative the headline implies — the growth is substantially Confluent, not organic, and NTNX is the one software-comp name this print doesn’t undercut. One level deeper: the print doesn’t support the neocloud leg of an AI-infrastructure trade (that group has its own, unrelated problem as of 7/1), but it adds a small corroborating brick to a chip/memory thesis that was already strong and whose key warning sign — hyperscaler capex deceleration — has not fired.
📊 Read this on SignalDeck.live.
Not financial advice. For informational purposes only. Do your own research.