28 October 2026 | Information cutoff: 06:48 Europe/London
Morning View
The overnight evidence shifted the AI debate back from financing structures to delivered earnings. SK hynix reported an 80% operating margin and said supply will remain structurally short through 2030, Seagate converted nearline scarcity into a 57.8% non-GAAP gross margin and record free cash flow, while Corning and Teradyne showed that optical connectivity and semiconductor test are compounding alongside accelerators. The common thread is operating leverage in bottlenecks that were initially treated as secondary beneficiaries of the AI buildout.
The most important cross-sector development is the broadening of compute architecture. Oracle is simultaneously scaling Nvidia Vera Rubin to multi-gigawatt deployments, becoming the first hyperscaler outside Google to offer Google TPUs and adding Cerebras inference. Separately, OpenAI is reportedly considering buying Google chips. Nvidia remains dominant in frontier training, but inference economics are creating a more heterogeneous market in which cloud distribution, networking and power efficiency matter as much as peak accelerator performance.
The morning is incrementally positive for technology fundamentals but less uniformly positive for positioning. Memory, storage, optics, test and cyber resilience support upward estimate revisions; NXP suggests the industrial and automotive semiconductor recovery is becoming broader; and American Tower’s CoreSite growth confirms scarce interconnected capacity still has pricing power. Enphase is the counterexample: a credible data-center power opportunity does not offset a sharp near-term contraction in its core solar business. Today’s Microsoft, Alphabet, ServiceNow, Fortinet, Tenable, KLA, Vertiv and Equinix results will determine whether end-customer revenue is catching up with the supplier earnings already visible.
1. SK hynix turns HBM scarcity into an 80% operating margin and locks in more than half of 2027–28 capacity
SK hynix reported on 28 October Q326 revenue of KRW104.6tn, up 32% sequentially, and operating profit of KRW83.3tn, implying an 80% margin, as HBM4 and high-value server products expanded. The more consequential disclosure is that long-term agreements with 14 customers now cover more than 50% of total 2027–28 production capacity, with contract pricing designed to move within agreed market ranges. Management expects HBM4 to exceed HBM3E in both shipments and revenue in Q426 and argues that memory supply will remain short through 2030. That combination strengthens the bull case that HBM is becoming a contracted infrastructure input rather than a normal spot-cycle product, supporting investment in M15X and advanced packaging without immediately surrendering pricing. The bear case is that an 80% margin and capacity commitments invite customer dual-sourcing, faster Samsung Electronics and Micron expansion, and eventual overbuild; long-dated contracts reduce volume risk but cannot eliminate technology-transition risk. Nvidia, Advanced Micro Devices, TSMC, Micron, Samsung Electronics, equipment suppliers and packaging providers all carry read-throughs. The next test is whether HBM4 qualification and ramp preserve yields as volume scales. Source: SK hynix Q326 results, 28 October.
2. Seagate’s record margins show storage scarcity has become an earnings event, not merely an exabyte narrative
Seagate reported fiscal Q127 revenue of $4.23bn, up 61% yoy, non-GAAP gross margin of 57.8% versus 40.1% a year earlier, non-GAAP EPS of $7.68 and record free cash flow of $1.5bn. Fiscal Q2 guidance of $4.8bn revenue and $9.20 non-GAAP EPS implies another material step up, while $1.2bn of debt retired during the quarter and a further $750m retired afterward sharply improve financial flexibility. The print challenges the view that Toshiba’s planned capacity expansion is enough to break current nearline pricing: Seagate’s HAMR density gains, long-term customer agreements and supply discipline are converting AI-driven exabyte demand into extraordinary incremental margins now. The bear case is precisely that those margins make future supply additions economically irresistible and that customers will resist value-based pricing as alternatives qualify. Western Digital and Toshiba receive positive near-term demand read-throughs but greater medium-term pricing pressure; cloud providers face higher storage bills, while Teradyne benefits from hard-drive test demand. The key evidence is whether Seagate can sustain gross margin as Toshiba’s additional capacity arrives and whether fiscal Q2 revenue growth is volume-led or price-led. Source: Seagate fiscal Q127 results, 27 October.
3. Oracle makes AI compute multi-architecture as Google TPUs and Cerebras challenge Nvidia at the inference layer
Oracle announced on 27 October that it will scale Nvidia Vera Rubin systems to multi-gigawatt capacity, offer Google TPUs in 2027 as the first hyperscaler outside Google to do so, and add Cerebras inference to Oracle Cloud Infrastructure. Reuters separately reported on 28 October that OpenAI plans to buy Google chips, citing The Information; neither OpenAI nor Google has confirmed that purchase. The combined development matters because Oracle is not replacing Nvidia in frontier training but is using Google and Cerebras to widen choice where inference latency, power and cost are more important. Google gains a route to monetize TPU intellectual property beyond its own cloud and potentially deeper OpenAI exposure; Oracle improves differentiation but increases the financing burden behind fiscal 2027 capex of roughly $90–95bn; Cerebras gains hyperscale distribution. The bull case for Nvidia is that heterogeneous inference expands the total market while Vera Rubin remains the anchor platform. The bear case is that Google’s external TPU strategy and purpose-built inference systems gradually compress Nvidia’s share and pricing outside training. Investors should monitor committed customer volumes, Oracle’s financing disclosures and whether TPU availability expands beyond Oracle in 2027. Sources: Oracle and Google TPU announcement, Nvidia and Oracle announcement, and Reuters on the reported OpenAI purchase.
4. Corning’s 84% enterprise-network growth says optical content is outrunning GPU units
Corning’s 27 October Q326 results delivered core sales of $5.08bn, up 19% yoy, and core EPS of $0.91, up 36%, with core gross margin expanding 150 basis points to 40.5%. Optical Communications revenue rose 44% to $2.38bn; Enterprise Networks grew 84%, and generative-AI product sales more than doubled. Management’s claim that AI sales are outpacing GPU unit growth is financially important because it supports the thesis that cluster scale and architecture increase optical content per accelerator rather than simply tracking accelerator volumes one-for-one. Q426 guidance for $5.2–5.3bn of core sales and $0.96–1.00 of core EPS indicates the strength is continuing, while customer-supported capacity additions reduce part of the expansion risk. The bear case is that long-term agreements and exceptional near-term demand can still produce digestion once campuses are equipped, especially if co-packaged optics or alternative interconnect designs change content allocation. Coherent, Lumentum, Amphenol, Broadcom, Marvell and networking vendors receive positive read-throughs; the next datapoints are hyperscaler capex, optical backlog and whether margin expands while Corning invests more than $500m per quarter. Source: Corning Q326 results, 27 October.
5. Teradyne’s 70% growth shows AI monetization spreading from accelerator test into memory, storage and interconnect
Teradyne reported Q326 revenue of $1.306bn, up 70% yoy and above the top of guidance, with non-GAAP EPS of $2.29 versus $0.85 a year earlier. Semiconductor Test contributed $1.072bn, while Memory and Integrated Systems Test reached record revenue on AI demand; management also cited hard-drive test and high-speed interconnect, and Robotics achieved its highest third-quarter revenue and returned to profitability. Q426 revenue guidance of $1.25–1.35bn suggests the elevated run rate is durable into year-end. This is a higher-quality read-through than a single accelerator program because test intensity rises with HBM complexity, advanced packaging, high-speed interfaces and storage qualification across multiple architectures. The bull case is a structurally larger test wallet per AI system and more diversified growth into 2027. The bear case is that test demand remains highly concentrated around a narrow set of capacity ramps and can reverse when customers absorb equipment. Advantest, FormFactor, Cohu, memory producers and packaging suppliers are exposed; the next catalyst is management’s 2027 demand framework on today’s earnings call and whether memory-test strength broadens beyond HBM. Source: Teradyne Q326 results, 27 October.
6. NXP’s recovery broadens beyond autos as data centers become a second growth engine
NXP Semiconductors reported Q326 revenue of $3.79bn, up 19% yoy and 8% sequentially, non-GAAP gross margin of 58.6%, non-GAAP EPS of $4.20 and free cash flow of $1.21bn, or 31.9% of revenue. Q426 revenue guidance of $3.75–3.95bn implies 12–18% yoy growth, while a new $5bn repurchase authorization signals confidence that the recovery can fund both capacity and capital returns. The narrative change is that NXP is no longer relying solely on an automotive inventory normalization: management identified data-center infrastructure and trusted physical AI alongside software-defined vehicles, while its new Singapore 300mm joint-venture fab expands supply. The bull case is a cyclical recovery becoming structural content growth across automotive compute, industrial edge systems and data centers. The bear case is that aggressive repurchases near an improving cycle and new manufacturing commitments reduce flexibility if auto demand slows again. Infineon, STMicroelectronics, Texas Instruments, Analog Devices and Microchip receive positive analog and embedded read-throughs. Segment growth, fab utilization and the mix of company-specific programs versus market recovery are the next tests. Source: NXP Q326 results, 27 October.
7. Commvault’s 20% ARR growth shows cyber resilience gaining budget as AI expands the recovery problem
Commvault reported fiscal Q227 subscription revenue of $269m, up 21% yoy, subscription ARR of $1.104bn, up 20%, and SaaS revenue of $106m, up 33%, with $71m of free cash flow. Constant-currency net new subscription ARR was $51m, and management raised the fiscal-year ARR midpoint to $1.21bn while guiding to $255–265m of free cash flow. The print supports a distinct cyber-resilience budget rather than treating recovery as a storage feature: enterprises need clean recovery across data and identity as AI increases data volume, machine-speed attacks and operational dependency. The bull case is that Active Directory recovery, threat scanning and FedRAMP-authorized services create a durable platform adjacent to backup. The bear case is that Microsoft, Rubrik, Cohesity, Dell and cloud platforms bundle overlapping recovery capabilities and pressure expansion economics. Rubrik, Veeam, Dell Technologies and identity-security vendors receive the clearest read-throughs. Net new ARR, SaaS margin and recovery-platform attach rates will determine whether 20% growth is sustainable rather than migration-driven. Source: Commvault fiscal Q227 results, 27 October.
8. Nvidia’s $3bn Meta investment ties supplier capital to the next agent platform
Nvidia and Meta said on 27 October that they will expand technical work on Meta’s Muse agents and models, deploy more Vera Rubin systems, Vera CPUs, Spectrum-X networking and BlueField-4, and explore Nvidia’s OpenShell and Sentry agent-security technologies. Nvidia will also invest $3bn in Meta’s equity offering. The hardware commitment is expected given Meta’s infrastructure plan; the new investor debate is whether Nvidia’s minority investment is a strategic ecosystem stake or another form of supplier-supported demand. The bull case is a deeper full-stack relationship spanning accelerators, networking, CPUs and runtime security, raising Nvidia content per Meta workload and improving Muse performance per watt. The bear case is customer concentration and circularity: Nvidia is investing in the equity of a buyer that will use part of its capital for Nvidia infrastructure, while Meta’s scale gives it leverage to develop more custom silicon. Broadcom, Arista Networks, Advanced Micro Devices and agent-security vendors carry read-throughs. Investors should monitor the final equity allocation, Meta’s custom-chip roadmap and whether OpenShell becomes an independent control standard or remains tied to Nvidia infrastructure. Source: Nvidia and Meta announcement, 27 October.
9. CoreSite’s 16% growth confirms interconnection scarcity is translating into data-center pricing power
American Tower reported Q326 data-center cash revenue growth of 16.3%, driven by record leasing and higher pricing at CoreSite, while consolidated AFFO rose only 2.4% to $1.335bn and free cash flow was essentially flat at $986m. Management raised full-year guidance and authorized a further $2bn of repurchases after quarter-end. The divergence matters: the tower portfolio remains burdened by churn and modest organic billings growth, but dense, interconnected data-center capacity is producing much stronger economics. That supports Equinix, Digital Realty and network-dense colocation assets, particularly where power and connectivity are already available, while weakening the argument that all data-center capacity is becoming commoditized. The bear case is that 16% growth partly reflects a favorable comparison and pricing that encourages customer migration or new supply. The next evidence is CoreSite leasing, power availability and renewal spreads, followed by Equinix results tonight. Source: American Tower Q326 results, 27 October.
10. Enphase’s data-center power option cannot yet offset a 29% contraction in the core business
Enphase reported Q326 revenue of $293.3m, down 29% yoy, and non-GAAP gross margin of 45.0%, then guided Q426 revenue to $225–255m with sell-through expected to fall 10% sequentially and non-GAAP gross margin of 40–43%. The weak outlook is the fundamental story despite management’s disclosure that it is developing solid-state transformer solutions for two global technology leaders and targeting 2027 pilots and 2028 commercial shipments. Converting 13.8kV AC to 800V DC could reduce conversion stages inside AI data centers, opening a potentially large new market for Enphase and Infineon’s GaN devices. The bull case is that power-electronics expertise creates a second growth engine with better structural demand than residential solar. The bear case is timing and distraction: commercial revenue is at least two years away while the core business is shrinking and margins are declining. Eaton, Vertiv, Schneider Electric, Infineon and Monolithic Power receive longer-term read-throughs, but customer names, megawatt-scale validation and unit economics are required before investors should capitalize the opportunity. Source: Enphase Q326 results, 27 October.
What to Watch
Vertiv reports before the U.S. open today, with orders, backlog, liquid-cooling capacity and 2027 margin commentary the most direct test of whether data-center power demand is outrunning supply without eroding execution.
Microsoft and Alphabet report after the close today. Azure revenue dollars, Google Cloud growth, AI revenue, capex, depreciation and free cash flow will test whether hyperscaler monetization is catching up with the supplier profits visible across memory, storage and optics.
ServiceNow, Fortinet, Tenable, KLA and Equinix also report tonight. The key variables are agent-driven workflow monetization, cybersecurity billings and ARR, wafer-fab-equipment demand, and the pricing and power availability of interconnected data-center capacity.
Amazon, Intel, Monolithic Power, Sandisk, Twilio and Cognizant report on 29 October, extending the read-through from cloud and semiconductors into storage, power management, communications software and IT services.
Bottom line
The physical AI stack is producing enough revenue, margin and cash flow to challenge the idea that supplier earnings are merely anticipating demand. SK hynix, Seagate, Corning and Teradyne are already converting scarcity and rising system complexity into exceptional economics, while NXP shows the benefit spreading into embedded and physical AI. The risk is that today’s margins pull forward supply and embed expectations that become difficult to beat.
Architecture is broadening at the same time. Nvidia remains the default platform for frontier training, but Oracle’s adoption of Google TPUs and Cerebras, together with the reported OpenAI interest in buying Google chips, shows inference becoming a contest among specialized architectures, cloud distribution and power efficiency. The most durable exposures are therefore not every announced AI project, but bottlenecks with contracted demand, differentiated technology and visible cash conversion. Today’s hyperscaler results must now prove that customer revenue can sustain the supplier prosperity already appearing upstream.