Tuesday, 3 November 2026. Information cutoff: 06:54 GMT.
Morning View
The overnight evidence sharpened the market’s new dividing line: AI demand remains strong, but access to capital and control of unit economics now matter as much as revenue growth. Oracle priced a $20bn equity-and-debt raise at borrowing costs as high as 8.65%, while Apple’s record quarter was overshadowed by a holiday gross-margin reset driven by memory and AI-server costs. The read-through is not that the AI cycle is ending; it is that investors are demanding proof that incremental compute, memory and model spending can earn returns above an increasingly visible cost of capital.
Two very different growth stories reinforced that point. Palantir Technologies nearly doubled revenue and produced a 53% adjusted free-cash-flow margin, yet its shares fell because US commercial contract-value growth slowed and early sovereign-AI deals were shorter than usual. Marvell Technology’s $115bn all-stock agreement for Lumentum offers a credible route to own more of the optical signal chain, but the negative reaction in both stocks shows that investors are unwilling to capitalize distant synergies at face value. Scale is still rewarded; scale purchased at peak-cycle valuations is not.
The fundamental balance is mixed. Lattice Semiconductor and onsemi showed that AI demand is widening from accelerators into rack control, programmable logic and power delivery, while Tempus AI demonstrated improving operating leverage in applied AI. Check Point Software exposed a slower security recovery despite healthy subscription growth, and the UK’s cyber agency warned that AI will accelerate exploitation of neglected infrastructure. The strongest positioning remains with companies converting AI demand into current cash flow or irreplaceable infrastructure; the weakest remains businesses financing long-duration commitments with uncertain utilization, pricing or residual value.
1. Oracle’s $20bn capital raise makes the cost of AI infrastructure impossible to ignore
Oracle priced concurrent offerings of roughly $10bn of common stock and $10bn of senior notes, generating expected net proceeds of about $19.9bn for capital expenditure, debt repayment, investment or acquisitions. The debt coupons range from 7.00% on 2029 notes to 8.65% on 2046 notes, while 86.2m shares were priced at $116. Oracle fell 8.6% during Monday’s session as the financing plan crystallized dilution and borrowing-cost concerns. The strategic logic is clear: Oracle is trying to fund a multi-gigawatt AI cloud buildout while demand and supplier capacity are available. The financial tension is equally clear: infrastructure returns must now exceed high-single-digit long-term debt costs before equity dilution, depreciation and execution risk. Bulls will argue that contracted customers and diversified accelerator offerings can support high utilization; bears will question whether long-duration capacity commitments are being funded ahead of proven free-cash-flow conversion. Nvidia, Broadcom, Google, Cerebras Systems, power suppliers, data-center developers and private-credit providers receive positive demand read-throughs, but Oracle shareholders absorb the financing risk. Closing of the offerings on 4 November, Oracle’s next capex update, customer prepayments and any revision to fiscal 2027 free-cash-flow expectations are the next catalysts. Sources: Oracle offering announcement, 2 November; Oracle SEC filings; Reuters market report, 2 November.
2. Apple beats the quarter but resets holiday margins as memory scarcity reaches the consumer stack
Apple reported fiscal fourth-quarter revenue of $114.5bn, up 12% yoy and above the roughly $113.4bn consensus, with diluted EPS of $2.10 versus approximately $1.99 expected. iPhone revenue set a September-quarter record and Services reached an all-time high. The quality of the beat was less clean than the headline: gross margin of 47.9% included about 100 basis points of tariff-refund benefit, and management guided December-quarter gross margin to 45%–46% as memory costs rise. Services growth guidance of 8%–9% also fell below the roughly 10.7% market expectation. Shares fell as much as 4% during the call and were down about 2% later in extended trading. The narrative changes from Apple being insulated from the AI capex cycle to Apple paying for it through component inflation and server expense, even with more restrained infrastructure spending than hyperscalers. Bulls can point to pricing power, strong iPhone demand and secured supply; bears will focus on purchasing commitments that rose by $33bn, slower Services growth and supply constraints around the iPhone Duo. Micron Technology, Samsung Electronics and SK Hynix benefit from the memory signal, while consumer hardware peers face similar margin risk. Holiday sell-through, memory pricing and the timing of supply normalization are the next decisive datapoints. Sources: Apple results, 2 November; Reuters, 2 November.
3. Marvell’s $115bn Lumentum deal buys optical scale, but investors reject peak-cycle execution risk
Marvell Technology agreed to acquire Lumentum in an all-stock transaction with an implied enterprise value of $115bn. Lumentum holders will receive 4.5 Marvell shares for each share owned and hold about 33% of the combined company; closing is targeted for the first half of 2028. Management expects more than $1bn of annual run-rate cost synergies, year-one non-GAAP EPS accretion and a combined addressable market of $550bn. The industrial logic is strong: AI clusters are moving from copper toward optical connectivity, and Lumentum brings indium-phosphide manufacturing, lasers and compound-semiconductor expertise that complement Marvell’s DSPs, networking and custom silicon. The price and timeline create a different debate. Marvell fell more than 10% and Lumentum more than 15%, indicating that investors are discounting regulatory risk, integration complexity and the possibility that current optical scarcity is being capitalized at the top of the cycle. Broadcom, Coherent, Fabrinet, Applied Optoelectronics and Nvidia receive major competitive read-throughs, while hyperscalers may gain a broader integrated supplier but less component competition. The key questions are whether the fixed exchange ratio remains economically attractive through closing, whether the $1bn synergy target requires aggressive restructuring, and whether Chinese-transceiver restrictions alter market share before 2028. Sources: Marvell and Lumentum announcement, 2 November; Reuters, 2 November.
4. Palantir nearly doubles revenue, but shorter contracts expose the duration risk inside sovereign AI
Palantir Technologies reported third-quarter revenue of $2.313bn, up 96% yoy and above the roughly $2.18bn consensus. US commercial revenue rose 148% to $984m and exceeded US government revenue for the first time, while adjusted operating income reached $1.427bn and adjusted free cash flow was $1.216bn, a 53% margin. Fourth-quarter revenue guidance of $2.649bn–$2.653bn was well above the approximately $2.46bn market estimate, and full-year guidance increased to $8.530bn–$8.534bn. Yet the shares fell nearly 3% after hours because US commercial total contract-value growth slowed to 11% yoy from more than 153% in the prior quarter. Management said several eight-figure sovereign-AI agreements were shorter than usual as customers adopted initial capabilities rapidly. Bulls will see exceptional revenue, margins and commercial adoption that justify continued estimate upgrades; bears will see duration, renewal and valuation risk masked by unusually fast near-term conversion. The results are positive for ontology, data-governance and implementation vendors, but they raise the bar for Snowflake, Databricks, C3 AI and defense software peers. Renewal rates, contract length, remaining deal value and conversion of pilots into multi-year deployments are the next catalysts. Sources: Palantir results filing, 2 November; Palantir investor presentation; Reuters, 2 November.
5. Anthropic’s amended filing reveals slower growth and $895.4bn of infrastructure commitments
Anthropic’s 2 November amended registration statement showed third-quarter revenue of $16.96bn, up 47% sequentially after 144% growth in the second quarter, while net loss narrowed to $23.29bn from $66.3bn. The company recorded an operating loss of $567m after prior-quarter operating income and incurred an additional $5bn of compute costs, weakening revenue generated per dollar of compute. Most consequentially, the filing describes $895.4bn of infrastructure spending commitments, with about $69.8bn due in the twelve months after 30 September and nearly half of the total due before 2031. The new disclosure changes the IPO debate from whether frontier-model revenue can scale to whether capital commitments can be financed without surrendering margin or strategic flexibility. Bulls can argue that enterprise adoption, more than 11,900 large customers and falling model costs provide a path to absorb capacity. Bears will focus on sequential deceleration, renewed operating losses and the company’s warning that third-party financing could delay or raise the cost of contracted infrastructure. Amazon, Google, Broadcom, Nvidia, cloud lessors and private-credit providers are the principal read-throughs. IPO pricing, cornerstone demand, compute-cost intensity and revisions to capacity commitments are the next catalysts. Sources: Anthropic S-1 amendment, 2 November; Reuters, 2 November.
6. Check Point’s subscription engine holds, but the recovery is slower and less visible than the market required
Check Point Software reported third-quarter revenue of $669.8m, down 1% yoy and slightly below the roughly $672m consensus, while non-GAAP EPS of $2.49 was broadly in line. Security-subscription revenue grew 11% to $339m and remaining performance obligations increased 5% to $2.5bn, but the company guided fourth-quarter revenue to $720m–$770m, a midpoint implying flat yoy growth and below the approximately $787m market expectation. Adjusted free cash flow remained excellent at $258m, or 38% of revenue. Management believes the third quarter marked the revenue trough, but explicitly said the recovery is taking longer than anticipated; the shares fell about 4.6% during Monday trading after a steeper premarket decline. The bull case is that email security, continuous exposure management and AI security can reaccelerate growth while high margins and cash generation protect downside. The bear case is that healthy subscriptions cannot offset weakness in products and maintenance quickly enough, leaving Check Point structurally behind faster platform consolidators. Palo Alto Networks, Fortinet, CrowdStrike and Zscaler gain competitive read-through, while resellers face slower appliance refresh. Fourth-quarter billings, product growth, RPO acceleration and evidence that new platforms displace rivals rather than bundle existing spend will determine whether the trough call is credible. Sources: Check Point results, 2 November; Market reaction and guidance context, 2 November.
7. Lattice and onsemi show AI demand widening into control logic and power, but end-market quality diverges
Lattice Semiconductor reported revenue of $291.0m, up 118% yoy, non-GAAP gross margin of 70.9% and adjusted EBITDA of $131.0m; its FPGA business grew about 90%, led by compute and communications demand. The comparison includes AMI from 27 July, which contributed approximately $39m, so the headline growth rate overstates the organic acceleration. Fourth-quarter revenue guidance of roughly $335m implies about 130% yoy growth, and the shares rose approximately 7% after hours. onsemi separately reported revenue of $1.722bn, up 11%, non-GAAP gross margin of 41.2% and free cash flow of $489m. AI data-center revenue grew 175% yoy and 39% sequentially, but the shares fell roughly 3% after hours as management forecast a high-single-digit sequential decline in automotive revenue while reallocating supply to AI and industrial customers. Together, the prints confirm that accelerator demand is pulling content into rack management, programmable control and power delivery. Bulls see a broad, high-margin content cycle; bears see acquisition effects at Lattice and cyclical substitution at onsemi rather than uniform end-market health. Microchip Technology, Texas Instruments, Monolithic Power Systems, NXP Semiconductors, Analog Devices and server original-design manufacturers are exposed. Organic backlog, automotive stabilization and AI revenue mix are the next key datapoints. Sources: Lattice Semiconductor results, 2 November; onsemi results, 2 November.
8. Tempus AI’s operating leverage improves, but diagnostics remains the proof point for data monetization
Tempus AI reported third-quarter revenue of $411.5m, up 23% yoy, with oncology volume growth accelerating to 32% and Data and Applications revenue growing 33%. Adjusted EBITDA rose to $25.6m from $1.5m, non-GAAP operating income reached $14.4m from an $8.7m loss, and the net loss narrowed to $14.9m from $80.0m. Management raised full-year revenue guidance to $1.600bn–$1.605bn, approximately 25% growth, while maintaining adjusted EBITDA guidance near $65m. The quarter matters because Tempus is testing whether an AI data asset can be funded by a scaled diagnostics channel rather than perpetual external capital. Bulls will argue that faster oncology volumes create more proprietary data and improve fixed-cost absorption, strengthening the flywheel between testing, data licensing and model development. Bears will question the durability of reimbursement, the quality of non-GAAP profitability and whether Data and Applications growth can outrun the more capital-intensive diagnostics base. Exact Sciences, Guardant Health, Illumina, health-data vendors and pharmaceutical research platforms receive the read-through. Cash conversion, reimbursement trends, Data and Applications bookings and evidence that diagnostic cohorts generate repeat pharmaceutical revenue are the next proof points. Source: Tempus AI results, 2 November.
9. OpenAI offers flexible load as a grid asset, shifting the data-center debate toward enforceable operating behavior
OpenAI introduced a Grid Protectors Program under which US data-center campuses it directly develops would reduce electricity use during extreme weather and other grid emergencies, beginning from commitments around its planned Georgia campus and expanding nationally. The announcement is strategically important because community and regulatory opposition increasingly centers on ratepayer costs and reliability, not simply total generation. Flexible load can improve project approval odds, reduce the need for peak infrastructure and support higher utilization of generation during normal periods. The bull case is that large AI campuses can become dispatchable demand resources, lowering political friction and accelerating interconnection. The bear case is that voluntary curtailment language lacks value without measurable capacity, compensation rules and enforceable performance; model training may be flexible, but latency-sensitive inference and contracted customer workloads are less interruptible. OpenAI, Microsoft, Oracle, CoreWeave, Crusoe and data-center developers are exposed, while utilities, battery providers and demand-response software vendors could gain a new revenue pool. Investors should watch for signed utility tariffs, disclosed megawatts available for curtailment, performance penalties and whether hyperscalers adopt comparable standards. Sources: OpenAI, 2 November; Axios, 2 November.
10. The UK cyber agency moves AI-enabled infrastructure attacks into the base case for 2028
The UK National Cyber Security Centre assessed that AI-powered malware is highly likely to be used in a ransomware attack on British critical national infrastructure by 2028 and that Russia, China and other states are highly likely to use AI-cyber routinely for geopolitical objectives. The agency expects open-source tools to lower attacker costs, accelerate vulnerability discovery and exploitation, and expand attacks against legacy infrastructure; it also warns that defensive implementation is lagging. This is more than a generic threat forecast because it accompanies £60m for the first phase of a national Cyber Shield, £70m for a National Exposure and Real-Time Vulnerability Centre, and a new government AI Cyber Security Centre of Excellence. The bull case for cybersecurity is a multi-year shift from episodic incident response toward continuous exposure management, identity control, segmentation and automated remediation. The bear case is that public budgets remain too small relative to legacy-system debt and that spending disperses across services rather than high-margin platforms. CrowdStrike, Palo Alto Networks, Microsoft, Cloudflare, Tenable, Rapid7, Qualys and industrial-security vendors are the main public read-throughs. Procurement awards, implementation timelines and measurable reductions in patch latency will show whether the warning becomes revenue. Sources: NCSC announcement, 3 November; NCSC AI and cyber intrusion report.
What to Watch
Advanced Micro Devices, Arista Networks, Astera Labs and Varonis Systems report after Tuesday’s close, testing accelerator share, networking demand, PCIe connectivity and security-spending durability. Datadog and GlobalFoundries report Wednesday morning, followed by Arm, Coherent, Qualcomm, Rapid7 and Figma after the close. Oracle’s $20bn offerings are expected to settle on 4 November. Anthropic begins marketing its IPO next week, making price range, cornerstone participation and infrastructure-risk disclosure the highest-value private-market catalysts. The US midterm result will also shape the policy path for data-center permitting, energy cost allocation, semiconductor support and AI oversight.
Bottom line
Demand is not the problem in this morning’s evidence. The problem is translating demand into durable returns after financing costs, memory inflation, acquisition premiums and shortened contract duration. Apple, Palantir, Lattice, onsemi and Tempus all showed real end-market strength, but their market reactions separated clean execution from narratives that require favorable assumptions.
The most attractive exposures remain businesses with current cash conversion and scarce technical control points—memory, optical connectivity, rack management, power delivery, data governance and continuous security. The least attractive are companies locking in long-duration infrastructure or transaction risk before utilization, margins and financing are proven.