NeutralCRWDCrowdStrikeCrowdStrikePANWPalo Alto NetworksPalo Alto Networks

Today

Deep|Cybersecurity: Enterprise AI Security Budgets Are Shifting Toward Large Platforms

AI security spend is shifting to platforms, boosting upsell for CRWD, PANW and NET as tools consolidate. AI security spend is shifting to platforms, boosting upsell for CRWD, PANW and NET as tools consolidate. Recent enterprise interviews show that AI applications are adding demand for identity, runtime and API security, with companies integrating these capabilities into existing platforms. We expect platforms that meet these needs and replace overlapping tools to grow spending within some accounts faster than the customers’ total security budgets. CrowdStrike, Palo Alto Networks and Cloudflare are among the vendors we are watching, although the outcome for each will depend on its products and customer purchases.

NeutralAMDAMZNAmazonAmazonNVDANVIDIANVIDIA

Today

Deep|Ibiden: The Substrate Seller’s Market Has Only Just Begun

Executive Summary Ibiden’s (4062 JP) core driver remains qualified high-end ABF capacity failing to keep pace with the area and layer count of AI silicon. Intel/AMD server CPU shipments grow +36%/+62% in 2027 on new platforms 56–63% larger in substrate area, widening the CPU gap further. GPUs and EMIB-T consume the highest-end capacity: Rubin yields only 12 units per panel, TPU V9 / Trainium 4 only 4. The next pricing step falls on 27Q1 contract renewals, where new products can price above legacy agreements. We forecast FY3/27 / FY3/28 revenue growth of 35% / 30% and EPS of JPY173 / JPY267; at ~45x FY3/28E PE the shares still have rerating room, with the FY27Q2 results call the near-term catalyst. CPU demand is back and widens the supply gap. Resident agents such as Muse push workloads back onto CPUs via headless browsers and high-frequency API calls, and Intel can currently fill only ~50% of orders; shipments have upside if capacity allows. FUNDA models Intel server CPU shipments of 26.5mn units in 2027E (+36%) and AMD 20.4mn (+62%); multiplied by substrate area 63% larger for Diamond Rapids vs. Granite Rapids and 56% larger for Venice vs. Turin, substrate-area demand grows far faster than unit count. Supply is not only limited but already allocated: tier-1 and tier-2 makers are all full as of 26Q4, lead times are 48–52 weeks, total 2027 capacity grows only ~10–15%, and a visible shortfall opens from 27Q1. Customers that did not co-fund capacity sit lower in the allocation queue. This is already visible at Ibiden: AMD substrate revenue declines sequentially through 2H FY3/27 as capacity temporarily allocated to AMD returns to Intel; this does not reflect weaker AMD demand. GPUs and EMIB-T consume the highest-end slice of capacity. Blackwell is 73×81mm at 14 layers; Rubin 83×97mm at 18 layers, an estimated 1.8x the area; Rubin Ultra more than 2x Blackwell’s area at ~3x the price. EMIB-T is more extreme: Google TPU V9 at ~120×120mm, 24 layers with a multi-layer core, and Amazon Trainium 4 at ~120×120mm, 24 layers with a multi-layer core, each yielding only 4 substrates per panel. Even if ASIC unit growth looks modest, what truly consumes capacity is area, layer count and yield (core-stage yield on a multi-layer core drops from 99.9% to 97–98%, and a core defect can render all subsequent build-up layers worthless). Substrate price increases continue, supporting Ibiden’s new-product pricing. This year’s increases came mainly from Taiwanese makers; Ibiden only removed its quarterly price-decline assumption and normalized Intel pricing to a fair level. Next year, with the industry shortfall in place and lead times at 48–52 weeks, the next customer contract renewals fall in 27Q1 and new products can secure better quotes; new T-glass suppliers are also expected to qualify in 27Q1, and with Chinese and Taiwanese suppliers raising prices Ibiden also has room to pass through costs, while substrates remain under 10% of chip cost so we expect customers to accept the increase. We have not yet seen specific quotes, but we think a price increase is likely. Business Overview Ibiden (4062 JP) is a Japanese electronic-materials company built around high-end IC package substrates (FC-BGA / ABF). Founded in 1912 and headquartered in Ogaki, Gifu Prefecture, its core product is high-end ABF substrates produced on a semi-additive process (SAP), used in AI GPUs, AI ASICs, server CPUs and networking switch ICs. Historically anchored on Intel CPU substrates, the company is now shifting rapidly toward AI accelerators, where its competitive strengths are yield on large, high-layer-count substrates, first-qualification capability on new designs, and close relationships with Japanese materials and equipment suppliers. The lowest unit price is not the main consideration. Products & Services Electronics (58% of FY3/26 revenue, ~68% in FY3/27E): the growth engine. IC package substrates dominate the segment. FY3/26 revenue JPY243.3b, operating profit JPY45.2b (18.6%); FY3/27 guidance JPY375.0b / JPY110.0b (29.3%). The customer base is led by NVIDIA, Intel and AMD. Ceramics (20%). Diesel particulate filters (DPF), SiC ceramic fiber and specialty carbon products for semiconductor processing. FY3/26 revenue JPY82.6b, operating profit JPY7.6b. FY27Q1 revenue rose 24% YoY on temporary DPF replacement demand, but this is not a medium-term growth driver. Others (22%). Construction materials, synthetic resins and petroleum products; operating margin ~10–12%, a stable but low-growth contributor. Near-Term Focus: FY27Q2 Results Subscribe now Read more

NeutralGOOGLAlphabetAlphabetNVDANVIDIANVIDIA

4d ago

Research|LLM: Gemini 4 Brings Google Back into Frontier Competition as Investment Accelerates

Google announced Gemini 4 Argon on September 30, roughly ten months after Gemini 3. On Google’s evaluations, Argon leads or ties GPT-6 Astra and Claude Opus 5.5 on 14 of 19 benchmarks. Artificial Analysis scores it at 53, level with Astra and below Opus 5.5 at 58 and Sonnet 5.5 at 56. Google is back in frontier competition, though it still trails OpenAI and Anthropic overall. Introductory pricing is $2 per million input tokens and $10 per million output tokens, rising to $4/$20 at the standard rate. We use this input/output convention throughout. Access is initially limited to trusted testers, with paid API customers and Google AI Ultra subscribers next in line. Two months ago, many investors thought Google was stepping back from frontier models. Gemini 3.5 Pro had missed its launch window and been shelved, leaving Google reliant on Flash-tier models. On August 5, Demis Hassabis stepped down as DeepMind CEO to become its chairman and Alphabet’s chief scientist. Jeff Dean left to found Discovery Loop with Oriol Vinyals, Quoc Le and others. Koray Kavukcuoglu took over DeepMind’s day-to-day operations, reporting to Sundar Pichai. Alphabet fell about 5% that day. Much of the commentary saw a shift toward selling compute and distributing AI applications. Tim O’Reilly drew a parallel with Westinghouse, suggesting Google might focus on infrastructure and wider AI adoption. Our view was different: Google would keep investing in frontier models, and a tighter focus would give it room to catch up. Argon’s results support that view. The training run began before the reorganization, however, so the effect of the management changes on R&D will only become clear over the next few model generations. Early this year, some in the market considered a scenario in which Anthropic kept extending its lead in coding and the enterprise, the other labs gradually fell away, and the frontier ended up with a single player. Today, OpenAI and Anthropic still lead, Google is back in contention, and Meta and xAI continue to invest. Several labs remain in the race. Google returns to frontier competition Argon’s Artificial Analysis score is 23 points above Gemini 3.1 Pro and 12 above Gemini 3.8 Flash, previously Google’s highest-scoring model. It ranks first among 41 models on the Vals Index at 68.9%, versus roughly 67% for Sonnet 5.5 and Opus 5.5, and tops the LMArena text leaderboard at 1525. Google disclosed on July 22 that its “most ambitious pretraining run yet” was in progress and announced Argon roughly ten weeks later. Our checks indicate the run used TPUs. We had pushed back on rumors of a move to Nvidia’s GB-series chips: Google has spent years optimizing its training systems around TPUs, and switching to GPUs would require extensive adaptation and revalidation. A training restart alone does not establish that there is a problem with the chips. Pretraining scaling continues to deliver Google has not disclosed the model’s size beyond calling this its “most ambitious pretraining run yet” and the model “significantly larger.” Our industry checks suggest a larger MoE base with a looped architecture: roughly 5–6 trillion total parameters, 400–500 billion active per token, and three loops. We estimate its effective scale at five to six times that of Gemini 3 Pro. This increases both model capacity and compute per token. Many in the industry argued over the past year that pretraining scaling had run its course and that the focus should shift to post-training and inference. Gemini 4 expands both pretraining and inference compute. The output limit rises from 64,000 tokens to 1 million, and Long Decode Continuation lets a response pause and resume across calls. The model can spend more tokens working through long tasks, but completing them reliably still depends on tools and state management. Long-context reasoning has improved as well. On GraphWalks with inputs of 256K–1M tokens, Argon scores 84.2%, versus 71.8% for Astra and 66.8% for Opus 5.5. This measures multi-hop reasoning over long inputs, a separate capability from generating longer outputs. Argon uses markedly more tokens per task. Across the Artificial Analysis index, Argon generated an average of 62,000 output tokens per task, more than twice Astra’s 27,000. Long agent runs can also be expensive: a CUA-bench run on Vals costs $193.78 at standard pricing with a 262K output cap. Pricing and the TPU cost advantage Subscribe now Read more

DRAMRoundhill Memory ETFBullishRoundhill Memory ETF

4d ago

Review|MU FY26Q4: Visibility Extends to 2028, Supply Tighter Than in 2026

Supply stays tighter through CY28; DRAM prices keep rising, pace capped by LTAs and affordability.

GEVGE Vernova Inc.NeutralGE Vernova Inc.

5d ago

Deep|FTAI: Aviation Aftermarket Meets AI Power

Executive Summary We initiate coverage of FTAI, an independent aviation aftermarket platform focused on the CFM56-5B/-7B and V2500 engine families. It is expanding into power generation by converting CFM56 engines into aeroderivative turbines. FTAI is the second BTM solution provider we have covered, after Bloom Energy. We remain bullish on off-grid power providers. As estimated in our AIDC Deep Dive, North America could face a ~15GW power shortfall in 2027. Securing grid power could take ~5 years, while traditional gas turbine OEMs such as GEV are booked out to 2032 with cash commitments. We are therefore looking for alternative BTM providers with reliable technology and confirmed orders. FTAI trades at ~10.5x forward EV/EBITDA and has yet to receive the re-rating seen at other BTM providers, despite the $1.465b Power order announced in late July. We see two reasons. Its traditional Leasing and MRE businesses face pressure on margins and industry demand. Its Power business has yet to deliver Mod-1 units to customers, making future earnings harder to assess. Bears still view Mod-1 as a prototype rather than a mature commercial product. In short, we believe that margin compression in its MRE business does not prevent it from maintaining earnings growth, supported by expanding market share from ~15% to ~26% based on our proprietary supply-demand industry model; and Power could deliver ~10 Mod-1 products in 26Q4 vs LSD consensus; ~80 in 2027 vs ~70 consensus and ~100 company maximum target. Management were guiding $450-750mn Adj. EBITDA in Power, suggesting a range of ~60 to 100 Mod-1 deliveries as guidance in 2027. We see the first batch of Mod-1 delivery in Q4 as the next rerating catalyst, and do not think the execution bar for Power is that high in the near term. Successful delivery and acceptance from customers should be enough for the street to catch up as it clears ambiguity at current valuation. Our reverse SOTP suggests that the market is paying nothing for Power, and we estimate it should be worth $82/share in our base case of 80 Mod-1 delivery in 2027. Why Now FTAI has yet to receive the valuation re-rating seen at other behind-the-meter (BTM) power providers in the current power shortage cycle. It trades ~30% below its two-year historical average and ~37% below peers, despite the launch of FTAI Power and management’s expectation of ~$450–750m in incremental EBITDA in 2027. We see this as a timely opportunity to examine the company’s fundamentals and quantify its position in the market. Key Debates & Our Views 1）Does AP margin compression in 26Q2 signal a de-rating, or is the market focused on the wrong issue? Debate Aerospace Products (AP) EBITDA margin declined from ~35% in 25Q4 to ~29% in 26Q2. Bears worry that FTAI is prioritizing market share among large Tier-1 customers over profits and cash flow, to the detriment of shareholder value. As the CFM56 fleet ages and its installed base eventually declines, they also question the durability of MRE demand and whether Power will compete with Aerospace MRE for the same feedstock. Our View We believe AP can maintain earnings growth through market-share gains even as margins weaken and the CFM56 aftermarket declines. Power could consume some of AP’s feedstock over the longer term, but its higher margins should still add to earnings: ~35–40% for Power versus ~30% for the company overall. We do not expect feedstock competition to become a constraint over the next three to five years. The greater risks are capacity expansion and coordination across Montreal, Miami, Rome and New York. Our Reasoning Margin compression mainly reflects a shift in mix towar d Tier-1 customers requiring heavier repairs. We expect margins to hold at ~30% over the next two years. Channel checks put gross margins on heavy restorations at ~23–24%, versus ~30–31% on standard engine sales, but with substantially more revenue per project. Heavier work requires more parts and forces FTAI to procure inventory ahead of demand. We also see a timing mismatch between costs and revenue recognition as a contributor to margin pressure. CFM56 retirements reduce future MRE demand but also increase feedstock for AP and Power Mod-1. Our proprietary supply-demand model examines both effects. We assume the CFM56 installed base declines at a five-year CAGR of ~4% through 2031E, while FTAI module output grows at ~9%, implying market share rises from ~15% to ~25%. More retirements and SCI engine returns expand available feedstock, although FTAI’s required share of total supply rises from ~23% today to ~45% as AP and Power grow. Gross feedstock coverage tightens from 4.4x in 2026E to ~2.2x from 2028E onward, without becoming a constraint in our base case. AP also has growth opportunities beyond CFM56. V2500 should support earnings as its restoration program expands, while LEAP offers the next source of growth in 2028–2029. We believe FTAI can apply much of its CFM56 sourcing and MRE expertise to other engine platforms. On Power, channel checks also suggest early Mod-1 production relies largely on existing inventory and lower-green-time material with limited aviation value, making one-for-one cannibalization of AP unlikely. The more meaningful overlap appears to be in shop capacity, selected modules, test-cell time and skilled labor, rather than in engines. Exhibit 1: FTAI Aerospace Products Supply-Demand Model Subscribe now Read more