1. Alphabet Pursues $80B Share Offering to Fund AI Infrastructure
• Core Source
“Alphabet (NAS: GOOG, GOOGL) announced on Monday the 2nd that it plans to issue shares worth a total of $80 billion, including a $10 billion investment from Berkshire Hathaway (NYS: BRK.B)”
“Strong demand for AI solutions and services is being seen from both enterprises and consumers, and current demand levels exceed the company’s supply capacity”
“Berkshire Hathaway purchased $5 billion in Class A common shares (at $351.81 per share) and $5 billion in Class C shares (at $348.20 per share). Berkshire Hathaway is reported to have been accumulating these stakes since Q3 2025”
• Expected Impact
Alphabet plans to raise a total of $80 billion through this share offering — $30 billion via a public offering, $40 billion through an at-the-market sales program beginning in Q3 2026, and $100 billion in private investment from Berkshire Hathaway — with all proceeds earmarked for expanding AI computing infrastructure. Berkshire purchased $5 billion in Class A common shares at $351.81 per share and $5 billion in Class C shares at $348.20 per share, having reportedly accumulated the position since Q3 2025.
Alphabet had already raised its annual capital expenditure forecast in April from $175–185 billion to $180–190 billion, and combined capital expenditure from Alphabet, Microsoft, Meta, and Amazon is expected to exceed $700 billion in 2026. Wall Street estimates that total AI-related capital expenditure will surpass $1 trillion by 2027. This fundraising underscores that AI infrastructure investment competition has intensified to a level that already exceeds companies’ own operating cash flows. The direct private participation of Warren Buffett’s Berkshire Hathaway is seen as a signal of institutional-level confidence in Alphabet’s AI infrastructure investment direction.
2. JPM on Google TPU v9 Supply Chain Reshuffling — MediaTek’s Rise
• Core Source
“MediaTek’s ‘Humufish’ (v9 TPU, based on TSMC N2 process with 4 compute dies + 12 HBM4e stacks) is emerging as the lead project for Google’s TPU v9 lineup in 2028–2029” (Gokul Hariharan)
“This effectively replaces Broadcom’s ‘Pumafish,’ which may be cancelled or delayed” (Gokul Hariharan)
“The more challenging but higher-margin and higher-volume Inference TPU remains with Broadcom, while the Training TPU is being handled by Google’s CoT (Chain of Thought) team and MediaTek” (Harlan Sur)
• Expected Impact
This issue is notable in that two analysts within JPM have presented analyses from different angles.
Gokul Hariharan, JPM’s Asia supply chain analyst, based his analysis on supply chain research findings, concluding that MediaTek’s Humufish is emerging as the lead project for Google’s TPU v9 in 2028–2029, while Broadcom’s Pumafish may be cancelled or delayed. Accordingly, he upgraded his outlook for MediaTek’s data center ASIC business and raised his price target, identifying EMIB-T packaging technology yield achievement as the key execution risk.
Harlan Sur, who directly covers Broadcom as JPM’s U.S. semiconductor analyst, maintained his view that Google is operating separate TPU roadmaps for training and inference. He emphasized that the “higher-margin and higher-volume” Inference TPU remains with Broadcom, adding that the two roadmaps are not dependent on each other. From this perspective, MediaTek’s rise does not directly translate into a threat to Broadcom’s entire Google business.
Taken together, MediaTek appears to be gaining a more prominent presence over Broadcom in certain projects within the Google TPU v9 lineup, and this is read as a signal that Asian fabless powerhouse MediaTek has made a full-scale entry into the custom AI chip market previously dominated by Broadcom.
3. Siemens and Fluence Collaborate with Nvidia to Standardize Power Infrastructure for Vera Rubin AI Factory
• Core Source
“Siemens, in collaboration with NVIDIA and Fluence, develops a dedicated Reference Architecture for NVIDIA’s next-generation AI platform ‘Vera Rubin NVL72′”
“System scale: total facility capacity of 136MW, supporting IT load of 100MW. Design objective: providing an end-to-end power and control solution from utility connection (34.5kV) to rack interface”
“Fluence Energy is the sole energy storage supplier for this data center design, supplying batteries to secure 2–3 hours of backup power instead of the typically required 1 hour”
“As of the end of FY2Q26, Fluence Energy’s data center pipeline backlog stands at approximately 12GW”
• Expected Impact
Siemens, together with Nvidia, Fluence Energy, and nVent, has jointly developed a 136MW-class AI Factory dedicated reference architecture for Nvidia’s Vera Rubin NVL72 platform. Roles are clearly delineated: Siemens handles medium- and low-voltage power distribution and control from the grid (34.5kV) to the rack interface; Fluence supplies battery energy storage systems (BESS) based on its ‘Smartstack’ platform; and nVent manages thermal solutions.
The key significance of this collaboration is that the power infrastructure standard for building Nvidia AI Factories has been officially formalized. Fluence has been officially recognized as the provider of the standard solution capable of resolving the ‘Quality of Power’ issue, which had been cited as the key variable in MSA (Master Service Agreement) contracts. While this announcement does not constitute a confirmed purchase order, it is interpreted as effectively guaranteeing Fluence’s inclusion in Nvidia data centers built by Siemens. With Fluence’s data center pipeline backlog already at 12GW, this standardization collaboration is expected to accelerate the conversion of that pipeline into actual orders.
4. Wells Fargo Projects Microsoft AI Revenue to Grow from $34.6B in FY26 to $125.9B in FY28
• Core Source
“Wells Fargo estimates Microsoft’s AI ARR at approximately $37B as of March 2026, with OpenAI- and Anthropic-related Azure usage and OpenAI revenue sharing accounting for the majority. OpenAI and Anthropic are estimated to account for over 70% of AI ARR, approximately 50% of Commercial RPO, and approximately 25% of Azure revenue”
“Microsoft has disclosed that M365 Copilot paid seats have exceeded 20 million, with guidance for more than 25 million seats by the end of FY26”
“Wells Fargo estimates that Microsoft’s total capacity will double from 11GW at end-2025 to 22GW by end-2027, reaching 27GW in 2028, 34GW in 2029, and 40GW by 2030”
“FY27 FCF estimates have been significantly reduced from $13.4B to $6.1B”
• Expected Impact
Wells Fargo maintained its Overweight rating on Microsoft while raising its price target from $625 to $650. The core thesis is that Microsoft’s AI revenue structure is transitioning from dependence on OpenAI and Anthropic’s Azure usage toward its own software and models.
M365 Copilot is the key driver. Wells Fargo estimates that of approximately 465 million M365 Commercial paid users, approximately 340 million are enterprise seats with high Copilot adoption potential, and projects paid seats to expand to 105 million by end of FY28. Accordingly, M365 Copilot revenue alone is expected to grow from approximately $4B in FY26 → approximately $10B in FY27 → approximately $20B in FY28. On the hardware side, the proprietary AI chip Maia 200 is already being used for some first-party inference workloads, with token costs improved by 30% versus the previous generation. However, due to capex burden from AI infrastructure expansion, FY27 FCF is projected to drop sharply from $13.4B to $6.1B, making near-term financial metric pressure unavoidable.
5. Kioxia Investor Day — CSPs Pursuing Long-Term Contracts Beyond 2029
• Core Source
“Management raised FY3/27 Capex to ¥470 billion, with a similar level guided for FY3/28”
“Citing AI inference driving KV cache demand, management raised its industry NAND CAGR forecast to +22%, in line with market expectations”
“Management emphasized that BiCS8 technology adopts CBA technology, as many investors had expected. CBA is a bonding process that separates memory and CMOS before bonding them together, enabling ultra-high capacity by stacking 32 dies in a single package”
“BiCS10 is expected to be based on a 332-layer structure, and management believes this demonstrates superior power, efficiency, and reliability compared to competitors’ 400-layer products”
“Hyperscale cloud service providers are seeking NAND flash memory contracts for 2029 and beyond”
• Expected Impact
Kioxia delivered a series of announcements at this Investor Day confirming its full-scale transformation into an AI infrastructure beneficiary. It raised FY3/27 CapEx to ¥470 billion with a similar level guided for FY3/28, and lifted its industry NAND CAGR forecast to +22%, targeting a data center revenue mix of 60% by CY28.
From a market size perspective, the most notable point is that hyperscale CSPs are already seeking long-term NAND contracts for 2029 and beyond. A transition from one-year contracts to multi-year LTAs extending through FY28–29 is underway, with management indicating progress in these negotiations. On the technology front, CBA (CMOS Bonded to Array) technology places Kioxia 4 years ahead of competitors, and BiCS10’s 332-layer structure is claimed to outperform competitors’ 400-layer products in power, efficiency, and reliability. Financially, Kioxia indicated the likelihood of paying its first dividend from FY27, with up to 50% of cash flow (or 100% absent M&A) available for shareholder returns.
6. HPE FY2Q26 Earnings — Revenue +40% YoY, Full-Year EPS Guidance Raised Sharply
• Core Source
“Q2 revenue was $10.7 billion, up 40% year-on-year, with non-GAAP EPS of $0.79, up 108% year-on-year”
“New orders more than doubled year-on-year, with backlog reaching the highest level in company history”
“AI order backlog reached a record high of $6.3 billion, of which 61% came from government and large enterprise customers”
“Management stated that shortages in DRAM, NAND, and wafers are the biggest constraint on converting backlog into actual revenue”
“The previously set FY28 target (minimum $3.00) is now expected to be exceeded two years ahead of schedule”
• Expected Impact
HPE recorded revenue of $10.68 billion (+40% YoY) and adjusted EPS of $0.79 (+108% YoY) this quarter, beating consensus by $930 million and $0.26 respectively. By segment, networking revenue — reflecting the Juniper acquisition effect — came in at $2.7 billion (+148% YoY), and Cloud & AI revenue at $7.7 billion (+23% YoY).
From a market size perspective, the key takeaway is that HPE management emphasized the strong results are driven by broadening AI demand. With 61% of the $6.3 billion AI order backlog coming from government and large enterprise customers, demand is clearly spreading beyond hyperscalers into the enterprise and public sector. The order pipeline continuously exceeding backlog, combined with management’s explicit statement that the current bottleneck is DRAM, NAND, and wafer supply constraints rather than demand, reinforces the structural nature of the growth.
Full-year guidance was raised sharply across the board: revenue growth of 29–33% (previously 17–22%), operating profit growth of 80–85% (previously 32–40%), FCF of at least $3.5 billion (previously $2.0 billion), and adjusted EPS of $3.35–3.45 (the previously set FY28 target of $3.00 achieved two years early). For FY2027, initial guidance includes FCF of at least $4.5 billion (substantially above consensus of $2.51 billion).
7. Credo FY4Q26 Earnings — FY2027 Optical Revenue Projected to Exceed $600M
• Core Source
“FY2026 revenue reached $1.3 billion, up 206% year-on-year. Non-GAAP net income reached $662 million, demonstrating strong operating leverage. Q4 revenue was $437 million, with a non-GAAP gross margin of 68.3%”
“Management projects each optical business segment to achieve over $100 million in revenue in FY2027, with total optical revenue expected to exceed $600 million”
“Agentic workloads present particularly large opportunities in terms of increasing front-end connectivity demand, with additional opportunities possible depending on architecture”
“Long-term, optical and copper product revenue mix is seen as capable of reaching 50:50. However, the total optical market is far larger than the AEC pluggable market even on a pluggable basis alone. Therefore, a scenario where optical revenue surpasses copper revenue in the long term is also possible”
• Expected Impact
Credo Technology recorded FY2026 annual revenue of $1.3 billion (+206% YoY), non-GAAP net income of $662 million, and Q4 gross margin of 68.3% (beating consensus of 65.4%), establishing itself as a core beneficiary of the AI data center interconnect market.
From a market size perspective, the key message from this earnings call is the diversification of demand sources. Management explained that agentic AI workloads are creating new front-end connectivity demand that was not previously anticipated, representing additional TAM beyond what was expected. All three optical portfolio segments (DSP, silicon photonics PIC, ZeroFlap optical modules) are each projected to independently exceed the overall FY27 growth rate, with ZF Optics, featuring a three-digit dollar ASP structure, expected to become the largest revenue contributor within the optical portfolio.
From a competitive landscape perspective, the transition from 800G to 1.6T is the key variable. This transition will simultaneously drive an increase in physical ports per GPU (adoption of 2-port designs) and ASP uplift, with management directly stating the possibility of a scenario where the long-term optical-to-copper revenue mix surpasses 50:50 in favor of optical. This means Credo’s TAM is expanding beyond copper interconnects into the broader optical communications market.
Comment [0]