1. Amazon, AI Shopping Technology Now Available to External Retailers
• Core Source
“Amazon has started opening its proprietary AI shopping technology to external companies through AWS. The company is expanding its business in the direction of selling ‘AI shopping guide capabilities’ themselves, moving beyond simply selling products directly.”
“This follows a similar trajectory to the strategy that gave rise to Amazon Web Services by commercializing internal infrastructure technology. Amazon states that retailers can build their own AI shopping assistant within approximately 60 days.”
“Amazon has started providing its AI shopping technology to other retailers, with Kate Spade becoming the first customer to sign a contract.”
“Amazon Rufus, as a shopping assistant, saw monthly active users grow +115% and engagement rate +400% year-over-year.”
• Expected Impact
Amazon’s move represents a strategic inflection point that could structurally reshape the e-commerce landscape, going far beyond simple B2B revenue expansion. The core is the same pattern AWS once followed. By opening internally refined technological infrastructure to the outside, Amazon has consistently created a structure that begins as a cost-reduction tool and ultimately leads to ecosystem lock-in.
The AI shopping technology now being sold to external retailers centers on an Alexa-based personalized recommendation engine. For retailers, the ability to deploy their own AI shopping assistant within approximately 60 days is a powerful incentive. If infrastructure that would otherwise take years to build internally can be implemented in a short timeframe on top of Amazon’s platform, adoption among small and mid-sized retailers is likely to accelerate rapidly.
Amazon’s internal metrics substantiate the strategy’s credibility. Rufus’s monthly active users growing 115% year-over-year and engagement surging 400% demonstrates that AI-driven shopping experiences are genuinely changing actual consumer behavior. Selling this proven technology externally means Amazon is converting its own performance metrics—validated on its own platform—directly into sales collateral.
The competitive dimension is equally significant. While OpenAI and Google are both actively entering the AI shopping and search space, Amazon is the only player capable of simultaneously leveraging existing AWS infrastructure, vast shopping data, and its logistics ecosystem. The more retailers adopt Amazon’s AI shopping infrastructure, the more Amazon entrenches itself as the sole entity simultaneously occupying the roles of seller, buyer, and technology provider.
Over the long term, this strategy could fundamentally transform Amazon’s revenue structure. On top of the direct-sales margin model, a recurring revenue stream based on technology platform subscriptions and usage fees is being layered in—a shift that could affect how Amazon’s enterprise value is assessed altogether.
2. AWS Pulls Ahead of Google and Microsoft in Cloud Profitability by Selling Anthropic Claude Tokens
• Core Source
“SemiAnalysis analyzed that AWS is quietly widening the gap in the cloud profitability race through a uniquely structured logic. The Bedrock platform accounts for only 4% of total AWS revenue, yet captures 30% of incremental gross profit growth, with an EBIT margin of 55%.”
“The core structure is a shift from the IaaS model of selling computing resources to the TaaS (Token-as-a-Service) model of securing model distribution rights. Anthropic’s quarterly net new ARR stands at $21 billion, and API revenue surged 13x year-over-year, serving as the key driver of this leverage structure.”
• Expected Impact
This analysis reveals that the nature of cloud competition is shifting from a battle over computing capacity to a battle over AI model distribution rights. The fact that AWS Bedrock accounts for only 4% of total revenue yet captures 30% of incremental gross profit growth means the token sales model carries dramatically higher margins than conventional IaaS. An EBIT margin of 55% substantially exceeds typical margins in cloud infrastructure businesses.
Understanding the structure is essential. AWS has executed large-scale investments in Anthropic (totaling $36.4 billion in Q1) while simultaneously securing rights effectively equivalent to exclusive distribution of Claude models. In a structure designed for enterprise customers to access the Claude API through AWS Bedrock, AWS operates not by directly bearing GPU compute costs, but by collecting fees as a middle-layer distributor of model access rights. This is the essence of TaaS.
Anthropic’s quarterly net new ARR of $21 billion and API revenue surging 13x year-over-year illustrate how rapidly this structure is scaling. As Claude usage grows, Bedrock revenue and margins expand in direct proportion—a compounding flywheel.
The implications for investors are clear. While Google Cloud and Microsoft Azure are pursuing analogous strategies through their own AI models (Gemini and Azure OpenAI respectively), the finding that AWS Bedrock leads on profitability metrics suggests the current AWS-Anthropic combination holds an advantage in model quality, adoption rate, and partnership structure. The emergence of measurable profitability gaps among the three cloud giants is an early signal that AI infrastructure investment returns will concentrate not in generic compute providers, but in model-platform integrated operators.
3. Snowflake Proves AI Is a Growth Engine, Not a Threat
• Core Source
“Snowflake delivered a landmark quarter, achieving product revenue of $1.33 billion, 34% year-over-year growth, and the strongest sequential dollar growth in the company’s history.”
“AI is acting as a powerful tailwind for Snowflake, and Q1 marked a definitive inflection point in that journey.”
“AI coding agent CoCo reached 7,100 accounts in just one quarter since launch.”
“Signed a new 5-year, $6 billion contract with AWS, with expected benefits from joint GTM investment and cost reduction.”
“FY2027 revenue growth guidance raised from 27% to 31%, an increase of 4 percentage points.”
• Expected Impact
Snowflake’s results provided a clear answer to a question that has hung over the enterprise software industry: “Will AI cannibalize data platform companies, or will it create demand?” Snowflake answered with numbers: the latter.
Product revenue growing 34% year-over-year and the fact that the sequential dollar increase in revenue was the largest in company history means that the slope of the growth curve itself has steepened—this is not merely a solid quarter. The key driver is AI-native products. The fact that AI coding agent CoCo secured 7,100 accounts in a single quarter since launch shows enterprise customers are integrating AI capabilities not as experiments, but into actual workflows. These users are drawn into a flywheel structure where increased AI feature usage drives incremental growth in existing platform consumption.
The full-year guidance raise (27%→31%) and the new 5-year, $6 billion contract with AWS signal that medium-to-long-term growth visibility has been secured, not merely a short-term improvement. The AWS contract is not a simple infrastructure deal—it includes joint GTM investment, which can translate into lower customer acquisition costs and an expanded pipeline of large deals.
A net revenue retention rate of 126% and 616 net new customers (up 38% year-over-year) demonstrate that both existing customer expansion and new customer acquisition are simultaneously accelerating. This signals that the role of a centralized data management platform is being reinforced, not eroded, in the AI era. Over the long term, Snowflake’s positioning at the intersection of data governance and AI agent operating infrastructure is likely to grow more entrenched.
4. Marvell Technology: Explosive Demand in ASIC and Interconnect
• Core Source
“Q1 revenue came in at $2.418 billion, growing 9% quarter-over-quarter and 28% year-over-year. Non-GAAP gross margin stood at approximately 58.9% and operating margin at approximately 35%. Operating cash flow also hit a record high of $639 million.”
“FY2027 revenue is projected at approximately $11.5 billion (+40% YoY), and FY2028 revenue at approximately $16.5 billion (+45% YoY).”
“Data center revenue reached $1.83 billion, accounting for approximately 76% of total revenue. The interconnect business is projected to grow more than 70% in FY2027.”
“The CEO described AI orders as ‘abnormally strong’ and the data center business as ‘completely explosive,’ reconfirming that custom ASIC revenue is on track to exceed $10 billion in FY2029.”
“The interconnect business growth outlook has been raised from 50% to more than 70%.”
• Expected Impact
Marvell’s results and guidance are a clear signal that AI infrastructure investment benefits are expanding beyond Nvidia GPUs to networking and interconnect semiconductor companies that serve as the connective tissue of AI data centers.
The fact that data centers account for 76% of total revenue means Marvell has effectively transformed into a dedicated AI infrastructure company. The revision of the interconnect business growth outlook from 50% to over 70% is particularly significant. As AI servers grow larger and denser, data transfer speed and bandwidth between GPUs and between servers emerge as critical bottlenecks—and the products that solve these problems are precisely Marvell’s core portfolio: 800G and 1.6T optical interconnects, 51.2T Ethernet switches, and DCI modules.
The reconfirmation of a $10 billion FY2029 revenue target for the custom ASIC business is also worth noting. As hyperscalers increase development of proprietary AI accelerators (custom ASICs) to reduce dependence on Nvidia GPUs, Marvell serves as a ‘custom ASIC design partner’ that receives customer specifications and translates them into silicon, connecting design through to mass production. Google’s TPU and Amazon’s Trainium are representative examples of this structure.
FY2027 revenue guidance of $11.5 billion (+40%) and FY2028 of $16.5 billion (+45%) shows that the pace of growth is actually accelerating over the next two years. Barclays raising its price target from $150 to $275 and projecting that custom AI semiconductor revenue could exceed $10 billion over the long term reflects the market’s increasing confidence in the credibility of this growth trajectory.
5. TSMC Considers Up to 15% Price Hike for 3nm in H2, with Additional 5–10% Increase Eyed for 2027
• Core Source
“TSMC is considering raising 3nm process prices by up to 15% in the second half of 2026, with a further increase of 5–10% possible in 2027 (TrendForce).”
“TSMC, AI·ASIC demand drives consideration of up to 15% 3nm price hike in 2H26… Additional 5–10% increase forecast for 2027 (TrendForce).”
“CEO Huang stated that ‘TSMC is the world’s best company, and given the difficulty of the work they perform and the value they create, the profits they earn are entirely justified,’ expressing unwavering partnership support.”
• Expected Impact
TSMC’s 3nm price hike is not a routine rate adjustment—it signals that the cost structure across the entire AI semiconductor supply chain is being reset to a new level. If a hike of up to 15% in H2 and an additional 5–10% in 2027 materialize, the cumulative increase will exceed 20%.
The demand backdrop underpinning this pricing move is exceptionally firm. Nvidia, AMD, Apple, Google, Amazon, Broadcom, and virtually every major fabless company are concentrating on advanced nodes at or below 3nm, with AI chip and ASIC demand exceeding TSMC’s available advanced-node capacity. In an environment where supply constraints are structurally sustained, TSMC’s pricing power is formidable. The fact that Jensen Huang publicly endorsed TSMC’s price increases as justified—with the world’s largest customer effectively accepting the hike—substantiates this reality.
For fab customers, this cost increase flows through in two directions. One is passing it on through higher final product prices. The other is diversifying production outsourcing to alternative foundries such as Intel Foundry and Samsung Foundry. However, given TSMC’s current yield advantages and technology gap, near-term switching is realistically difficult. As a result, TSMC’s profitability is set to improve structurally, while upward cost pressure propagates across AI semiconductor systems broadly. As AI infrastructure investment costs rise, the relative value of efficient inference chips and software optimization companies may simultaneously come into sharper focus.
6. Power Semiconductors: A Wave of Sequential Price Increases
• Core Source
“Infineon will raise prices on certain products again from July 1, 2026, following an earlier increase in April of this year.”
“Texas Instruments (TI): Raising prices on PMIC and MOSFET product lines effective July 1—this is the second increase of the year.”
“MacMic: Plans to raise IGBT product prices by 10%.”
“Jiangsu JieJie: Plans to raise MOSFET and IGBT product prices by 10% to up to 20%.”
“Texas Instruments’ Q1 data center revenue surged approximately 90% year-over-year, underscoring this trend.”
• Expected Impact
The wave of sequential price increases delivers a core message: the AI semiconductor boom is spreading beyond GPUs and HBM into the power semiconductor market. The back-to-back price hikes from Infineon, TI, MacMic, and Jiangsu JieJie reflect not individual company margin strategy, but a structural supply-demand imbalance across the entire supply chain.
TI’s Q1 data center revenue surging 90% year-over-year explains this directly. As AI data centers are built at scale, demand for power management (PMIC), high-voltage switching (MOSFET), and power conversion (IGBT) components in each server rack is growing explosively. The number of power semiconductors required to build a single data center hall is orders of magnitude greater than the number of GPUs—making it structurally inevitable that AI infrastructure investment benefits concentrate in this space.
The character of these price hikes matters. Analysts characterize this not as temporary inventory restocking but as a reflection of structural demand change. As Nvidia’s GB300 platform rollout and high-voltage direct current (HVDC) architecture deployment accelerate, demand for high-voltage MOSFETs and IGBTs will increase further. In other words, the current price increases are closer to the beginning of a cycle than the end. Infineon and TI executing their second price hike of the year reflects the power semiconductor market’s characteristic: supply capacity additions lag demand growth significantly. The operating margin improvement and strengthening pricing power of these companies are likely to persist for some time.
7. ByteDance Begins Development of Proprietary CPU
• Core Source
“Amid rising chip prices and prolonged supply shortages constraining business expansion, Chinese tech giant ByteDance is reportedly developing its own CPU to meet growing AI infrastructure demand.”
“ByteDance currently purchases CPUs from Intel and AMD, but both suppliers have recently raised prices sharply, with increases of 10–35% on a quarterly basis, which is also accelerating the push for proprietary CPU development.”
“ByteDance plans to deploy its proprietary CPU in internal servers and data centers to support internal operations, while simultaneously preparing for the large-scale launch of agent-type products including the Coze platform. The CPU is currently being evaluated under two architectural directions: Arm-based and open-source RISC-V.”
• Expected Impact
ByteDance’s move to develop its own CPU operates on two levels. In the short term, it is a cost-reduction strategy to reduce dependence on Intel and AMD CPUs. Over the medium to long term, it represents a structural shift toward self-sourcing the core infrastructure of the AI agent era.
The intensity of the cost pressure is spelled out in the numbers. Intel and AMD raising CPU prices by 10–35% on a quarterly basis translates to an annualized rate of 40–140%—an acute surge in unit costs. At ByteDance’s scale of operating hundreds of thousands of servers, this cost increase directly translates into hundreds of millions of dollars in additional expenditure. Even with the substantial upfront investment required for proprietary chip development, the long-term cost savings are sufficient to justify it at this scale.
The technical direction is also worth noting. Simultaneously evaluating both Arm-based and RISC-V architectures implies a potential shift toward greater reliance on open-source RISC-V in preparation for tightening U.S. semiconductor export controls. Given that Arm licenses could themselves become subject to U.S. regulatory action, this is a strategic choice that goes beyond cost reduction into supply chain risk management.
The larger implication lies in the nature of AI agent workloads. This move reflects that AI is rapidly transitioning from a GPU-centric training (Training) phase to an inference (Inference) phase—and within that, for the lightweight, agent-type inference workloads ByteDance is targeting, CPUs are considered relatively more suitable than GPUs. The Coze platform and similar agent-type services generate large volumes of workloads that are better suited to lightweight CPU-based inference than to GPU processing. ByteDance’s proprietary CPU development is a structural signal that the broader trend of big tech semiconductor internalization is now spreading to Chinese companies as well.
Comment [0]