AI agents ignite the cloud computing second growth curve! Wall Street major banks break down the IaaS expansion wave and PaaS value re-rating who is pocketing the real gold from the token frenzy?

Wall Street financial giant Bernstein's latest research report shows that the AI cloud computing sector is entering a new phase characterized by the simultaneous occurrence of "explosive expansion of AI computing power demand driven by the AI agent frenzy, upgrading of AI inference workloads, and redistribution of platform value," with investment opportunities extending from GPU computing power supply to cloud computing leaders that are critical to enterprise data operations, core enterprise operational processes, and integrated platforms connecting business workflows with agent development, deployment, and operation. Bernstein states that the global cloud computing industry is currently undergoing its most significant round of competitive change since Microsoft Corporation Azure rose to challenge Amazon.com, Inc. AWS: training demand has spawned GPU-centric data center expansion, opening up the IaaS infrastructure market for new cloud service providers such as CoreWeave; as AI inference workloads gradually move into enterprise production environments, the demand for coordinated computing power operations across CPUs, GPUs, databases, storage, and software platforms expands; OpenAI and Anthropic are further shifting from AI large model suppliers to AI agent development platform suppliers, contributing larger infrastructure orders to partner cloud computing vendors while also beginning to compete for the software entry points, customer relationships, and profit margins of PaaS cloud computing vendors. Bernstein's analyst team is therefore more bullish on Microsoft Corporation, Oracle Corporation, and MongoDB, but more cautious about the long-term competitiveness of new cloud enterprises that simply rent out AI GPU computing power infrastructure. The institution's core investment judgment is that growth in AI computing power resource demand can expand the AI cloud rental market, but will not give all cloud enterprises the same pricing power and strong capital returns in the AI era. This judgment echoes recent industry data: Microsoft Corporation's July earnings disclosure showed Azure and other cloud services quarterly revenue growth of 43%, with demand still exceeding available capacity; Oracle Corporation's September quarterly IaaS revenue grew 121% to $7.4 billion, with remaining performance obligations reaching $664 billion, indicating that demand is being converted into revenue and long-term contracts. How do AI agents gradually amplify cloud computing demand? Combining Bernstein's research report with AI inference engineering deductions, the changes represented by Muse and Astra are expanding a single question-and-answer into a task process that includes planning, retrieval, tool invocation, execution, checking, and retrying: Meta disclosed that Muse runs in a dedicated secure virtual machine with a browser, and users can continue processing tasks after closing the application; OpenAI disclosed that Astra enhances computer operation and multi-step professional work capabilities. These capability advances themselves do not equal confirmation that AGI has been achieved. As usage scale expands, model calls increase GPU inference loads, browser, code, and tool execution increase CPU demand, long context and concurrent sessions raise memory and cache requirements, persistent state and enterprise knowledge access increase database and storage demand, and production deployment further drives PaaS services such as identity, security, monitoring, and workflow orchestration. The seven-year $11.6 billion agreement signed between Anthropic and Akamai explicitly targets CPU workloads, which is an example of AI demand spreading beyond GPUs; its cooperation with Amazon.com, Inc. (Amazon) AWS, Google Cloud/Broadcom Inc. (Broadcom), Microsoft Corporation Azure cloud computing, and SpaceX's AI cloud computing rental platform also reflects multi-vendor infrastructure configuration. For potential IaaS/PaaS winners in the AI cloud computing sector facing the strong rise of new cloud forces such as CoreWeave, Bernstein gives Microsoft Corporation (MSFT.US), Oracle Corporation (ORCL.US), and MongoDB (MDB.US) its most positive "Outperform" rating, with the latest target prices implying potential upside of approximately 28.7%, 135%, and 35% over the next 12 months, respectively. The institution gives Snowflake (SNOW.US), a leader in AI cloud data warehousing and enterprise cloud databases, a "Market Perform" rating, with the target price implying potential upside of about 10%. Undoubtedly, AI cloud remains an important direction for industrial capital investment and profit expectations, while equity funds are screening for the segments that can retain profits. OpenAI seeking new financing at a valuation of about $1.4 trillion, and Anthropic seeking an IPO valuation of more than $2 trillion, reflect extremely high expectations in global capital markets for AI-related revenue creation space. Compared with AI application valuation narratives, the AI computing power resource supply chain focused on the inference side already has more concrete evidence of strong demand: media disclosed that Anthropic's infrastructure arrangements of approximately $518 billion over the next decade, of which about 80% is non-cancelable or payable as agreed; Anthropic's 2025 revenue increased to about 12 times the previous year, approaching $4.6 billion, with operating losses exceeding $8 billion, and computing power and infrastructure spending reaching $7.33 billion, about three times that of 2024, accounting for about 58% of total operating expenses of $12.65 billion;; South Korea's September semiconductor exports were $60.3 billion, up 262.8% year over year, and officials also noted that memory export volumes and contract prices rose. The computing power bill is growing larger and larger: growth and divergence in the IaaS expansion wave Bernstein's investment logic for IaaS in the AI era is the strong incremental growth brought by the continued explosion of AI computing power demand on top of the continued growth of traditional cloud computing. Gartner forecasts cited in the report show that the global IaaS market will grow from about $223 billion in 2025 to $287 billion in 2026 and $662 billion in 2030; within that, traditional IaaS will increase from $201 billion to $488 billion, while AI-optimized IaaS will increase from $22 billion to $174 billion. Calculations based on the values shown in Bernstein's research report charts indicate that the overall market compound growth rate from 2025 to 2030 is about 24.3%, while the AI-optimized portion is about 51.2%, with its share rising from about 9.9% to 26.3%. Bernstein's analyst team states that this also means AI not only creates a high-growth new AI cloud computing infrastructure market, but enterprises' existing migration of computing, storage, and applications to the cloud also continues to contribute a massive absolute increment. In terms of the competitive landscape, under Gartner's 2025 methodology, the IaaS shares of AWS, Microsoft Corporation, Alphabet Inc. Class C, and Alibaba Group Holding Limited Sponsored ADR are 35%, 24%, 10%, and 8%, respectively; the corresponding IDC figures are 41%, 16%, 8%, and 4%, and both major research institutions count CoreWeave's market share at about 2%. The differences mainly involve the revenue division between IaaS and PaaS; for example, IDC classifies more Oracle Corporation (Oracle) revenue under PaaS, so shares from different institutions cannot be stitched together into the same equivalent competitive map. Bernstein's research report also raises a question worth tracking for investors: how can $174 billion in AI-optimized IaaS revenue in 2030 be reconciled with the trillion-dollar-scale GPU investment discussed in the market? This may involve enterprises building their own facilities, some investment ultimately being monetized through PaaS, and industry forecasts being too high or benchmark forecasts from market research institutions such as Gartner being too low. The advantages of new clouds come from computing power scarcity, but the long-term competitiveness of enterprise inference requires a complete software and data system. Bernstein's research report divides industry evolution into three interconnected changes: first, GPU training clusters place higher demands on power supply, cooling, and high-bandwidth low-latency networks, and traditional data centers need to be retrofitted or rebuilt, so the supply-demand gap creates entry opportunities for new cloud enterprises and Oracle Corporation; second, AI spending expands from training to inference, and Bernstein estimates that annual inference spending may ultimately reach 5-20 times that of training, but does not give a clear year for realization; third, enterprise inference needs to connect to actual business, and its infrastructure must support model computation, CPU application execution, and enterprise data access. Bernstein states that the "data gravity" generated by this trend will attract some AI workloads to platforms where enterprises have already deployed databases, permission systems, and applications. The report raises five constraints on new clouds: it is difficult to surpass GPU-based IaaS; large cloud customers may move workloads back to their own facilities once supply is sufficient; adopting open-source databases and development tools does not automatically create differentiation; enterprise data migration faces resistance; and the current supply-demand imbalance may improve over the next two years. Bernstein states that for investors, the logic that truly needs long-term testing is whether new cloud forces can convert short-term delivery advantages into long-term customer retention, utilization, and cash returns; at the same time, the research report believes that Oracle Corporation, with sovereign cloud, private cloud, and AI data centers forming continuous growth momentum, has the opportunity to move further toward the first tier of AI cloud computing infrastructure players such as Microsoft Corporation, Amazon.com, Inc., and Alphabet Inc. Class C. Whoever controls the agent entry point shares strongly in cloud profits: PaaS ushers in value re-rating AI labs are becoming important competitors in PaaS, but revenue classification is still in a formative stage. Data cited in Bernstein's report shows that the global PaaS market in 2025 is about $225.2 billion under Gartner's methodology and about $227.5 billion under IDC's methodology; calculated based on their respective 2024 data, year-over-year growth is about 25.8% and 42.3%, respectively. IDC ranks Microsoft Corporation first with a 21% share, followed by AWS at 12%, Alphabet Inc. Class C at 9%, OpenAI at about 6%, Salesforce at about 5%, Oracle at about 4%, and Anthropic, Snowflake, Databricks, and others at about 2% each; after Gartner combines traditional and AI PaaS, AWS leads with 19%, Microsoft Corporation has 15%, Alphabet Inc. Class C has 10%, OpenAI has about 2%, and Anthropic has less than 2%. More directionally significant is that under the same methodology, AI lab shares are all rising: IDC statistics show OpenAI rising from 2.1% in 2024 to 5.7% in 2025, and Anthropic rising from 0.4% to 1.9%; the corresponding Gartner changes are 1.0% to 1.7% and 0.3% to 1.1%. The revenue measurement boundaries for model APIs, development tools, coding products, and agent software have not yet been unified, but the direction of the two AI labs expanding their platform businesses is relatively clear. Bernstein expects their 2026 shares may further rise to high single digits or even low double digits, depending on revenue classification and the growth rate of the entire PaaS market; this remains a forward-looking judgment. Moving from models to platforms means AI labs are beginning to compete for the development and operation entry points of enterprise software. Bernstein's research report believes that Anthropic extends from the Claude model to Claude Code, and then into automation and agent application development; OpenAI, in addition to ChatGPT, search, and research products, is accelerating the development of Codex and agent development capabilities. These tools not only help build AI applications, but can also improve the efficiency of traditional software development, testing, and maintenance. Platformization can increase workflow integration and customer stickiness, thereby alleviating the pressure on pure model businesses from competition among open-source or open-weight models and falling token prices. The Gartner report cited by Bernstein shows that in 2025 OpenAI's enterprise revenue was about $5 billion, of which foundation generative models accounted for 57%, specialized models 4%, AI development platforms 15%, and other application software 24%; Anthropic's enterprise revenue was about $3.3 billion, corresponding to 70% foundation models, 7% specialized models, and 23% other application software. However, these are estimates by enterprise business classification and cannot be equated with the company's total revenue or annualized revenue run rate. In the same period, the AI application development platform market was about $6.8 billion, with Microsoft Corporation, AWS, and Alphabet Inc. Class C accounting for 24%, 14%, and 14%, respectively, and OpenAI accounting for 11%, corresponding to about $750 million; in the foundation generative model market, OpenAI, Anthropic, and Alphabet Inc. Class C accounted for 25%, 20%, and 18%, respectively, totaling 63%. This set of data shows that competition at both the model and platform layers has already begun simultaneously. Bernstein even expects that development platform revenue may exceed model revenue in 2026, but whether coding products are classified under PaaS or SaaS still needs to be clarified; its greater strategic opportunity is to become the underlying platform on which other software enterprises build products, rather than entering vertical markets such as CRM and HCM one by one. AI labs have advantages in model understanding, while cloud giants and existing SaaS vendors have customer data, business semantics, deployment, security, and management systems, so competition will extend to the entire set of enterprise delivery capabilities. Bernstein's stock selection focus in AI cloud computing is on enterprises that can convert AI usage into sustained revenue, customer retention, and profit. Microsoft Corporation's advantage lies in simultaneously hosting OpenAI, Anthropic, multiple third-party models, and its own models, and attracting new AI workloads with a broad range of development tools and enterprise data accumulation; Oracle Corporation covers AI data centers, OCI Gen2 public cloud, private cloud, Alloy white-label IaaS/PaaS, and sovereign cloud, and OpenAI's platform business development can strengthen demand from its important customers while at present not directly replacing Oracle's core database products. The report also notes that multi-cloud cooperation has driven triple-digit growth in its cloud database in recent quarters. Bernstein states that MongoDB's logic is more directthat is, AI applications or the large-scale penetration of AI agents into various industries will require persistent data, queries, and state management over the long term, and an increase in applications may expand database usage. The report believes that AI labs and new clouds are not currently directly targeting its database market. By contrast, Snowflake's extension into an AI platform also exposes it to dual competition from AI labs and hyperscale cloud vendors. Bernstein emphasizes that investors should look for revenue realization along IaaS delivery capabilities, PaaS development entry points, and enterprise data infrastructure, while also testing free cash flow after depreciation, power, financing, and competitive costs; on investment risks, Bernstein adds that Microsoft Corporation faces cloud price competition and security risks; Oracle Corporation needs to address customer concentration, AI demand realization, and cloud business growth risks; MongoDB needs to guard against slowing cloud growth and rising infrastructure costs and expenses.