AI Agents Ignite the Second Growth Curve of Cloud Computing! Wall Street Giants Analyze the IaaS Expansion Wave and PaaS Value Reassessment—Who Profits from the Real Value Behind the Token Craze?

According to English Financial News APP, a latest research report from Wall Street financial giant Bernstein indicates that the AI cloud computing sector is entering a new stage where an "AI agent craze dominates, driving an explosive surge in demand for AI computing power, upgrades in AI inference workloads, and a simultaneous reallocation of platform value." Correspondingly, investment opportunities are expanding from GPU computing power supply to leading cloud computing providers, which are crucial for enterprise data operations, core operational processes, connecting business workflows, and comprehensive platforms for AI agent development, deployment, and operation.

Bernstein notes that the global cloud computing industry is currently experiencing the most significant competitive shift since the rise of Microsoft Azure challenging Amazon AWS: training needs are driving data center expansions centered on GPUs, opening up the IaaS infrastructure market for new cloud service providers like CoreWeave; as AI inference workloads gradually enter enterprise production environments, there is an expanding need for coordinated CPU, GPU, database, storage, and software platform computing operations; OpenAI and Anthropic are further shifting from AI large model providers to AI agent development platform suppliers. This transition not only brings larger infrastructure orders to partner cloud providers but also sparks competition for PaaS entry points, customer relationships, and profit pools.

The Bernstein analyst team therefore favors Microsoft, Oracle, and MongoDB, while being more cautious about the long-term competitiveness of new cloud companies that only rent out AI GPU computing infrastructure. Their core investment thesis is that while growing demand for AI computing resources can expand the AI cloud rental market, not all cloud companies will gain the same pricing power or robust capital returns in the AI era. This judgment is reflected in recent industry data: Microsoft’s July financial report disclosed a 43% quarterly revenue growth in Azure and other cloud services, with demand still exceeding available capacity; Oracle’s September quarterly IaaS revenue increased 121% to $7.4 billion, with remaining performance obligations reaching $664 billion, indicating a conversion of demand into revenue and long-term contracts.

How do AI agents amplify cloud computing demand step by step? Based on the Bernstein report and engineering analysis of AI inference, the changes represented by Muse and Astra involve extending a single Q&A interaction into a task process encompassing planning, retrieval, tool invocation, execution, validation, and retry: Meta revealed Muse runs in a dedicated secure virtual machine with a browser, allowing users to continue processing tasks after closing the application; OpenAI disclosed that Astra enhances computer operation and multi-step professional work capabilities, but these advancements do not confirm AGI has been achieved.

As usage scales up, model calls increase GPU inference loads, browser, code, and tool executions raise CPU requirements, long contexts and concurrent sessions heighten memory and cache needs, and persistent state and enterprise knowledge access intensify database and storage demand. Production deployments further boost demand for identity, security, monitoring, and workflow orchestration PaaS services. The seven-year, $11.6 billion agreement between Anthropic and Akamai clearly targets CPU workloads, exemplifying the diffusion of AI demand beyond GPUs; partnerships with Amazon AWS, Google Cloud/Broadcom, Microsoft Azure, and SpaceX's AI cloud rental platform also demonstrate multi-vendor infrastructure deployments.

Regarding potential IaaS/PaaS winners in the AI cloud realm facing the rise of new players like CoreWeave, Bernstein assigns a top “Outperform” rating to Microsoft (MSFT.US), Oracle (ORCL.US), and MongoDB (MDB.US). Bernstein’s latest target prices suggest potential upside of roughly 28.7%, 135%, and 35% over the next 12 months, respectively. AI data warehouse and enterprise cloud database leader Snowflake (SNOW.US) receives a “Market Perform” rating, with a target of around 10% potential upside.

Undoubtedly, the AI cloud remains a key destination for industrial capital investment and profit expectations. Equity funds are seeking segments able to retain profits. OpenAI is seeking new funding at a valuation of around $1.4 trillion, and Anthropic is aiming for an IPO valuation above $2 trillion, reflecting the capital market’s high expectations around AI’s revenue creation prospects. Compared to AI application valuation narratives, the focus on the inference side of the AI compute resource supply chain already has more concrete and robust demand drivers: media reports reveal Anthropic’s infrastructure commitments could total $518 billion over the next decade, with about 80% non-cancellable or contractually guaranteed; Anthropic’s 2025 revenue is projected to multiply 12 times to nearly $4.6 billion year-on-year, with operating losses exceeding $8 billion, compute and infrastructure costs hitting $7.33 billion (about triple that of 2024), accounting for roughly 58% of total operating expenses of $12.65 billion; South Korea’s semiconductor exports in September reached $60.3 billion, a year-on-year increase of 262.8%, with official communications highlighting growth in memory unit volume and contract prices.

Soaring Compute Bills: Growth and Divergence Amid IaaS Expansion

Bernstein’s investment logic for IaaS in the AI era is that explosive incremental growth in AI computing demand is layered atop ongoing growth in traditional cloud computing. Citing Gartner projections, the report shows the global IaaS market is expected to grow from about $223 billion in 2025 to $287 billion in 2026, and to $662 billion in 2030. Traditional IaaS will grow from $201 billion to $488 billion, while AI-optimized IaaS climbs from $22 billion to $174 billion.

According to Bernstein’s calculations based on the report’s charted values, the market’s CAGR for 2025–2030 is about 24.3%, with the AI-optimized segment at about 51.2%, rising from roughly 9.9% to 26.3% of the total. The analyst team notes that while AI is creating a new, fast-growing AI cloud infrastructure market, migration of enterprises’ existing computing, storage, and applications to the cloud will continue to drive substantial absolute growth.

In terms of market landscape, Gartner projects AWS, Microsoft, Google, and Alibaba to have IaaS shares of 35%, 24%, 10%, and 8%, respectively, in 2025; the IDC equivalents are 41%, 16%, 8%, and 4%. Both firms estimate CoreWeave’s market share around 2%. The main differences stem from how IaaS and PaaS revenues are categorized—for example, IDC assigns more Oracle revenue to PaaS. Therefore, the different agencies’ figures are not directly comparable in constructing a unified competitive landscape.

Bernstein’s report also raises a key issue for investors: how will the $174 billion in AI-optimized IaaS revenues forecasted for 2030 coordinate with the market’s trillion-dollar GPU investment discussions? This may involve enterprise in-house builds, some investments ultimately monetized through PaaS, as well as possibly optimistic industry forecasts or conservative baseline estimates from Gartner and other researchers.

The temporary advantage of new clouds arises from compute resource scarcity, but long-term enterprise inference competitiveness requires complete software and data systems. Bernstein divides the sector’s evolution into three linked transitions: First, GPU training clusters have higher requirements for power, cooling, and high-bandwidth, low-latency networks, necessitating traditional data center retrofits or rebuilds, creating supply-demand gaps which allow new entrants and Oracle to gain traction; Second, AI spending expands from training to inference—Bernstein estimates inference annual expenditures may ultimately be 5–20 times those of training, though it does not specify when; Third, enterprise inference must connect to real business operations, requiring infrastructure that supports model computation, CPU-based application execution, and enterprise data access.

Bernstein notes that the resulting "data gravity" will attract some AI loads to platforms where enterprises already have databases, permissions, and applications deployed. The report sets five constraints on new clouds: difficulty surpassing GPU-centric IaaS; large cloud clients may repatriate workloads once supply normalizes; adopting open-source databases and development tools does not automatically create differentiation; enterprise data migration faces resistance; and the current supply-demand imbalance may abate in the next two years.

Bernstein states that for investors, the key logic to test long term is whether new cloud vendors can convert short-term delivery advantages into lasting customer retention, utilization, and cash returns; meanwhile, Oracle is seen as generating sustained growth momentum through sovereign cloud, private cloud, and AI data centers, potentially narrowing the gap with the likes of Microsoft, Amazon, and Google, the first tier of AI cloud infrastructure giants.

Whoever Controls the AI Agent Portal Shares Cloud Profits: PaaS Awaits Value Reappraisal

AI labs are emerging as major PaaS competitors, though revenue classification is still evolving. Bernstein’s report cites data showing the global PaaS market in 2025 at approximately $225.2 billion by Gartner and $227.5 billion by IDC—with year-on-year growth of about 25.8% and 42.3%, respectively, based on their 2024 estimates. IDC lists Microsoft first by share (21%), followed by AWS (12%), Google (9%), OpenAI (6%), Salesforce (5%), Oracle (4%), and about 2% each for Anthropic, Snowflake, and Databricks. Gartner, merging traditional and AI PaaS, gives AWS 19%, Microsoft 15%, Google 10%, OpenAI about 2%, and Anthropic less than 2%.

More directionally significant is that AI lab shares are rising across both agencies: IDC sees OpenAI climbing from 2.1% in 2024 to 5.7% in 2025, Anthropic from 0.4% to 1.9%; Gartner’s corresponding changes are 1.0% to 1.7% and 0.3% to 1.1%.

The boundary for measuring revenues from model APIs, development tools, coding products, and AI agent software is not yet unified, but both AI labs have a clear direction in expanding platform businesses. Bernstein expects their 2026 market shares could further rise to high single or low double digits, depending on revenue classification and the PaaS market’s growth rate, though this remains a forward-looking projection.

The model-to-platform transition means AI labs are competing for the entry point for enterprise software development and operation. According to Bernstein, Anthropic is extending from its Claude model to Claude Code, then into automation and agent application development; OpenAI, apart from ChatGPT, search, and research offerings, is rapidly building out Codex and agent development capabilities. These tools aid not only in constructing AI applications but also in enhancing traditional software development, testing, and maintenance efficiency. Platformization can boost workflow integration and customer stickiness, thereby mitigating competitive pressures from open-source or open-weight models and token price declines in pure model businesses.

According to the Gartner data cited by Bernstein, OpenAI’s enterprise revenues in 2025 will be about $5 billion, with base generative models accounting for 57%, dedicated models 4%, AI development platforms 15%, and other application software 24%; Anthropic’s enterprise revenues will be about $3.3 billion, with base models at 70%, dedicated models 7%, and other application software 23%. However, these are estimates for enterprise business segments and not equivalent to company-wide or annualized revenue run rates.

In the same period, the AI application development platform market will be about $6.8 billion, with Microsoft, AWS, and Google each taking 24%, 14%, and 14% shares, and OpenAI 11%, or about $750 million. In the foundational generative model market, OpenAI, Anthropic, and Google together account for 63% (25%, 20%, and 18% respectively). These numbers indicate that the dual-layer competition of models and platforms is underway. Bernstein even anticipates development platform revenues may outpace model revenues in 2026, though whether coding products fall under PaaS or SaaS is still undetermined; the larger strategic opportunity is to become the underlying platform for other software companies’ products, rather than entering verticals like CRM or HCM one by one. AI labs have model reasoning strengths; cloud giants and established SaaS vendors possess customer data, business semantics, deployment, security, and management systems, so competition will expand to the full enterprise delivery stack.

Bernstein’s stock selection in the AI cloud sector focuses on companies able to translate AI usage into sustained revenue, customer retention, and profit. Microsoft’s advantage lies in supporting OpenAI, Anthropic, various third-party and proprietary models, and leveraging extensive development tools and enterprise data to attract new AI workloads. Oracle covers AI data centers, OCI Gen2 public cloud, private cloud, Alloy white-label IaaS/PaaS, and sovereign cloud; the development of OpenAI’s platform services can reinforce its key client requirements, while currently not directly substituting for Oracle’s core database products. The report notes that multi-cloud partnerships have enabled triple-digit growth for Oracle’s cloud database segment in recent quarters.

Bernstein states the logic for MongoDB is more straightforward—AI applications or large-scale AI agent adoption across industries will require persistent data, querying, and state management for the long term. Increased application should expand database usage, and the report sees AI labs and new cloud players as not yet directly targeting its database market. By contrast, Snowflake’s pivot towards the AI platform increases its exposure to competition from both AI labs and hyperscale cloud players.

Bernstein emphasizes that investors should look for revenue realization across IaaS delivery capability, PaaS development entry points, and enterprise data infrastructure, while assessing free cash flow after depreciation, power, financing, and competitive costs. On risk, Bernstein adds: Microsoft faces cloud pricing competition and security risks; Oracle must manage customer concentration, realization of AI demand, and cloud business growth risks; MongoDB should watch for any slowdown in its cloud growth, as well as rising infrastructure costs and expenses.