Cloud Providers Surge While Platforms Face Pressure: AI Returns Diverge Further
According to Woofun AI, JPMorgan (JPM.US) revealed in its latest research report a significant divergence in investment returns within the artificial intelligence sector: infrastructure giants Google (GOOGL.US) and Amazon (AMZN.US) have received positive revaluations, while application-layer and platform companies like Meta (META.US) and Fiverr (FVRR.US) are facing profit downgrades or traffic crises.
This structural difference indicates that the commercialization process of AI technology is not distributed evenly but is profoundly impacting the revenue structure, cost efficiency, and cash flow performance under various business models.
The growth logic of Google Cloud (GOOGL.US) is undergoing reconstruction, with the core driver now stemming from the inclusion of self-developed chip TPU sales. JPMorgan (JPM.US) has significantly raised its year-on-year revenue growth forecast for Google Cloud (GOOGL.US) in Q3 2026 from 88% to 100%, and in Q4 from 90% to 105%. More notable adjustments appear in the medium- to long-term forecasts: full-year growth in 2027 is raised from 70% to 107%, and for 2028 from 55% to 63%. These figures are institutional forecasts and not yet the company’s disclosed results. JPMorgan (JPM.US) notes that the main reason for the upward revision is the inclusion of more Tensor Processing Unit (TPU) sales in the model, and believes that the adjusted forecasts may still be conservative. TPU is Google’s (GOOGL.US) dedicated chip for machine learning computation, and its sales growth is becoming an important variable in the bank’s assessment of Google Cloud (GOOGL.US) revenues.
This means that understanding Google Cloud’s (GOOGL.US) growth requires a closer look at revenue composition; the 107% growth forecast includes more contributions from TPU sales and cannot be attributed solely to the acceleration of traditional cloud service consumption. Differences in profitability, delivery pace, and cash recovery characteristics across business lines will also affect how much profit revenue growth ultimately brings. At the group level, JPMorgan (JPM.US) raised Alphabet’s (GOOGL.US) 2027 U.S. GAAP EPS forecast by around 3% to $14.05 and its 2028 forecast by about 4% to $17.29; the December 2027 price target was raised from $420 to $425, maintaining an ‘Overweight’ rating, corresponding to about 25 times the 2028 expected EPS.
While the revenue growth outlook for the cloud business has been substantially raised, the upward adjustment of the group's profit forecast is relatively mild. The elasticity of profit from single-business growth still depends on revenue scale, cost structure, and performance of other group businesses. Financing was also included in this revision: JPMorgan (JPM.US) incorporated the $5 billion debt financing completed by Waymo (GOOGL.US) in early October and assumed an additional $80 billion in debt financing in 2027. The latter is a forecast assumption and should not be interpreted as announced or completed financing.
AWS (AMZN.US) is also highly regarded for its ability to monetize computing power, benefiting from the expansion of its partnership with OpenAI, large-scale GPU deployments, and sequential price increases for reserved computing power. JPMorgan (JPM.US) raised its growth forecast for AWS (AMZN.US) but emphasized the need to monitor delivery execution and margin risks.
Woofun AI’s aggregated data reveals that JPMorgan (JPM.US) expects AWS (AMZN.US) revenue to accelerate to 40% growth in 2027, possibly still a conservative estimate. This judgment is mainly supported by two factors: first, the expanded scale of cooperation with OpenAI and the additional deployment of 2 million GPUs between 2027 and 2028; second, the pricing power of AI computing capacity. Added computing power determines potential delivery scale, while pricing affects how much revenue can be generated from the same resources. According to the report, AWS (AMZN.US) raised prices for reserved GPU computing capacity by 15% starting October 7, marking the fourth consecutive quarter of price hikes for relevant AI computing capacity.
The report also notes that prices for H100/H200 and B200/B300 GPUs have risen between 40% and 75% since the start of the year. Reserved computing capacity refers to service arrangements where customers book certain compute capacity in advance. The price increases mentioned in the report are concentrated on specific GPUs and reserved AI computing power; it cannot be concluded that all AWS (AMZN.US) products have increased prices to the same extent. JPMorgan (JPM.US) also stresses that customer demand is relatively broad: AWS (AMZN.US) allocates computing power quite diversely, with the largest client accounting for only a single-digit percentage of the total. Thus, their growth argument includes not only the increment from large AI partnerships but also broader demand across a wide client base.
Together, these factors led the bank to raise its 2027 and 2028 forecasts for Amazon (AMZN.US): group sales are increased by about 1% to 2%, operating profit by about 3% to 5%, and GAAP EPS by about 2% to 3%. JPMorgan (JPM.US) maintains a December 2027 price target of $365 and an ‘Overweight’ rating, using a valuation of about 27.5 times the 2028 expected EPS of $13.34. The upward revision in revenue and profit supports their positive stance, but AWS (AMZN.US) still needs to translate partnership expansion and deployment plans into actual delivery. The report also lists the early-stage and limited visibility of generative AI monetization and ongoing investment that may pressure margins as risks. Although price increases provide revenue support, ultimate profit contributions will depend on equipment, operations, and delivery costs.
In contrast to the upward revisions for cloud businesses, the adjustments for Meta (META.US) mainly reflect increased expenses. Despite improvements in advertising and AI monetization, large accruals for legal expenses and high capital expenditures have led to negative free cash flow. JPMorgan (JPM.US) has downgraded its EPS forecast for META (META.US) and cautions about short-term expenditure pressure. JPMorgan (JPM.US) added a $10 billion legal expense accrual to its Q3 2026 model, attributing it to a settlement with 52 State Attorneys General, and notes that this expense was not included in the company’s Q2 full-year expense guidance of $165–169 billion. As a result, the bank lowered Meta’s (META.US) 2026 GAAP EPS forecast by about 10% to $27.99 and the 2027 forecast by about 2% to $31.08.
Meanwhile, JPMorgan (JPM.US) raised the 2027 GAAP total expense forecast for Meta (META.US) to about $210 billion, up approximately 18% year-on-year; excluding the $10 billion legal accrual from the 2026 base, the expense increase for 2027 would be about 25%. The moderate headline growth rate partly reflects the higher base due to a one-off expense the previous year. The revenue side saw marginal improvement, with the bank raising Meta’s (META.US) 2027 revenue forecast by about 0.9% to $307.9 billion, though this growth is insufficient to fully offset the impact of higher expense forecasts. Capex and cash flow offer another perspective: according to the report’s model, Meta’s (META.US) 2027 capital expenditure is around $242.7 billion, with free cash flow (FCF) at negative $72.8 billion.
Free cash flow reflects the balance of cash generated from operations after capital expenditures; the negative value in the model indicates that operating cash is expected to be insufficient to cover capital expenditures during the same period. Meta’s (META.US) core contradiction is that improvements in advertising and AI monetization need to catch up to massive investment levels. JPMorgan (JPM.US) believes that AI-powered ad ranking and recommendation are boosting user engagement and ad returns, and sees potential for monetizing consumer and enterprise agents as well as model interface services, thus maintaining an ‘Overweight’ rating and a $920 price target. However, the bank also clearly warns that if AI monetization lags, ongoing investment will weigh on operating profits, EPS, and free cash flow. Maintaining the rating reflects long-term return confidence, but near-term spending pressure remains.
Fiverr’s (FVRR.US) traffic crisis and transformation challenges further highlight the impact of AI on traditional service platforms. The shift in search habits due to AI summaries has caused a substantial drop in platform visits; the transition towards high-value services has been slow. JPMorgan (JPM.US) has downgraded its forecasts for revenue and profit, maintaining a Neutral rating. Citing SimilarWeb data, JPMorgan (JPM.US) notes Fiverr’s (FVRR.US) U.S. website traffic dropping 10% year-on-year in Q2 and expanding to a 22% decline during the reported period for Q3 to date, with monthly declines of 19% in July and 24% in August. Thus, this data should not be interpreted as the complete Q3 result.
According to JPMorgan (JPM.US), AI-generated summaries and related changes in Google’s (GOOGL.US) search diminish opportunities for users to click on paid or organic search links. Even though Fiverr (FVRR.US) has invested in Generative Engine Optimization (GEO) to increase exposure in AI-generated answers, click-through rates from large language models remain below that of traditional search, thus pressuring the platform’s ability to acquire potential customers. The supply side of services is also adjusting, with Fiverr (FVRR.US) shifting from a marketplace focused on low-value projects to a platform for higher-end, higher-value work.
However, the report notes that high-value business currently accounts for only 15% of the platform’s gross merchandise volume (GMV), and management expects the transformation to take at least six quarters. Based on this, JPMorgan (JPM.US) projects Fiverr’s (FVRR.US) 2027 revenue will decline about 15%, the same as in 2026, and has cut its 2027 and 2028 revenue forecasts by about 6% each. The adjusted EBITDA forecasts are reduced by 21% and 14%, respectively. The bank maintains a ‘Neutral’ rating and lowers the price target from $11 to $10.
Other ride-hailing platform adjustments and overall summary indicate that assessing AI returns requires a granular examination of revenue, cost, and cash; each entity’s logic differs. Uber (UBER.US) restructuring involves about 3,300 employees, and JPMorgan (JPM.US) estimates annual savings of more than $600 million. However, since the company plans to reinvest in price reductions, service choices, and growth businesses, these are not yet counted as net cost savings. Lyft (LYFT.US) faces additional costs and cash outflows due to a settlement over driver classification disputes; the bank maintains a Neutral rating and a $19 price target. These adjustments collectively demonstrate that evaluating the returns of AI and platform businesses requires meticulous inspection of revenue sources, cost changes, and cash flow direction. Google Cloud’s (GOOGL.US) TPU sales, AWS’s (AMZN.US) computing power delivery and pricing, Meta’s (META.US) expenses and capital expenditures, as well as Fiverr’s (FVRR.US) traffic and transformation progress, will each determine whether their respective growth logics are fulfilled. This is a deep stress test for the monetization ability of the AI application layer following the infrastructure arms race, and the diverging trend may continue throughout the cycle.
