The latest earnings reports from the ‘Magnificent 4 hyperscalers’ prompted some of the weakest market reactions that we have seen in recent years. The performance between Meta, Amazon, Alphabet, and Microsoft diverged significantly. This upholds our team’s long-held mantra that the Mag 7 should not be treated as a ‘monolith’, each business has different earnings drivers, business models, management teams, pathways to AI monetisation – reinforcing the importance of active investment across these names.
Here we identify the core themes and takeaways from the results and some of the associated key concerns:
1 AI demand remains strong and broad. Cloud backlogs are providing much stronger visibility into contracted demand.
Company earnings reported in the third quarter of 2026 saw the fastest growth in revenues for AWS ( Amazon Web Services) in five years, Microsoft’s Azure cloud business grew 48%, while Google Cloud’s business more than doubled (+ 103%). Token growth has continued to accelerate strongly with enterprises and consumers using both frontier and open weight models managed on these platforms. That growth is driving higher backlogs (performance obligations), which has tripled this year for Alphabet and Amazon. Microsoft guided to further Azure revenue acceleration of 49-50% next quarter, while Amazon was able to guide revenue growth for AWS not only for 2027 but also for 2028. Putting a floor (not a ceiling) on its AWS revenue growth of 40% offers strong visibility on AI capex monetisation. There is a notable change in the breadth of customers that make up that backlog. This demand is no longer concentrated around a limited number of AI frontier labs like OpenAI and Anthropic. It is being driven by a wide range of governments and enterprises across the broader economy who are scaling their use of AI from training into inferencing workloads. So, unlike in the dot.com era, customer demand is not coming from IT capex budgets alone. It’s also from operating expenditure, research and development (R&D), and marketing : the entire knowledge work that is being disrupted. A more than US$30 trillion potential opportunity (according to Anthropic estimates)1 is currently showing as a backlog of US$1.7 trillion between the largest hyperscalers.
2 Focus on FCF inflection. Patience is required given bottlenecks mean there is a time lag between capex turning into additional compute capacity that will later translate into revenues.
Capital expenditure (capex) for the recent quarter exceeded operating cash flow for Amazon, Alphabet and Meta, with deteriorating free cash flow (FCF) overhanging the valuation that investors are willing to pay. Given the total backlog has more or less tripled this year, all three companies indicated that capex would increase “significantly”, “meaningfully”, or “aggressively” in 2027, but post-earnings capex estimates only moved a modest circa 5% higher. We expect that when companies announce capex in 2027, we may likely see higher growth in capex than revenue growth once again. There is a time lag between capex turning into additional compute capacity due to power availability, permitting, and time to first token. These bottlenecks continue to dictate the pace of revenue growth for the hyperscalers. We anticipate that as we move into the end of 2027, and 2028, the pace of capex growth may slow down, while additional capacity could more rapidly convert backlog into revenues. While operating margins for the three cloud divisions have continued to rise, for the overall businesses they held steady for Alphabet, moved higher for Amazon, but declined for Meta. Revenue acceleration should ultimately feed into higher earnings growth, but will take time.
How AI infrastructure investment could translate into revenue and earnings
Source: Janus Henderson Investors. For illustrative purposes only.
3 Heightened focus on compute efficiency is driving more compute capacity for capex dollars across the ecosystem with vertical integration, silicon and higher utilisation.
Meta’s management noted compute cost efficiency for Muse, its integrated agentic personal AI assistant is expected to improve more than tenfold over the next year through better utilisation of peak and spare capacity deployment of smaller more efficient modders as well as inference optimisation and higher GPU utilisation. Meta also called out AI-driven internal efficiency improvements from software output per engineer is expected to more than double year-over-year. The search for efficiency has benefited the hyperscalers, both in their role of managing and orchestrating model use, but also the increasing demand for their own silicon (chips). Amazon noted that usage of Bedrock, which enables faster build and scale of generative AI applications grew 17x year-over-year, making it the fastest-growing multi-billion-dollar service in AWS history. Meanwhile, Microsoft highlighted broader model adoption on Foundry, as customers increasingly build multi-model, agentic AI solutions.
All of the hyperscalers are increasingly focused on custom silicon to complement NVIDIA GPUs, creating more vertically integrated, heterogeneous AI infrastructure platforms. For Alphabet, and increasingly Amazon, selling proprietary chips to external customers is emerging as a business in its own right. However, business models and margin structures have yet to be fully disclosed, while the build out is weighing on near-term free cash flow as inventories increase alongside the broader capex push.
Encouragingly, Alphabet and Amazon noted that AI compute remains highly utilised with Amazon AI accelerators being installed at a rate of one every 14 seconds. Typically, as quickly as Amazon installs that capacity, customers are consuming it and turning it into revenue.
4 Funding of capex growth – will external financing needs continue?
Funding the build out of AI infrastructure is a key focus given the issuance of debt and equity from these mega caps has increased significantly in 2026. As well as heavy debt issuance, Alphabet and Meta have now tapped equity markets for financing. Alphabet did so earlier this year but still has US$25 billion of at-the-market offering capacity, while Meta raised US$35 billion post results and announced a new C share class to raise a further US$60 billion in late January. We view Meta as being most likely to tap equity shareholders for funding given the extent of its contractual commitments – Meta has more than US$300 billion of long-duration infrastructure, energy, content and purchase obligations, and no existing cloud business with backlog visibility. External financing needs creates an overhang for equity shareholders in the short term, but the level of dilution has been modest and can be rationalised if growth is accelerating as result of the capacity addition. Given the strength of balance sheets, cash generation and multi-year demand visibility, we think funding needs are unlikely to constrain the mega-cap hyperscalers. However, the size of these capital raises combined with a large initial public offering (IPO) in the pipeline could create a challenge for neoclouds and others seeking financing as investors digest overall supply.
5 Varied impact of AI on core businesses
Aggregate ‘Mag 4’ hyperscaler growth slowed from 22% last quarter to 18% this quarter. Although the companies expect this to pick up in Q4, we think investors should be mindful of seasonality and some idiosyncratic issues impacting the latest quarter’s earnings:
Alphabet
Investors were concerned with search revenue growth slowing to 12% and the company’s indication of ongoing slower growth. We think it’s important to differentiate the slowdown this quarter from that of early 2025 when the market was concerned about cannibalisation of search from Chat GPT. Despite record growth in search queries the slowdown was driven by a lag in monetisation – something Alphabet has historically demonstrated it can solve, supported by its industry-leading return on ad spend. We see a parallel with Meta’s Reels monetisation journey. Alphabet is trying to improve its user experience and engagement, ad placement and optimisation, while more competitive advertising auctions could progressively close the monetisation gap as they scale high-quality ads in AI Overviews and AI Mode. Alphabet also alluded to evolving its Gemini app into a more holistic agent while continuing to build out the infrastructure needed for AI agents to interact securely with shopping and payment systems.
Amazon
The ecommerce and advertising business slowed during the quarter, but this was expected given Amazon had previously flagged the move of its Prime Day shopping event from the third quarter to the second quarter, which moved more significant sales and ad revenue into the second quarter. Amazon online fresh grocery business has grown rapidly, outperforming the broader retail business. The company has continued to reduce its costs to serve customers, and has taken some share and grown faster than select peers.
Microsoft
The company’s business covers a broad range of enterprise and consumer markets with growth rates varying across these areas. The cloud business, Azure and M365 (the cloud-based subscription service that combines applications like Word, Excel, Outlook etc, collaboration tools, device management, and security features) remain as key drivers of the business. But other areas have shown slower growth rates, owing to product cycle stages and business maturity. MS Dynamics 365, its enterprise resource planning (ERP) and customer relationship management (CRM) applications, and LinkedIn reported moderating bookings; they remain a small part of the overall business. Importantly, Microsoft remained free cash flow (FCF) positive, a benefit of such a diversified business.
Meta
Advertising is one of the first clear beneficiaries of AI; Meta stood out for its AI spend demonstrating sustained growth in its core business, when most digital platforms decelerated. Management expects to see continued strong growth ( > 20% into 2027). AI continues to have a strong positive impact on engagement and advertising monetisation. Over the quarter, Meta launched Muse, and laid out a monetisation path that could tap into search advertising revenues. We also saw the launch of Meta Enterprise, which sells AI tools, models, and infrastructure directly to organisations and developers. While the advertising core business is strong, there is a lack of certainty and visibility in building out these non-core businesses where they face tough competition.
IMPORTANT INFORMATION
References made to individual securities do not constitute a recommendation to buy, sell or hold any security, investment strategy or market sector, and should not be assumed to be profitable. Janus Henderson Investors, its affiliated advisor, or its employees, may have a position in the securities mentioned.
There is no guarantee that past trends will continue, or forecasts will be realised.
All company information sourced from earnings results releases and call transcripts, unless otherwise specified. Amazon Investor Relations; Q3 2026 earnings results; 29 October 2026; Google Blog; Q3 2026 earnings call28 October 2026; Microsoft Investor Relations; Earnings Release FY27 Q1; 28 October 2026; Meta Q3 2026 earnings call; 21 October 2026.
1 Fortune.com; 26 August 2026.
Advertising auction: An automated process an advertising platform uses to decide which ads may appear, where they are placed, and how much each advertiser may pay.
Agentic AI: AI systems capable of planning and completing multistep tasks, often by interacting with external software, data or services with limited human intervention.
At-the-market (ATM) offering: Allows a listed company to issue and sell new shares gradually into the public market at prevailing market prices, rather than raising a large amount through a single underwritten equity offering.
Backlog (Remaining Performance Obligations or RPO): The value of contracted revenue that a company has not yet recognised because the services have not yet been delivered. A growing backlog can provide greater visibility into future revenues.
Capital expenditure (capex): Money invested to acquire or upgrade fixed assets such as buildings, machinery, equipment, vehicles, technology in order to maintain or improve operations and foster future growth.
Cloud computing: The delivery of computing services, including data storage, processing power and software, over the internet rather than through local computers or servers.
Custom silicon: Computer chips designed for a specific purpose. Hyperscalers develop proprietary AI chips to improve performance, supply availability and cost efficiency.
Free cash flow: Cash generated after operating expenses and capital expenditure. It can be used to repay debt, make acquisitions, repurchase shares or pay dividends.
Frontier model: An advanced artificial intelligence model that represents the leading edge of AI capabilities, typically able to perform a wide range of complex tasks.
Generative AI: Artificial intelligence capable of creating new content, including text, images, audio and software code, based on patterns learned from large datasets.
GPU: A graphics processing unit performs complex mathematical and geometric calculations that are necessary for graphics rendering and are also used in gaming, content creation and machine learning (AI).
Hyperscaler: A large cloud computing company that operates vast data-centre networks and provides computing infrastructure at global scale. Examples include Amazon Web Services (AWS), Microsoft Azure and Google Cloud.
Inference: The process through which a trained AI model generates an output, such as text, an image, software code or a prediction.
Initial public offering (IPO): The process of issuing shares in a private company to the public for the first time.
Magnificent 7 (Mag 7): A group of large US technology-related companies comprising Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla.
Monetisation: The process of generating revenue from a product, service or investment.
Open-weight model: An AI model for which the trained parameters are publicly accessible, although the complete training data and methodology may not be disclosed.
Time to first token: The period between submitting a request to an AI model and receiving the beginning of its response. In an infrastructure context, the phrase can also be used more broadly to describe the time needed to bring new compute resources into productive use.