Q4 2023 Hyperscaler Earnings Review

After over a year of headwinds from customer workload optimization, it appears the hyperscalers are finally reaching a point where future growth will be driven by a more normalized expansion cycle. All three hyperscalers delivered annual growth in Q4 that was either equal to or above the prior quarter. Sequential revenue growth rates for AWS and GCP even accelerated.

With cloud workload utilization returning to being driven by new applications and the expansion of existing ones, software and data infrastructure investors can breathe a small sigh of relief. The strong headwind of workload optimization is yielding to the renewed secular tailwind of cloud migration and digital transformation projects. Lest we get too excited, though, it still isn’t clear what the post-Covid, steady state revenue growth rate will be. It’s likely that the natural growth rate has moderated from the rush to the cloud in 2020, forced by work from home and interruption of physical channels.

Making the recovery picture more complicated, AI initiatives are introducing a new demand tailwind for cloud infrastructure. Just as the obfuscation from workload optimization abates, IT enterprise spend appears to be shifting towards capitalizing on new AI-driven capabilities. Enterprises are delivering tangible workforce efficiencies, productivity improvements and better customer experiences by applying foundational models to their own data troves and business processes.

Whether this AI investment represents incremental budget or is being pulled from existing digital transformation work remains to be seen. If there has been borrowing for AI, that funding may be reset in 2024 as updated annual budgets are rolled out. Additionally, as worker productivity prototypes transition into full scale deployment across the employee base, realized cost savings can be shifted back into budgets for further investment.

The recent earnings results from the hyperscalers provide some hints. The market is certainly interpreting the re-acceleration of revenue growth as a positive signal for software and data infrastructure companies across the board. Stocks for these companies were up significantly the day after Amazon’s earnings (in spite of a strong Jobs report). In the time since I started this post, we have received better than expected earnings reports from CFLT and NET, resulting in outsized jumps in those stocks.

Overall, I think the hyperscaler results portend well for the basket of software and data infrastructure companies going forward. The big hindrance over the past year has been pressure from cost cutting and delayed expansion of utilization. Further, the reset phase for digital natives that spent big during Covid has likely reached its end, with investment picking up again going forward. Additionally, start-up investment (outside of AI) may re-emerge later this year or into 2025, which will bring another tailwind of infrastructure spend to the cloud providers.

In this post, I review the Q4 earnings reports from Microsoft, Google and Amazon with particular focus on their cloud divisions. I also digest commentary on their AI initiatives and speculate what this might mean for budget allocations. As we have already seen, the activity on the hyperscalers has implications for supporting software and data infrastructure providers.

Background

What a difference a year can make. In the first half of 2023, the hyperscaler earnings reports were chock full of commentary on the slowdown in enterprise IT spend and focus on cost cutting through workload optimization. Given the inherent elasticity of cloud workloads (which is really the whole point), it’s not surprising that enterprise IT teams were able to quickly reduce the cost of existing application workloads. In many cases, they had deferred the normal post-launch tuning or over-provisioned capacity in anticipation of maximum utilization during the ZIRP days of Covid.

With higher interest rates and an uncertain demand environment, enterprise IT teams could revisit existing workloads with an eye towards cutting costs. For new workloads, they extended timelines or prioritized only the highest impact efforts. For consumption businesses like the hyperscalers and the independent infrastructure providers, utilization reductions on existing workloads would have an immediate impact on revenue. Combined with postponement of cloud migrations and digital transformation projects, it’s no wonder that sequential quarterly revenue stagnated.

Eventually, all cycles come to an end. As we enter 2024, enterprises appear more optimistic. Their IT teams have realized the majority of benefit from workload optimizations, which have a finite limit. As the headwind of cost cutting abates and new investment picks up, resource consumption at the hyperscalers would naturally increase.

Reflecting this inflection, here is a sampling of quotes from prepared remarks by each hyperscaler in their Q4 reports:

• Microsoft: “That period of massive, I’ll call it, optimization only and no new workloads start, that I think has ended at this point.”
• Alphabet: “And second, I think there are regional variations, but the cost optimizations in many parts are something we have mostly worked through.”
• AWS: “While cost optimization continued to attenuate, larger new deals also accelerated, evidenced by recently inked agreements with Salesforce, BMW, NVIDIA, LG, Hyundai, Merck, MUFG, Axiata, Cathay, BYD, Accor, Amgen, and SAIC. Our customer pipeline remains strong as existing customers are renewing at larger commitments over longer periods and migrations are growing.”

So, at least the picture for software and data infrastructure demand from “traditional” digital transformation and cloud migration projects appears to be back on track. The open question is at what growth rate do we snap back to. This will likely not be the 30% – 50% annual revenue increases enjoyed by the hyperscalers in 2021. Perhaps more like 20% – 30% a year.

That would be the likely scenario, if not for AI. The problem (or opportunity) with AI initiatives is that the ROI is more immediately clear, and spans more than the IT department. As enterprises are collecting data from experiments with new AI services, many are reporting worker productivity gains of 10%, 20% or more. The implication is that they would need fewer of these workers (cost savings) or can process the work faster (more revenue). In either case, the impact on the bottom line is immediate.

Those dollars can be re-invested into new AI initiatives. As most of these AI-driven applications are delivered over the Internet, they consume similar software and data infrastructure resources as the traditional applications. This “spill-over effect” will create incremental demand for those services, even outside of what the hyperscalers capture from the AI model training and inference.

AI Impact and the Value Add

Over the past 6 months, there has been significant debate about the real-world economic impact of AI for enterprises. Early on, consumer-facing generative AI implementations, like ChatGPT, were viewed as clever gimmicks that might disintegrate into a passing fad. The investing implication was that AI infrastructure spend was largely a pull-forward, which would never be sustainable by real enterprise application investment.

Yet, the more time that passes, the more evidence emerges that AI services are creating real impact for enterprises that embrace them. The primary benefits to date revolve around productivity improvements and cost reductions. If companies can deploy co-pilots, expert systems and other types of “assistants” that save their employees time, that translates into real financial impact.

As an example, Microsoft referenced their enterprise customer KPMG, who built an AI assistant for their consultants. This partnership was announced back in July 2023 with the intent for KPMG to leverage Azure AI Services to create new tools for their tax, audit and advisory businesses. KPMG committed to investing multiple billions of dollars in Microsoft cloud and AI services over 5 years. They expected an ROI of $12B in new revenue streams and savings.

The industry-leading collaboration between the two global organizations includes a KPMG multi-billion dollar commitment in Microsoft cloud and AI services over the next 5 years that will help to unlock potential incremental growth opportunity for KPMG of over US $12 billion. The expanded alliance will enhance KPMG client engagements and supercharge the employee experience in a way that is responsible, trustworthy and safe.

The Microsoft cloud and Azure OpenAI Service capabilities will empower the KPMG global workforce of 265,000 to unleash their creativity, provide faster analysis and spend more time on strategic advice. This will enable them to help clients, including more than 2,500 KPMG & Microsoft joint clients, keep pace with the rapidly evolving AI landscape and solve their greatest business challenges while positioning them for success in the future world of work.

KPMG and Microsoft Press release, July 2023

On their earnings call, the Microsoft leadership team provided an update on this relationship. They said that in one case KPMG leveraged the Azure OpenAI service to power an AI assistant for their consultants. This drove a 50% increase in productivity.

Obviously, a 50% increase in productivity would have a significant financial impact on KPMG’s bottom line. As a consultancy, the majority of their production cost is associated with the salaries of knowledge workers. If these consultants can perform their work 50% more efficiently, then KPMG’s effective cost to deliver services just cut in half. They could apply the savings to become more profitable, or double the number of engagements they take on (or some combination).

We are starting to see examples like this repeated over and over again. Enterprises are rolling out co-pilots and other assistants that generate new efficiencies for their employees. These savings (by needing fewer employees to do the same amount of work) are then re-invested into other AI projects to create new efficiencies. This cycle will likely repeat for a few years.

Because the benefit to enterprises is real, they won’t mind investing significant budget into these initiatives. Similar to KPMG, these types of engagements could involve budgets in the billions of dollars. Multiplied across the Global 2000, this makes it easy to appreciate how we may just be seeing the tip of the iceberg for AI spend.

More broadly, the Microsoft team cited their own research that elucidates the role AI will play in transforming work. They claim to have facilitated outcomes with as much as 70% improvement in productivity, using generative AI for specific work tasks. Early Copilot for Microsoft 365 users were 29% faster in the series of tasks, like searching, writing, and summarizing. 

The thing that you brought up is a little bit of a continuation to how Amy talked about, right, so you are going to start seeing people think of these tools as productivity enhancers, right? I mean, if I look at it, our ARPUs have been great, but they’re pretty low. Even though we’ve had a lot of success, it’s not like we had a high-priced ARPU company. I think what you’re going to start finding is, whether it’s Sales Copilot or Service Copilot or GitHub Copilot or Security Copilot, they are going to fundamentally capture some of the value they drive in terms of the productivity of the OpEx, right? So it’s like 2 points, 3 points of OpEx leverage would be goal is on software spend. I think that’s a pretty straightforward value equation.

And so that’s the first time, I mean, this is not something we’ve been able to make the case for before whereas now I think we have that case. Then even the horizontal copilot is what Amy was talking about, which is at the Office 365 or Microsoft 365 level, even there, you can make the same argument whatever ARPU we may even have with E5, now, you can see incrementally as a percentage of the OpEx, how much would you pay for a copilot to give you more time savings for example. And so yes, I think all up, I do see this as a new vector for us in what I’ll call the next phase of knowledge work and frontline work, even in their productivity and how we participate. And I think GitHub Copilot, I never thought of the tools business as fundamentally participating in the operating expenses of a company’s spend on, let’s say, development activity and now you’re seeing that transition.

It is just not tools. It’s about productivity of your dev team.

Microsoft Q2 FY2024 Earnings Call, january 2024

Given these productivity gains, Microsoft leadership points out that they feel they can make the case to customers that spend on these AI productivity tools is justified by their savings in operation expense. They weren’t able to make this direct correlation previously with software tools. For GitHub Copilot, as an example, the cost would fold directly into a customer’s staffing budget for their development team. Previously, it fell into the allocation for tooling separate from salaries. Now, the productivity improvements allow the cost of Copilot tooling to be pulled from the actual headcount line item.

That represents a subtle, but important distinction. Headcount has often been the largest expense in a knowledge worker department budget. Usually, the leadership team determines the number of staff needed in each role and then allocates a couple percentage of cost for workforce tooling. Now, Copilot can be lumped into the headcount line item itself, with the argument being that the productivity improvements generated by the copilots reduces the number of headcount needed.

And this is just for knowledge work. The next and I think bigger application of AI services will be in the physical world. The combination of AI-driven software services, plus devices (robots, cars, IoT, cameras, equipment, etc.) would likely consume an order of magnitude more compute, data and software infrastructure resources.

Just look at this collaboration between the NFL and AWS to improve player safety (lower costs from injuries). The amount of data from real-world interactions is enormous. The inputs are from multiple sources that span several modes – video, RFID, weather, physical equipment, sound, etc.

Digital Athlete is a platform that leverages AI and machine learning (ML) to predict from plays and body positions which players are at the highest risk of injury. The platform draws data from the players’ RFID tags, 38 5K optical tracking cameras placed around the field capturing 60 frames per second, as well as other data such as weather, equipment, and play type to build a complete view of players’ experiences. One of those data sources is the Next Generation Stats System (NGS), which captures real-time location, speed, and acceleration data for every player.

During each week of games, Digital Athlete captures and processes 6.8 million video frames and documents about 100 million locations and positions of players on the field. During practices, it processes around 15,000 miles of player tracking data per week — translating to more than 500 million data points.

“We’re running millions of simulations on in-game scenarios to tell teams which players are at the highest risk of potential injury, and they use that information to develop individualized injury prevention courses,” says Julie Souza, global head of sports at AWS.

The amount of compute and storage to handle all this data would be massive. And it’s just one use case within one sports league. As these types of examples, with real-world benefits, are socialized, I think we will witness an explosion of additional applications. Each one will be bring incremental and unique demand for software, data and AI processing resources. They will also be paid for out of tangible cost savings, revenue opportunities or better customer experiences. This direct correlation of ROI will make further investment easy for companies and organizations to greenlight, versus the sometimes amorphous benefits of digital transformation projects.

Hyperscaler Results

With that background, let’s look at the earnings results from the hyperscalers and signals for the path forward. Comparing their performance isn’t exact, as the three hyperscalers report their cloud division results differently. Amazon provides the clearest indicator, reporting the total revenue from just AWS. Microsoft reports the annual growth rate for Azure, but not the amount. Finally, Google reveals the total revenue for Google Cloud, but GCP is mixed in with other cloud-based SaaS offerings, like Google Suite.

Coming into the Q4 earnings reports, all three stocks had appreciated nicely over the course of 2023. This isn’t surprising, as it is becoming clear that late 2022 was a bottom for most software and cloud infrastructure stocks.

Looking at the stock charts for the hyperscalers going back 3 years, we can make a few observations. While all 3 stocks have appreciated nicely over the past year, only MSFT has passed its historical ATH price. As of this writing, MSFT is about 20% above its prior peak in November 2021 and almost 86% higher than its opening price of 2021.

AMZN and GOOG have not fared as well over the past 3 years. AMZN is about 6% above its price at the start of 2021, but is still slightly below the peak in November 2021. GOOG has done better for a 3 year period, up 66%, but is also just below its November 2021 peak.

MSFT stock really pulled away over the last 4 months, when it became clear that their AI products would meaningfully drive new revenue growth for the company. Of the three, Microsoft moved the fastest to capitalize on their AI capabilities to generate new revenue streams. Co-pilots for Office 365 and GitHub already have measurable revenue streams that are separate from AI services in Azure. Most of the uplift for Amazon and Google is associated with their cloud offerings, which include AI processing services, but not stand-alone end products.

After watching revenue growth rates declining for most of 2022 and 2023, the Q4 2023 earnings results for all three hyperscalers showed evidence of a bottoming and slight acceleration. For both AWS and GCP, sequential growth rates in Q4 ticked up nicely. Azure growth was stable after recovering a bit in Q3.

This represented a welcome change after several quarters of slower growth, driven by pressure on IT budgets and cost optimization from enterprises. Leadership commentary was consistently upbeat for the quarter ending in December 2023. While some macro pressure still exists, hyperscaler leadership seemed confident that the overall demand situation is improving going forward, versus continuing to deteriorate, as had been the message in the past.

For all three hyperscalers, AI was a common theme. All three spoke to the enormous opportunity to leverage AI to drive a whole new suite of services. The AWS team talked about “tens of billions in potential revenue” in the future. Leadership cited many examples of enterprise customers making use of AI services to launch new capabilities for their employees and end consumers. AI is quickly moving past being visionary towards real-world applications that generate tangible ROI for enterprises.

Personally, I think we are just scratching the surface and many more use cases for AI technology exist going forward. While applications like ChatGPT may exhaust consumer demand soon, the real benefit comes for enterprises in applying LLMs to make their workforce more productive and efficient, reducing costs. That generates real enterprise budget savings. Those dollars can be invested into more AI services, which in turn drives incremental benefit.

The key point is that less spending on employee salaries can offset the expense of new AI services. Enterprises won’t need to create new budget to fund visionary AI products. Rather, they can pay for them from existing workforce expense.

Microsoft

Microsoft reported Q2 FY2024 earnings (quarter ended December 31, 2023) on January 30th. Overall, the company delivered $62.0B in revenue, which was up 18% y/y. Analysts were looking for $61.1B. Similarly, EPS beat estimates by $0.16, realizing $2.93 EPS for the quarter. The stock finished down by 2.7% the next day, which turned out to be a bad day for the market overall. Since then, MSFT stock is back above the price before earnings.

For Azure, in the prior quarter of Q1 (ended September 2023), the company delivered growth of 29% overall and 28% in constant currency. This result beat their estimate for 25%-26% growth in constant currency coming into the quarter by a healthy 2-3% margin. This also represented a slight acceleration in annual growth as compared to the 27% constant currency growth rate in Q4 (ended June 2023).

Microsoft provides an additional and very interesting data point relative to AI with their Azure results, which is the number of points of the overall growth rate that is represented by their AI services. Their definition of AI services is a little broad, including those offerings that are direct extensions of OpenAI products, as well as Azure AI services. In Q4 (ended June 2023), this AI contribution was 1% (or 1 point of the 27% growth in Q4 revenue). For Q1, this value tripled to 3%, meaning that 3 points of the 28% cc revenue growth came from AI services.

For Q2 (quarter ended December 31st and just reported), Microsoft leadership had projected Azure revenue growth of 26%-27% in constant currency. Microsoft actually delivered Azure revenue growth of 30% overall and 28% in cc, beating their estimate by 1-2%. Further, the contribution from AI services jumped to 6 points. That is a huge increase from the 3% in the prior quarter.

The interesting aspect of this is that a 6 point attribution to AI services implies that “regular, non-AI” services in Azure grew by just 22% in cc. This would represent a decrease from the adjusted 25% in Q1 (28% overall in cc, minus the AI contribution of 3%). That implies a deceleration in growth from 25% (adjusted) to 22% in the most recent quarter for non-AI Azure infrastructure services.

The implication is that enterprises are increasing their spend on AI services, potentially at the expense of non-AI services. On the earnings call, analysts probed this point. They asked for a definition of AI services. Microsoft leadership pointed them to Azure AI Services, which allows customers to infuse generative AI into their applications to create new product offerings. A big part is Azure OpenAI Service, which facilitates access to LLM’s from OpenAI through an API or SDK. This programmatic interface makes it very easy to add AI capabilities to an existing application, much like API’s to third party services that enable payments and SMS messages.

Azure OpenAI and then OpenAIs on APIs on top of Azure would be the sort of the major drivers, but there is a lot of the small batch training that goes on, whether it’s a graph or fine-tuning, and then lot of people who are starting to use models as a service with all the other new models, but it’s predominantly Azure OpenAI today.

Microsoft Q2 FY2024 earnings Call, January 2024

Looking forward, the Microsoft CFO expects Azure revenue growth in constant currency to “remain stable” to the Q2 results. This implies a revenue growth rate of 28% in constant currency for Q3 (ending March 2024). Given that they just beat their prior Q2 estimate by 1-2% of growth, we could see an actual revenue growth rate in Q3 of 29%-30%, which would represent a slight acceleration.

On the earnings call, Microsoft leadership shared a few other updates, specific to AI services and their surge in customer demand.

• Azure AI now has 53,000 customers. Of these, over 1/3 are new to Azure in the last 12 months. That is an important metric, implying that getting access to Azure AI Services is driving customers to Azure itself. This metric was 11,000 just two quarters ago (Q4 FY2023, ended June 2023).
• Over half of the Fortune 500 use Azure OpenAI, including Ally Financial, Coca-Cola and Rockwell Automation. As an example, at CES in January, Walmart shared how it’s using Azure OpenAI Service, along with its own proprietary data and models, to streamline how more than 50,000 associates work and transform how its millions of customers shop. 
• GitHub revenue growth accelerated to over 40% year/year, driven by adoption of GitHub Copilot. Copilot has 1.3M paid subscribers, up 30% q/q, spanning more than 50,000 organizations. Based on published research, Microsoft claims that developers can see a 55% increase in coding speed, at least in one study. The key contribution is that GitHub Copilot handles the more mundane coding tasks (utility functions, interface, state management, bootstrap code), allowing the developer to focus on the code unique to their business.
• More than 230k organizations have used AI capabilities in Power Platform, up over 80% q/q. With Copilot Studio, organizations can customize existing copilots for MS 365 or create brand new ones on their data. In just weeks, for example, both PayPal and Tata Digital built copilots to answer common employee queries, increasing productivity and reducing support costs. 
• The Sales Copilot has helped more than 30k organizations, including Lumen Technologies and Schneider Electric. They use this service enrich their customer interactions using data from Dynamics 365 or Salesforce. Customer service employees at companies like Northern Trust can resolve client queries faster. The copilot includes out-of-the-box integrations to apps like Salesforce, ServiceNow and Zendesk.

Outside of AI, Microsoft leadership noted demand for their data infrastructure services, packaged as the Microsoft Intelligent Data Platform. This data platform brings together operational databases, sophisticated analytics and governance within a data lakehouse architecture. These capabilities can be enhanced with AI services and machine learning to power applications.

One of the operational database offerings is Cosmos DB, which is a multi-modal data storage engine. It can handle document-oriented, relational, wide column and key-value workloads, among others. CosmosDB can serve as the metadata store for AI services and also offers a vector database with vector search capability. This allows users to store vector embeddings and query them, which is often leveraged to support retrieval-augmented generation (RAG).

Microsoft leadership highlighted the popularity of Cosmos DB among customers. Data transactions for Cosmos DB increased by 42% y/y in the quarter. They mentioned customer workloads from AXA, Kohl’s, Mitsubishi and TomTom. KPMG used CosmosDB to built an AI assistant for its consultants. They credit the AI assistant with driving a 50% increase in productivity.

Looking outside of the hyperscalers, CosmosDB shares a number of capabilities with MongoDB (MDB). The MongoDB team recently added support for a vector database and search capability to the Atlas cloud offering. This strong demand for CosmosDB on Azure implies that enterprises would likely be looking for similar workload growth on their MongoDB clusters. This may provide a positive demand signal for MongoDB, as we await their earnings report in early March.

Alphabet

Alphabet obfuscates the performance of their Cloud Platform (GCP) somewhat, by bundling it with the line item called Google Cloud. This includes their workforce application business (Google Workspace – formally G Suite) in addition to their hyperscaler services. For Q3 2023 (ended September 2023), Google Cloud revenue growth dropped to 22.5% year/year and 4.7% sequentially. This was after two quarters of 28% annual growth and a nice sequential acceleration in Q2 of nearly 8%.

In the Q4 2023 report from January 2024 (quarter ended December 2023), revenue ticked back up to 25.7% annually and a strong 9.3% sequential increase. The sequential growth rate matched the highest rate going all the way back to Q4 2021. While we don’t know the exact contribution from Google Cloud Platform, management has commented in the past that the relative growth rate of GCP exceeded that of Google Cloud overall. They didn’t provide that comparison on this quarter’s earnings call.

Like Microsoft, Google has been investing heavily in the creation of new AI services for customers to consume. For generative AI use cases, these capabilities are made available through the Vertex AI platform. This exposes APIs for Google’s Gemini multimodal model, capable of understanding virtually any input, combining different types of information and generating almost any output. These span text, images, video and code.

For developers, Google offers Duet AI, which is their AI-powered development assistant. This is similar to Microsoft’s GitHub Copilot. Through a natural language chat interface, developers can interact with Duet AI to get answers to coding questions, generate starter code and receive guidance on GCP configuration and administration.

Similar to Microsoft, the Google leadership team highlighted some AI-related wins during Q4:

• In Q4, landed or expanded relationships with Hugging Face, McDonald’s, Motorola Mobility and Verizon.
• Hightlighted AI specialist companies that are leveraging Google’s AI Hypercomputer, which provides a supercomputing architecture that is appropriate for model training. It combines TPUs and GPUs, AI, software, and multi-slice and multi-host technology to provide performance and cost advantages for training and serving models. AI centric start-ups like Anthropic, Character.ai, Essential AI and Mistral AI are building and serving models on the AI Hypercomputer.
• Vertex AI has experienced strong adoption. API requests increased nearly 6x from H1 to H2 in 2023.
• Duet AI is helping employees benefit from improved productivity and creativity at thousands of paying customers around the world, including Singapore Post, Uber and Woolworths. In Google Cloud Platform, Duet AI assists software developers and cybersecurity analysts.

Google doesn’t break out the contribution of AI services to overall Cloud revenue growth in the way that Microsoft does. I suspect that these services contribute less to overall Google Cloud revenue at this point than Microsoft’s more mature offering in OpenAI. However, Google will likely catch up quickly. They were, in fact, the earliest of the hyperscalers to recognize the potential for AI and created Google Deepmind, which has been a source of top talent for the newer breed of AI companies.

Amazon

Similar to Google, Amazon Web Services (AWS) delivered another quarter of sequential revenue growth acceleration in their Q4 results. Total revenue was $24.2B, up 13.2% y/y and 4.9% sequentially. This is an improvement from Q3’s results of 12.3% annual and 4.2% sequential growth. While 13% annual growth isn’t very exciting, if we annualize the 4.9% sequential growth, it implies a run rate of 21%. We have to go all the way back to early 2022 to find a sequential growth rate higher than this.

Like Google, Amazon didn’t break out the contribution from AI to the AWS growth rate. In the prepared remarks, Amazon’s CEO commented “AWS’s continued long-term focus on customers and feature delivery, coupled with new GenAI capabilities like Bedrock, Q and Trainium have resonated with customers and are starting to be reflected in our overall results.”

I think both Google and Amazon are seeing a smaller contribution from AI to their overall cloud infrastructure growth than Microsoft because it required more time for them to spin up their dedicated AI services. Microsoft had the benefit of OpenAI and GitHub already operating at scale. They simply layered an API over OpenAI and rolled out copilot offerings for developers, sales teams and content creators. The monetization path was easier to ramp. Amazon and Google have a similar opportunity – it’s just taking more time to realize.

With that said, the Amazon team shared a huge number of examples of customers using AI to power real-world use cases in their quarterly report. I will include them all below, as I think this provides important evidence for the thesis that enterprises are finding real-world applications for AI. This early success will reinforce their commitment to AI use and lead to more investment.

• Amgen expanded its work with AWS to create generative AI–based solutions, using Amazon SageMaker to help discover, develop and accelerate the manufacturing of medicines for patients suffering from serious illnesses. Amgen will also use AWS’s global infrastructure and advanced services to power a new digital data and analytics platform.
• Salesforce significantly expanded its global partnership with AWS to make it easier for customers to bring AWS data into Salesforce products and deepens data and AI integrations between AWS’s and Salesforce’s products. Salesforce will support Amazon Bedrock as part of Salesforce’s open model ecosystem strategy and will expand its use of AWS, including compute, storage, data and AI technologies.
• Hospitality company Accor S.A. expanded its relationship with AWS to launch a generative AI Travel Assistant using Amazon Bedrock and Amazon SageMaker. The Travel Assistant aims to reinvent the guest experience by enhancing travel planning and booking, while reducing call volumes at Accor’s contact center.
• Financial services provider Mitsubishi UFJ Financial Group (MUFG) signed a multi-year global agreement with AWS as its preferred cloud provider to innovate personalized financial services and use Amazon Bedrock to experiment with more than 100 potential generative AI use cases. Using Amazon QuickSight, MUFG lowered operational costs by approximately 70% compared to legacy technology.
• Merck will work with AWS and Accenture to move a substantial portion of Merck’s IT infrastructure to AWS as its preferred cloud services provider. Merck is using services, including high-performance computing on AWS and AWS HealthOmics to speed drug discovery and accelerate therapeutic discovery and Amazon SageMaker for generative AI and machine learning (ML) to improve product manufacturing and availability.
• Ecommerce retailer The Very Group expanded its collaboration with AWS, which included the launch of a Gen AI Innovation Lab. The lab will trial new generative AI retail solutions built on Amazon Bedrock to deliver interactive and personalized digital shopping experiences.
• Automaker SAIC MOTOR selected AWS as its strategic cloud provider for its i-SMART connected vehicle platform. SAIC MOTOR is using AWS’s global infrastructure services to enhance the driving experience and build an elastic and agile connected vehicle architecture that can scale globally. SAIC MOTOR will use Amazon Bedrock to personalize the in-vehicle experience and to automatically diagnose vehicle issues.
• Multinational telecom Axiata Group Berhad selected AWS as its primary cloud provider to accelerate its digital transformation across its operating companies by using AWS’s generative AI and ML capabilities, including Amazon Bedrock and Amazon SageMaker. By the end of 2024, Axiata will migrate more than 650 services spanning customer service, enterprise resource planning, and human resources, along with 80 ML applications.
• Energy company HD Hyundai Oilbank will transform its refining and petrochemical business into a platform for eco-friendly energy by going all-in on AWS, forecasting to reduce IT cost by 20%. The company will launch an electric vehicle charging business on AWS, optimize the refinery process using ML and generative AI with Amazon Bedrock, deliver data-driven insights such as energy-demand forecasting, and improve operations like inventory management.
• LG AI Research, the research hub of LG Group, launched an AI image-to-text captioning solution on AWS at 66% lower cost and with 83% faster data processing than on-premises infrastructure. The new captioning solution uses a multimodal foundation model (FM) built using Amazon Elastic Compute Cloud (Amazon EC2) and Amazon SageMaker that creates more accurate content than previous solutions.
• Travel lifestyle brand Cathay selected AWS as its cloud provider and migrated more than 50% of its legacy and critical enterprise workloads to AWS, reducing IT infrastructure costs by 40%. The new Cathay Machine Learning Innovation Hub will help automate operations, train employees on ML and enhance air travel experiences. Cathay’s cargo business is using Amazon SageMaker, Amazon OpenSearch Service, and Amazon Redshift to build a revenue management model to increase profitability and customer loyalty.

These examples highlight the emerging use cases for enterprises to leverage AI to generate cost savings and better customer experiences. They also demonstrate how a desire to harness AI is driving these companies to upgrade their software and data infrastructure. In most cases, that leads to a cloud migration. We are witnessing a “spill-over effect”, where AI brings new urgency to planned digital transforming and infrastructure modernization efforts. This benefits the hyperscalers and independent software service providers.

Additionally, Amazon announced a number of applications of AI to improve their internal operations and product offerings across business lines outside of AWS. This included an AI tool for fashion shopping that helps customers find the best fit. They also launched Rufus, a generative AI-powered conversational shopping assistant. For advertising clients, they developed a generative AI solution to help brands produce lifestyle imagery for their ads.

One of the cooler applications introduced was the Automated Vehicle Inspection (AVI) system. This AI-powered technology performs a full-vehicle inspection on Amazon delivery vans. It flags issues like tire deformities, undercarriage wear and bent body pieces that might presage an on-road problem.

What is impressive about all this detail is the broad scope of applications for AI. Amazon’s customers are using AI services at scale and Amazon is internalizing the capabilities to enhance their own operations. This provides a feedback loop to further improve the effectiveness of AI services offered through AWS. Additionally, Amazon is leveraging these AI services to generate their own efficiencies, saving costs and time, as well as proactively addressing potential problems (like when a delivery vehicle might break down).

I think these examples from Amazon really solidify the case that AI is impacting real-world productivity and business results for enterprise customers. There are just too many examples to write this off as a non-productive experiment. Companies one after the other are reporting real use cases with tangible savings from applying foundation models towards their internal operations and customer offerings. These are lowering costs and improving service.

As these benefits are directly tied to the bottom line, we can comfortably assume that the investment will continue. Additionally, I think we at the tip of the iceberg in terms of these types of applications. Normally, small wins of this type within a large enterprise are publicized and then quickly duplicated across many business lines.

Department heads scramble to demonstrate that they too can apply AI to cut costs and improve productivity, leading to more use cases. In my view, this makes it unlikely that these AI investments represent a one-time, pull-forward in utilization that will quickly plateau. I think it will require several years for enterprises to fully take advantage of these capabilities. And that is just for knowledge work, versus applications for autonomy and robotics in the physical world.

Take-Aways and Investment Plan

The Q4 results from the hyperscalers provided investors with a few useful signals as we enter 2024. These generally mark an inflection from the IT demand doldrums of 2022 and much of 2023. While we won’t return to the halcyon days of 2021, where low interest rates, Covid restrictions and start-up investment drove record growth rates, investors can look forward to a demand environment influenced more by tailwinds than headwinds in 2024.

First, all three hyperscalers demonstrated a trough and even slight re-acceleration of their revenue growth rates. Management uniformly pointed to the end of heavy spend optimization efforts by their enterprise customers and a renewed focus on new application use cases. The headwind of continuous cuts in the resource consumption of existing workloads is finally abating. It is giving way to the resumption of the secular tailwind from planned digital transformation and cloud migration projects.

By itself, the end of heavy optimization cycles would help reinvigorate revenue growth for the hyperscalers. However, the rapid emergence of generative AI and useful foundation models has introduced another tailwind. Enterprises are scrambling to harness AI capabilities to create operational efficiencies, boost employee productivity and deliver better customer service. These AI offerings have measurable ROI, allowing enterprises to justify the budget allocation and increase investment to drive new use cases.

As we see with Azure, the contribution of AI services to their overall revenue growth is scaling quickly, doubling from 3 points to 6 points of total growth in the last quarter. AWS and GCP are spinning up their own AI offerings rapidly with management providing many examples of enterprise wins using these new offerings.

The deployment of new AI-powered applications and digital experiences will drive a “spill over” effect for other aspects of software and data infrastructure providers. As these AI use cases often take the form of a software application delivered over the Internet, they will consume the same supporting services as the traditional web or mobile application, needing operational data, delivery, app security and monitoring. This should increase demand for providers of these supporting services, in addition to the core AI capabilities themselves.

As I discussed in a prior post, I think data infrastructure providers will be one of the primary beneficiaries. Confluent (CFLT) released a better than expected Q4 report and the stock is up about 35% so far in 2024. The earnings report eluded to the fact that the demand for AI services is causing enterprises to reconsider their data infrastructure, including how they distribute data internally. We will hear from MongoDB, Snowflake and other data providers in the next month. A similar carry-over effect will likely emerge.

As enterprises pursue more initiatives to incorporate AI into offerings for their customers, employees and suppliers, it will drive more hyperscaler resource consumption, both of AI services and the broader rails of software infrastructure. These efforts will go beyond core AI model generation, cascading out towards inference and all standard cloud application support services (data, storage, security, monitoring, etc.).

As Microsoft’s CEO once quipped, AI services will also “spin the Azure meter.” After two years of IT spending headwinds, software and data infrastructure companies may finally shift back towards enjoying tailwinds. Investors can relax a little relative to the overall demand environment, but will still need to be selective about which providers may enjoy the most benefit as we move through 2024. Identifying the real winners in the AI era may be a little more nuanced than during the prior Covid investment cycle.

NOTE: This article does not represent investment advice and is solely the author’s opinion for managing his own investment portfolio. Readers are expected to perform their own due diligence before making investment decisions. Please see the Disclaimer for more detail.