One of the main features of the third quarter report is that tech giants continue to "burn money," unwaveringly increasing AI spending, with total capital expenditures of Microsoft, Alphabet, and Meta growing by 48% year-on-year. However, the returns from these massive investments are "mixed," and the market may need to wait longer. Additionally, the three major trends of "hiring is not as good as buying cards," training shifting to inference, and Microsoft marginalizing OpenAI are becoming increasingly clear
The financial reports of major U.S. tech giants for the third quarter have been released, showcasing the latest developments in AI.
What are the concerns regarding AI investments, such as capital expenditures and the returns behind the massive investments? What answers do the financial reports provide? What trends in AI development do the reports reveal?
Tech Giants Continue to "Burn Cash," Unwaveringly Increasing AI Spending
In the third quarter, major tech giants continued to follow the narrative of "increasing spending" on AI, with Microsoft, Google, and Meta reporting a quarter-on-quarter capital expenditure increase of 3% and a year-on-year increase of 48%, committing to further increase spending on servers and other AI-related costs in the coming quarters.
Specifically:
Google: The capital expenditure for the third quarter was $13 billion (a quarter-on-quarter decrease of 1%, a year-on-year increase of 62%), exceeding market consensus by 3%, driven mainly by replenishment of general servers and strong spending on AI servers. The company expects capital expenditure in the fourth quarter of 2024 to be similar to that of the third quarter, indicating that total spending in 2024 will reach approximately $51 billion, a year-on-year increase of about 60%. Management expects significant growth in capital expenditure in 2025, primarily driven by investments in AI infrastructure.
Microsoft: Capital expenditure in the third quarter reached $15 billion (a quarter-on-quarter increase of 8%, a year-on-year increase of 50%), exceeding market expectations by 3%, mainly attributed to the acceleration of the H200 AI server project and increased infrastructure investments. Due to investments in Azure and AI infrastructure, capital expenditure in the next quarter is expected to increase by 32% year-on-year. Capital expenditure for the fiscal year 2024 is expected to grow by over 55%, primarily driven by aggressive AI investments and replenishment of general servers.
Meta: Capital expenditure in the third quarter was $8.3 billion (a quarter-on-quarter increase of 1%, a year-on-year increase of 26%), lower than Bloomberg's expectation of $11 billion due to timing differences in server deliveries. However, the company narrowed its guidance for capital expenditure in 2024 from the previous range of $37 billion to $40 billion (midpoint year-on-year growth of 41%) to $38 billion to $40 billion (midpoint year-on-year growth of 43%), primarily driven by investments in AI infrastructure. Meta's management reiterated plans for significant capital expenditure increases in AI infrastructure next year.
Amazon: In the third quarter, the company spent $22.6 billion on real estate and equipment, an 81% increase compared to the same period last year, bringing total capital expenditure to $51.9 billion year-to-date. The company expects capital expenditure in 2024 to be $75 billion, an increase of about 55% from $48.4 billion in 2023, driven by the AWS cloud division, with capital expenditure potentially higher in 2025.
In response, JP Morgan analysts believe:
From the financial reports of Google, Microsoft, and Meta, there are optimistic comments on capital expenditure, solid core business performance, and a positive outlook for AI business, which may enhance investor confidence in the sustainability of AI spending and alleviate market concerns about the impact of AI servers. It is expected that the optimistic spending guidance from Google, Microsoft, and Meta will boost downstream AI server sentiment in the short term
The Mixed Returns Behind Huge Investments
Regarding the capital returns of AI, which the market has been closely watching, tech giants have seen some positive growth signs in conversational AI services, but this growth seems to have fallen short of expectations and may be offset by a slowdown in spending in other areas, limiting the overall growth of tech companies. Shareholders will have to wait longer to know the specific return answers.
Microsoft reported a slight decline in the growth of its cloud computing division, Azure, in the third quarter, with a larger decline expected in the fourth quarter. However, Microsoft stated that the slowdown in Azure's growth is not related to weakened customer demand, especially in this quarter, but rather due to the company's ongoing capacity constraints. Analysts expect that starting in the second half of 2025, as more AI servers come online, Azure's growth will accelerate, indicating that Microsoft’s AI server supply chain revenue will speed up in the coming quarters.
At the same time, on the positive side, there is a growing demand for AI products. Microsoft stated that revenue from all its AI products (such as OpenAI models sold through Azure, GitHub Copilot, and Office Copilot subscriptions) is expected to exceed $2.5 billion per quarter in the next two months, while Microsoft’s total revenue in the third quarter was $66 billion.
Additionally, Google reported that growth in its cloud computing division is accelerating, with Google Cloud's revenue growing 35% year-on-year in the third quarter, up from 29% growth in the previous quarter, primarily due to strong core/AI GCP demand and expansion in Google Workspace's average revenue. Improved advertising demand across industries also indicates a good momentum for general server spending.
However, this is partly attributed to non-AI-related factors. Google Cloud began executing contracts signed by some of its largest customers earlier this year, who committed to spending a certain amount annually to rent servers. Sources involved in these deals indicated that, in some cases, customers would pay a lump sum for unused cloud services they had previously committed to purchasing, marking a departure from past practices where Google Cloud allowed customers to carry over their spending commitments to the next year.
"Hiring is Not as Good as Buying Cards," Training Shifts to Inference, Microsoft Marginalizes OpenAI
This financial report also revealed three major trends: first, "hiring is not as good as buying cards"; second, the shift from training to inference; and third, Microsoft is "diluting" the importance of OpenAI.
First, although tech giants are significantly increasing their investments in artificial intelligence, they are also trying to control expenses as much as possible, which includes layoffs and improving employee efficiency. Specifically, Amazon has laid off tens of thousands of employees since 2022. Controlling expenses has slowed the growth of total operating costs from 17.6% in Q3 2022 to 7.3% in Q3 2024.
Second, the importance of inference surpasses that of training. Microsoft has officially decided to prioritize inference work over AI training, expecting annual revenue from its AI business to reach $10 billion next quarter. Microsoft CEO Satya Nadella stated that this growth is driven by inference rather than training, and Nadella also mentioned that Microsoft is not selling GPUs to others for training, as there is significant demand for inference that supports Copilots and other AI servicesAnalysis indicates that the key to Microsoft's strategic modification lies in its decision to refuse customers eager to rent GPUs for training new AI models. Instead, Microsoft's goal is to utilize these resources for inference, as the demand in this area is booming with more enterprises integrating AI components into their operations. This shift demonstrates a keen understanding of market dynamics; inference can provide more direct and stable revenue compared to model training.
Finally, it is noteworthy that Microsoft is downplaying the importance of OpenAI. Some analysts suggest that with Microsoft's announcement of the introduction of Claude 3.5 Sonnet and Gemini 1.5 pro, OpenAI is no longer the only option. This indicates that Microsoft aims to build more orchestration layers on top of the models, capturing greater value from the intermediate orchestration layer, similar to the past cloud computing era, such as the recently released Agent creation tool. The value of the underlying model layer, which is OpenAI's, will be downgraded