AI’s Price Tag: Tech Giants Shift Costs to Consumers
After a year of aggressively integrating generative AI into their core products, tech giants like Microsoft and Google are now grappling with a harsh reality: making their AI investments pay off. The initial enthusiasm is giving way to a more pragmatic approach, characterized by price hikes, the introduction of advertising, and other strategies that effectively shift the financial burden onto consumers.
Is this a sign that the AI boom is losing steam? The situation is more complex than a simple reversal. While the commitment to AI remains unwavering, companies are struggling to monetize the technology. Their response? To find subtle, often less obvious ways, to make consumers foot the bill.
Shifting the Costs
Last week, Microsoft quietly scaled back its planned data center expansions. Concurrently, the company increased subscription prices for its flagship 365 software by up to 45% and introduced an ad-supported version of some products. Additionally, Microsoft CEO Satya Nadella has recently downplayed the tangible value generated by AI so far.
These actions might seem counterintuitive in the face of the current AI frenzy, especially when contrasted with the splashy announcements from companies like OpenAI regarding their $500 billion Stargate data center project. However, a closer look reveals that these moves do not indicate a retreat from AI. Instead, Microsoft is adapting its strategy to make AI profitable by subtly shifting costs onto consumers.
The High Cost of Generative AI
Generative AI is, undeniably, expensive. OpenAI, a market leader with approximately 400 million active monthly users, is incurring massive losses. Last year, the company generated US$3.7 billion in revenue but spent nearly US$9 billion, leading to a net loss of about US$5 billion.
OpenAI CEO Sam Altman has stated that the company is losing money on its US$200 per month ChatGPT Pro subscriptions. Microsoft, OpenAI’s largest investor and cloud computing provider, also indirectly bears the brunt of these expenditures.
What factors contribute to the high cost? Beyond human labor, there are two critical expenses associated with AI models: training (building the model) and inference (using the model). While training is a substantial upfront investment often involving significant resources, inference costs escalate with the user base. Furthermore, the larger and more sophisticated the AI model, the more expensive it becomes to operate.
Searching for Cheaper Workarounds
A single query on OpenAI’s most advanced models can cost up to US$1,000 in compute power alone. In January, CEO Sam Altman admitted the company’s $200 per month subscription is not profitable. This suggests a significant financial shortfall, not just related to the free models, but also with their premium subscription services.
Both training and inference rely heavily on data centers, which pose significant cost factors. The specialized chips required for operation are exceedingly expensive and the costs for electricity, cooling, and hardware depreciation are substantial. Tech companies face a growing problem in recouping these costs associated with running data centers to power generative AI products.
To date, innovation in AI has largely been a matter of scale. OpenAI describes its newest model as “a giant, expensive model”. There are signals that this relentless pursuit of scale may not be necessary. DeepSeek, a Chinese company, created comparable models for just a fraction of the traditional training expense. Researchers at the Allen Institute for AI (Ai2) and Stanford University claim to have trained a model for as little as US$50.
Essentially, big tech AI might not be profitable due to the expense of building and maintaining the necessary data centers.
Microsoft’s Strategy
Having invested billions into generative AI, Microsoft is focusing on a viable business model to turn the technology profitable. Over the past year, the tech giant has integrated its Copilot generative AI chatbot into its products. Today, it is impossible to purchase a Microsoft 365 subscription without Copilot. Consequently, subscribers are experiencing significant price increases.
The cost of running generative AI models in data centers is substantial. Microsoft is attempting to move more of the processing burden onto users’ devices – where the user is responsible for device-specific hardware and operational costs. Microsoft highlights that Copilot’s key will enable people to participate in the AI transformation.
As evidence of this strategy, Microsoft has integrated a dedicated Copilot key onto their devices. Apple is taking a similar approach, focusing on on-device AI processing rather than cloud-based services. Apple’s new devices offer AI capabilities along with the promise of consumer data privacy benefits.
Pushing Costs to the Edge: Opportunities and Tradeoffs
There are key advantages to doing generative AI inference on personal devices, including phones, laptops, and smartwatches – a concept known as “edge computing.” It can mitigate the costs of data centers, diminishing the environmental impact related to resources, waste, heat, and water use, which can lower AI’s carbon footprint. It may decrease bandwidth requirements and improve user privacy.
However, edge computing introduces its own set of challenges. Edge computing shifts computation costs to consumers, which can drive demand for new devices, despite concerns regarding environmental impact and economic factors. This shift could intensify with newer, more demanding generative AI models, potentially creating more electronic waste. Furthermore, if the use of a certain type of AI is based on a device, it has the potential of creating a digital divide for users, especially in educational settings. Despite the appearance of a more “decentralized” approach, a scenario might emerge where a handful of companies control the transition, diminishing the value promised on the shift to decentralization. The financial and environmental cost impacts of these evolving methods are noteworthy.
As the expenses surrounding AI infrastructure rise and model development advances, shifting costs to consumers has become an increasingly attractive strategy for AI companies. While bigger organizations such as universities or governments may manage these costs, most consumers and small businesses may find this increasingly challenging.
