This article is archived. For updated research and insights, please visit our new site Small Island Research Notes on Tech and Future.
AI Strategy Shifts Among the Big Six: Four Core Trends from Compute Scale to Efficiency Competition
In less than three years, the focus of the AI race has shifted three times. It began with a contest to build the largest and most capable models, moved into a rush to secure computing power, and has now arrived at a phase defined by efficiency, the rise of AI agents, and the first real tests of commercial viability. Based on insights from the most recent earnings calls of six leading technology companies — Microsoft, Amazon, Google, Meta, Apple, and Tesla — the next 12 to 18 months will revolve around four core trends shaping the AI landscape.
- Optimizing AI infrastructure: Cloud-oriented companies are entering the multi-gigawatt data center era and focusing on improving tokens-per-GPU efficiency, energy use, and latency. Hardware-oriented players are deepening their on-device AI strategies and embedding AI into their products.
- The era of AI agents: AI is moving from conversational tools to agents that can take initiative, connect to tools, and carry out tasks in daily workflows. Three main paths are emerging: purely digital enterprise agents, hardware-enabled agents, and physical-world automation.
- The commercial validation phase: From the second half of 2025 through the first half of 2026, companies will face proof points in high-stakes arenas, including enterprise AI agents, autonomous driving and robotics, AI wearables, and AI-powered advertising and e-commerce.
- Efficiency as the new battleground: Competition is shifting from sheer GPU volume to performance per unit of resource, spanning hardware architecture (Tesla’s “intelligence per GB”), model-level efficiency (Microsoft and Google’s tokens-per-GPU gains), and algorithmic optimization in applications (Meta and Amazon).
Cloud-oriented giants are competing fiercely in enterprise AI agents, infrastructure build-out, and efficiency gains, while hardware-oriented companies are seeking breakthroughs in consumer access points and real-world automation. The year 2026 will be a pivotal test of commercial viability. Success in high-commitment use cases could spark a second wave of enthusiasm. Failure may slow both investment and technological momentum. In the end, leadership will not be decided by who has the largest models or the most GPUs, but by who can integrate AI most effectively into daily life and industry, turning it into sustainable business value.
Introduction
In less than three years, the focus of the AI race has shifted through several distinct phases. It began with a competition to build ever-larger models, moved into a rush to secure computing power, and has now reached a stage defined by efficiency, the rise of AI agents, and the first tests of commercial viability.
From 2022 to 2023, the generative AI wave ignited by ChatGPT pushed technology leaders into a model-building contest. Companies raced to release larger, faster, and more capable large language models. Victory was often measured by parameter counts and benchmark scores. Yet this contest came at an extraordinary cost and lacked sufficient commercial grounding.
From 2023 through the first half of 2025, companies began to recognize that the real bottleneck in AI development lay in computing resources. This led to a phase of capacity accumulation. Microsoft, Google, Meta, and Amazon made massive purchases of NVIDIA GPUs, locking in multi-year supply agreements and building multi-gigawatt data centers to meet training and inference demands. But simply stacking more compute proved costly, and performance gains did not always match the scale of investment.
In the second half of 2025, attention began to turn toward efficiency, the deployment of AI agents, and the validation of commercial models. The focus shifted from adding more GPUs to finding ways to accomplish more with the same resources. This included improving tokens-per-GPU throughput and strengthening inference performance. At the same time, AI began to move beyond conversational formats toward agents capable of taking action, connecting to tools, and embedding themselves in daily workflows, ranging from enterprise operations and autonomous driving to AI-enabled eyewear and e-commerce advertising.
While Apple, Amazon, Google, Meta, Microsoft, and Tesla have pursued different paths in AI investment since the generative AI wave began, these differences were less apparent in previous quarters. This quarter, as deployment models take shape, investment priorities diverge, and commercialization timelines become clearer, those distinctions have come sharply into focus.
Cloud-oriented vs. Hardware-oriented AI Leaders
Looking at the AI strategies of the six major technology companies, it is clear that while all are investing heavily, their deployment models and investment structures follow two distinct paths. These differences did not emerge overnight; they reflect long-standing business foundations and competitive strengths.
1. Cloud-oriented leaders (Microsoft, Amazon, Google, Meta)
Their strength lies in global cloud computing platforms, large-scale data center networks, and robust software ecosystems. Their AI strategies focus on building massive computing capacity while continuously improving infrastructure efficiency. In recent years, they have introduced proprietary AI chips such as Microsoft’s Maia AI accelerator and Cobalt cloud CPU, Google’s TPU v5, and Amazon’s Trainium 2 and Inferentia 2. These chips operate alongside NVIDIA GPUs, balancing performance with cost while reducing supply chain dependence. Their business models center on subscriptions and API usage, with advertising serving as an important AI monetization channel.
2. Hardware-oriented leaders (Apple, Tesla)
Their strength lies in integrating hardware products, ecosystems, and specialized computing architectures. Their AI strategies lean toward embedding AI deeply into devices (on-device AI) or physical products such as autonomous driving systems and humanoid robots. This approach reduces reliance on cloud infrastructure while strengthening user experience and ecosystem stickiness. Their business models are driven primarily by hardware sales and value-added services, with AI features playing a central role in driving device upgrades and product adoption.
Table 1. Classification of Cloud-oriented and Hardware-oriented AI Leaders
| Company Type | Representative Companies | Core Business Strengths | AI Strategic Focus | Commercialization Model |
|---|---|---|---|---|
| Cloud-oriented | Microsoft, Amazon, Google, Meta | Global cloud computing platforms, extensive data center networks, platform ecosystems | Build multi-GW data centers, develop proprietary AI chips (TPU, Trainium), provide cloud-based generative AI models and agent services (Copilot, Gemini, Bedrock, Business AI) | Enterprise AI subscriptions, API usage-based revenue, advertising monetization |
| Hardware-oriented | Apple, Tesla | Hardware products and ecosystems, specialized computing architectures | On-device AI (Apple Silicon), physical AI (FSD, Robotaxi, Optimus) to reduce cloud dependence and deeply integrate with hardware experiences | Hardware sales, value-added services, AI features driving hardware upgrades |
These two distinct models mean that, on the road to AI commercialization, they will face very different validation timelines, capital expenditure structures, and return profiles. Understanding this distinction not only helps interpret the signals emerging from recent earnings calls but also offers a clearer view of how each is likely to compete in the AI market over the next one to two years.
Four Core AI Trends
As generative AI moves from technical exploration to the race for deployment, the strategies of the six leading technology companies are becoming more focused and increasingly distinct. Over the next 12 to 18 months, four core trends will shape the landscape:
- Optimization of AI infrastructure
- The rise of the agent era
- The start of the commercial validation phase
- Computing efficiency as the new battleground
The sequence of these trends reflects the full arc of AI development, from building the foundation to deployment, then to validation, and finally to long-term optimization.
Trend 1: From Stacking to Optimizing AI Infrastructure
The transition from the “compute accumulation” phase of 2023–2024 to 2025 marks a shift in focus. The question is no longer simply who has more GPUs, but how to build infrastructure that is more efficient and more adaptable to support long-term AI commercialization.
For the cloud-oriented leaders (Microsoft, Amazon, Google, Meta), the past year has brought them into the multi-gigawatt data center era. Their priorities are moving from expanding GPU counts to improving tokens-per-GPU efficiency, reducing energy consumption, and lowering latency. At the same time, sovereign AI clouds, low-latency cloud services, and private deployments have become important directions, ensuring that key customers can run generative AI in secure and compliant environments.
For the hardware-oriented leaders (Apple, Tesla), Apple is pursuing an on-device AI plus private cloud architecture, keeping much of the AI processing on Apple Silicon devices to reduce cloud load and protect privacy. Tesla is embedding AI directly into its products, from Full Self-Driving (FSD) and Robotaxi to the Optimus humanoid robot, using physical AI as a core differentiator.
Trend 2: The Era of AI Agents
Over the past year, generative AI has largely taken the form of chatbots. Yet conversational AI often lacks stickiness, tending to remain in one-off interactions or experimental use. In contrast, AI agents can connect to tools, take initiative, and embed themselves in daily work and life. This ability to act within real workflows is central to their long-term commercial potential.
From the latest earnings calls, it is clear that the six major companies are shifting their focus toward AI agents capable of carrying out tasks, with three primary paths emerging in different market dimensions:
1. Cloud-native Enterprise Agents
These agents operate entirely in cloud environments, focusing on enterprise workflows and data processing without relying on specific hardware as an entry point.
- Google Agentspace: A foundational enterprise agent platform that enables companies and developers to build their own corporate AI agents.
- Microsoft Foundry Agent Service: Also a cloud-based enterprise agent platform, but deeply integrated with Microsoft 365 and Copilot to strengthen workflow capabilities within Microsoft’s ecosystem.
- Amazon Bedrock Agent: A cloud-based agent with a more vertical focus, specializing in e-commerce, customer service, and logistics.
2. From Digital Agents to Consumer Hardware Entry Points
These agents retain the core capabilities of digital agents but rely on hardware devices as the main interface, making interactions more immediate and natural.
- Meta Business AI: Essentially still an AI agent, but accessed through AI-enabled glasses, marking the first step from pure cloud to hardware-based entry.
- Apple Personalized Siri: Also a hardware-enabled agent, deeply integrated with the iPhone and the broader Apple ecosystem, enhanced by Apple Intelligence to deliver personalized task handling.
3. Physical-world Automation
These agents do more than act in the digital realm; they can operate in the physical world, performing real-world tasks.
- Tesla FSD and Robotaxi: AI agents in the transportation domain that can perceive their surroundings, make driving decisions, and carry out mobility services, representing a fundamentally different market dimension from digital agents.
Trend 3: The Commercial Validation Phase Begins
From the second half of 2025 through the first half of 2026, the six leading technology companies will enter a decisive period for AI commercialization. Over the past two years, they have committed unprecedented capital to infrastructure, model development, and product design. These investments must now begin to translate into measurable business returns, such as return on investment (ROI).
The earliest results will emerge in a few high-stickiness application areas. We can rank them by their alignment with each company’s AI strategy, their maturity, and the urgency of market validation.
First, enterprise-grade AI agents are at the core of nearly every cloud-oriented company’s strategy. They represent the largest investment areas and are tightly integrated with existing enterprise cloud services. These will be the first to enter real-world usage and face evaluation, testing whether they can truly become indispensable daily work partners.
Second, autonomous driving, Robotaxi services, and the production of Optimus robots, led by Tesla, will be closely watched. Although they face significant regulatory and technical hurdles, success in scaling operations could create landmark commercialization cases.
Third, AI glasses and wearable devices, championed by Meta and Apple, have long-term potential for high user engagement but remain in the early adoption stage. Market acceptance, retention, and conversion to paid usage will require more time to observe.
Finally, AI-powered advertising and e-commerce, already widely applied in the ad and recommendation systems of Meta, Google, and Amazon, are primarily efficiency improvements within existing businesses. Their potential for transformative impact is lower than the other applications, and thus they have a lower priority for immediate validation.
The outcomes of this stage will directly determine the pace of future capital spending and product strategy. If commercial validation falls short, both investment enthusiasm and the speed of product expansion may slow significantly. If it succeeds, strong case studies will fuel the next wave of AI growth.
Trend 4: From Computing Scale to Computing Efficiency
Computing power remains the foundation of generative AI, yet the focus is shifting toward achieving more with fewer resources. AI cannot rely indefinitely on buying more GPUs to expand capacity, especially as power availability, cost, and supply chain constraints become pressing bottlenecks. In this context, efficiency is emerging as the sustainable basis for competition. This shift is a natural evolution from the “compute accumulation” era to a more mature stage.
In the latest strategies of the six leading companies, improvements in computing efficiency can be grouped into three layers. Together, they form a bottom-up chain of optimization that spans from hardware architecture to commercial applications.
1. Hardware and System Architecture Level
Tesla has introduced a new metric for measuring AI efficiency called “intelligence per GB,” which reflects how effectively AI systems use memory to deliver intelligence. This metric represents the most fundamental layer of efficiency measurement, focusing on improving the density of intelligence at the physical resource level.
2. Model Inference and Training Efficiency Level
One level higher, Microsoft and Google are working to improve tokens-per-GPU processing efficiency so that the same hardware can handle more generative tasks. This metric targets the optimization of generative AI model performance within existing hardware limits. Compared with Tesla’s metric, it sits closer to the application layer but still focuses on maximizing the use of core computing resources.
3. Application and Algorithm Optimization Level
At the layer closest to business applications, Meta and Amazon are improving efficiency in algorithms and recommendation systems, such as reducing inference costs and speeding up ad-serving computations. Although these optimizations take place at the application level, they can significantly lower AI operating costs and directly enhance ROI in advertising and e-commerce.
Summary of the Four Core Trends
As shown in Table 2, these four trends together provide a framework for understanding how the six companies are shaping the AI landscape over the next one to two years. They also reveal the roles that different types of companies may play in this evolution. The next phase of AI infrastructure competition will not be decided by who has the most GPUs, but by who can achieve the highest performance and commercial efficiency with finite resources.
Table 2. Four Core AI Trends
| AI Trend | Signals from Earnings Calls | Representative Companies |
|---|---|---|
| 1. Optimization of AI Infrastructure |
|
|
| 2. The Era of AI Agents |
|
|
| 3. The Commercial Validation Phase |
|
|
| 4. Computing Efficiency as the New Battleground |
|
|
Conclusion
Generative AI is moving beyond its early phase of model competition and compute accumulation into a new stage driven by efficiency and commercial validation. Optimization of AI infrastructure, the rise of AI agents, the start of the commercial validation phase, and computing efficiency as the new battleground will be the key trends shaping the industry over the next one to two years.
As shown in Table 3, cloud-oriented leaders are competing intensely in enterprise AI agents, infrastructure build-out, and efficiency gains. Hardware-oriented leaders are seeking breakthroughs in consumer access points and real-world automation. The success or failure of these different approaches will determine who can sustain leadership in the AI era.
Despite their varied strategies, the six companies share a clear consensus: AI is the primary arena for the next phase of competition. While the cloud-oriented and hardware-oriented paths are diverging, both sides are working to strengthen their positions in infrastructure and agent applications at the same time.
The year 2026 will serve as a defining year for commercial validation. If agents and hardware-based AI can prove their value in high-engagement scenarios, it could spark a second wave of AI enthusiasm. If not, the market may enter a period of narrative fatigue, slowing both investment and technological progress.
Over the next 12 to 18 months, the key developments to watch include:
- Whether enterprise AI agents can become indispensable daily work tools
- Whether autonomous driving and Robotaxi services can overcome regulatory and production hurdles
- Whether AI wearables can achieve lasting engagement and paid adoption
- Whether AI-powered advertising and e-commerce can deliver meaningful revenue growth
Ultimately, leadership in AI will be decided not by who has the largest models or the most GPUs, but by who can integrate AI most effectively into everyday life and industry, turning it into sustainable business value.
Table 3. AI Development Types and Trend Positioning of the Six Leaders
| Company Type | Company | AI Focus Areas | Investment and Deployment Directions | Key Commercial Validation Points | Current Trend Positioning* |
|---|---|---|---|---|---|
| Cloud-oriented | Microsoft | Azure AI infrastructure, Copilot enterprise agents | Multi-gigawatt data centers, tokens-per-GPU efficiency improvements | Whether Copilot becomes an indispensable daily enterprise tool | Accelerating deployment |
| Amazon | AWS Bedrock, AI-driven advertising monetization | Proprietary AI chips (Trainium 2 / Inferentia 2), Bedrock Agent | Sustained high demand for AWS AI, integration of DSP advertising | Accelerating deployment | |
| Gemini, multimodal search agents | AI Overviews, Agentspace | Improvement in AI search performance and ad conversion rates | Accelerating deployment | ||
| Meta | AI personal assistant (Business AI), AI glasses | Large-scale AI training clusters (Prometheus / Hyperion), Business AI | Retention and monetization model for AI glasses | High-expectation phase | |
| Hardware-oriented | Apple | On-device AI, personalized Siri | Apple Silicon plus private cloud | 2026 Siri upgrade driving hardware refresh cycle | Initial validation |
| Tesla | Robotaxi, Optimus humanoid robot | FSD upgrades, autonomous driving agents | Geographic coverage and production scale of Robotaxi | Initial validation |
*Definition of Current Trend Positioning
- Accelerating Deployment: The product has completed core development and entered large-scale deployment, with adoption rates rising quickly and becoming part of regular daily use.
- High-Expectation Phase: The market and the company hold high expectations for the product’s potential, but large-scale adoption and a proven business model have yet to be established.
- Initial Validation: The product has completed core technical development and has entered small-scale pilot operations or regional rollout, with commercial viability and scalability still being tested.
This article is part of our Global Business Dynamics series.
It explores how companies, industries, and ecosystems are responding to global forces such as supply chain shifts, geopolitical changes, cross-border strategies, and market realignments.
See more in this category, or explore more notes here.