Key Events of the Week: What Happened

From May 17 to May 23, 2026, the most visible shift in the U.S. technology sector was that AI competition moved more clearly from model capability and infrastructure expansion toward platform entry points, agent workflows, compute architecture, and geopolitical boundaries.

Google I/O stood out as the most important technology event of the week. Google introduced a broad set of updates across Gemini, AI Search, developer tools, agent-related features, and consumer applications. These updates showed that Google is placing AI back at the center of its core platforms. For Google, AI is not only about model competition. It is also a reintegration of Search, Android, Workspace, Cloud, the developer ecosystem, and everyday user interfaces. Google’s advantage may not lie only in the capability of a single model. It may lie in its ability to distribute AI across large existing entry points and usage contexts.

At the same time, Anthropic sent several signals worth watching. It continued to attract important AI talent, and it was also reported to be considering the use of Microsoft’s in-house AI chip servers. This suggests that frontier model companies are competing on model research capability, compute access, and enterprise deployment opportunities at the same time. In the past, competition among model companies was easier to understand through benchmarks and model capability. Now they need to prove that they can secure enough compute and enter more concrete enterprise and professional workflows.

NVIDIA’s signals centered on CPU and agentic AI. Jensen Huang connected the future CPU market opportunity with agentic AI, extending NVIDIA’s CPU narrative from Grace’s large-scale deployment toward the Vera CPU and Rubin platform. This broadens the scope of AI infrastructure beyond GPUs into a more complete system architecture. As AI moves from chatbots toward agent workflows, compute demand is no longer only about model training or single-step inference. It is increasingly about longer tasks, lower latency, more complex data movement, and broader system coordination.

China’s AI chip substitution narrative also continued to gain momentum. Alibaba introduced a new AI chip, showing that Chinese technology companies are accelerating the development of domestic AI compute alternatives under U.S. export controls and supply chain constraints. This does not mean China can immediately replace NVIDIA. But it will make the relationship among NVIDIA, the Chinese market, and U.S. technology controls more complex. Future AI chip competition will not only take place around product performance. It will also involve export licenses, domestic substitution, customer adoption, software ecosystems, and supply chain security.

Taken together, the U.S. technology news from May 17 to May 23 was not simply a continuation of the AI boom. It showed that AI competition is entering a more concrete deployment phase. Model companies need to find compute and enterprise use cases. Platform companies need to control entry points and distribution power. Chip companies need to expand their system role. Geopolitics continues to reshape the boundaries of the global AI supply chain.

Narrative Observation: What It Means

The most important development last week was a shift in the way the AI industry is being understood.

Over the past year, the market has often used two questions to understand the AI industry. The first was whether model capability would continue to improve. The second was whether compute infrastructure would continue to expand. But the events of last week suggest that the core question may now be changing. The question is who can put model capability into workable systems. These systems are not only technical systems. They also include platform entry points, enterprise workflows, data permissions, compute access, user habits, business models, and policy boundaries.

Google’s role reminds us that truly valuable AI may be embedded inside search, browsers, operating systems, developer tools, enterprise software, and cloud platforms. In the future, users may not always be aware of which model they are using. They may simply encounter AI naturally through existing platforms. This means that large platform companies still have a chance to regain a degree of distribution and governance power in AI.

Anthropic’s moves suggest that model companies are moving from capability demonstration toward the test of deployment capability. Model companies need talent and compute, but more importantly, they need to enter enterprise workflows, professional use cases, and high-trust applications. This is why coding, legal work, finance, cybersecurity, and enterprise workflows are becoming important. These are not only places where AI can be used. They are also testing grounds for whether AI can be paid for by enterprises, improve efficiency, and form long-term contracts.

NVIDIA’s signals show that competition in AI infrastructure is moving from single-chip supply toward system design capability. As agentic AI brings longer tasks, more real-time interaction, and more complex data movement, GPUs remain important. But CPUs, memory, networking, software stacks, rack systems, and data center architecture are also becoming part of the competitive equation. This is also why NVIDIA continues to move from being seen as a GPU company toward being seen as an AI factory and AI infrastructure platform.

The role of geopolitics is also becoming clearer. AI is no longer only a commercial competition among technology companies. It is gradually becoming a question of national supply chains, export controls, domestic substitution, and market boundaries. The development of China’s AI chip substitution will create a more complex situation for U.S. technology companies. On the one hand, China remains an important source of demand. On the other hand, policy restrictions and domestic alternatives will change the accessibility and predictability of that market.

As a result, the AI boom is being broken down into more specific questions.

• Who controls the entry points?
• Who controls compute?
• Who controls deployment scenarios?
• Who can manage costs?
• Who can put models into enterprise workflows?
• Who can continue to expand within policy boundaries?

The next stage of AI may be shaped by system integration capability. In other words, AI competition is moving toward workable systems, rather than staying only at the level of model capability or infrastructure expansion.

The Momentum of Trust: Why It Matters

Trust momentum still leans toward AI infrastructure, but the market is beginning to distinguish more carefully among different layers of AI capability.

NVIDIA remains at the center of market trust because it represents the most direct, visible, and verifiable form of AI infrastructure demand. Orders and supply chains can confirm that demand more clearly than many other parts of the AI economy. What mattered more last week, however, was that the market’s trust in NVIDIA is not only about chips. It is also about NVIDIA increasingly being seen as a system designer for AI infrastructure.

Google’s trust momentum comes from platform entry points. If Google can bring AI agents and AI Search into everyday usage, it may be able to turn AI capability into user behavior that is more frequent, more natural, and harder for any single model company to replace.

Anthropic’s narrative is no longer only about being a model company. It also needs to move toward a combination of enterprise trust, professional deployment, and diversified compute access. If Anthropic does use Microsoft’s in-house AI chip servers, it would suggest that Microsoft’s AI strategy is not only about OpenAI or Azure GPU capacity. Microsoft is also trying to bring in-house chips, cloud infrastructure, and multi-model partnerships into the same system. This would give Microsoft greater strategic flexibility between AI infrastructure and model distribution.

China’s AI chip substitution will also affect how the market prices risk for NVIDIA and U.S. AI chip companies. The development of domestic AI chips in China may not immediately change the global market structure, but it reminds the market that export controls can create substitution demand. It may also encourage Chinese customers to gradually test non-NVIDIA pathways. This could turn the China market for U.S. AI chip companies from a simple demand story into a market shaped by policy, compliance, approvals, and the pace of substitution.

Overall, the conditions for market trust in AI continue to become more specific. Infrastructure companies need to convince the market that demand can continue. Model companies need to show deployment capability and a path to commercialization. Cloud platforms need to prove that in-house chips and multi-model strategies can generate returns. Platform companies need to show that they still control user entry points.

The Coming Weeks: What to Watch

In the next few weeks, the first area to watch is whether the market will reassess Google’s AI entry-point capability after Google I/O.

The second area to watch is whether Anthropic will continue to expand its enterprise and compute strategy. The addition of important talent may strengthen expectations around Claude’s model research capabilities. If there is further progress around the reported Microsoft chip server arrangement, it would also show more clearly that frontier model companies are looking for more diversified sources of compute.

The third area to watch is whether Microsoft’s in-house AI chips begin to receive external customer validation. If Anthropic becomes an important user of Microsoft’s in-house AI chips, Microsoft’s AI infrastructure narrative would become more complete. It would also lead the market to reassess the commercial value of in-house chips developed by cloud platforms.

The fourth area to watch is whether NVIDIA further strengthens the narrative around Vera, Rubin, CPU, and agentic AI at Computex and in its upcoming earnings results.

The fifth area to watch is whether China’s AI chip substitution begins to affect the China narrative for U.S. AI chip companies. The next things to watch are not only chip performance, but also customer adoption, supply capability, software ecosystems, and whether these chips can truly absorb part of the demand constrained by export controls.

The sixth area to watch is whether AI agents move from product demonstrations into clearer usage data. The market will begin to care whether these agent features actually increase usage frequency, willingness to pay, enterprise efficiency, or new revenue creation. If agents remain only product demos, trust momentum will be limited. If agents can enter real workflows, the market’s view of AI software may gradually begin to change.

Summary

From May 17 to May 23, 2026, the AI narrative in the U.S. technology sector showed that competition is entering a more concrete and more system-oriented stage.

Google is embedding AI into Search, developer tools, consumer entry points, and everyday workflows. Anthropic may need to face the test of enterprise trust, deployment scenarios, and compute flexibility. NVIDIA is expanding the narrative of AI infrastructure toward the broader AI factory. China’s AI chip substitution also reminds the market that AI compute competition is being reshaped by export controls and domestic alternatives.

The key point of the week is that AI competition is moving toward workable systems. These systems are not only about model capability, and they are not only about compute supply. They include platform entry points, enterprise deployment, chip architecture, cloud strategy, user habits, and policy boundaries. The next stage of AI competition may no longer be only about who can demonstrate the strongest capability. It may be about who can place those capabilities into structures that are stable, deployable, monetizable, and able to retain market trust over time.