Google, with a full-stack deployment of "agent-based AI" as a winning strategy... Cloud performance season faces a test

The “Google Cloud Next” held in Las Vegas, USA, is an event showcasing this year’s enterprise AI competition direction once again. Google has launched a comprehensive “full-stack” strategy covering all layers—from semiconductors, large language models, data management, networks to security—and emphasized its role as an “AI agent” operating system.

Google Cloud CEO Thomas Kurian stated, “We have clearly entered the era of ‘agent-based AI’,” and claimed that “every employee in every organization can become a developer.” Despite the eloquence, the message is clear: Google’s vision is to enable enterprises to use AI as autonomous software to perform actual work, rather than just simple chatbots.

At this event, Google released numerous products based on Gemini, including enterprise agent platforms, new TPU chips, ultra-large network technologies, Workspace AI features, data access tools, and security operation agents. Google also announced collaborations with Salesforce, Oracle, Atlassian, among others, accelerating ecosystem expansion. This is interpreted as a strategy: providing a one-stop service for enterprise customers when adopting AI, including models, infrastructure, data, and security.

The core is that “agent-based AI” could change the role of existing SaaS software. Users no longer need to switch between multiple applications manually but instead rely on a structure where multiple AI agents communicate and collaborate to handle tasks. Google Data Cloud Vice President Andy Gutmans explained, “The primary users will no longer be humans, but agents,” and pointed out that data platforms need to be redesigned from “human-facing” to “agent-scale.”

This shift greatly increases infrastructure demands. Google Cloud Vice President Mark Romayer said that a token required by an agent could be 20 to 50 times that of a typical chatbot. This is because a single request can trigger a series of chained tasks. Therefore, networks, storage, computing resources, and even power—every link in the AI stack—could become bottlenecks. This is also why Google has repeatedly emphasized network optimization and TPU upgrades recently.

AI investment confidence is recovering… corporate performance shows divergence

This week, the performance of US tech stocks indicates that market expectations for AI remain strong. Intel’s stock surged about 24% in a single day on Friday, driven by better-than-expected results. Texas Instruments also benefited from demand for AI data center simulation chips, with a 19% rise. SAP delivered steady results based on strong AI business performance.

However, not all companies received the same evaluation. IBM, despite posting better-than-expected results, failed to fully meet market expectations for its AI business, causing its stock to plummet about 21%. ServiceNow’s stock performance was weak due to concerns that large contracts might be delayed amid Middle East conflicts. This shows that although AI optimism has rebounded, market reactions will vary significantly depending on whether actual sales and orders can be realized.

Particularly noteworthy is that next week, major cloud giants like Alphabet ($GOOGL), Microsoft ($MSFT), and Amazon ($AMZN) will release earnings reports. Additionally, Apple ($AAPL), Meta ($META), Samsung Electronics, and Qualcomm ($QCOM) are also scheduled, which will be key indicators of how large-scale AI investment costs are reflected in financial statements.

OpenAI, Anthropic, DeepSeek… model competition intensifies

Competition in the generative AI model field is heating up. OpenAI released new models in an attempt to regain dominance; meanwhile, Anthropic has secured commitments of up to $25 billion from Amazon and up to $40 billion from Google. In Korean won, that’s approximately 36.9375 trillion KRW and 59.1 trillion KRW respectively. Both companies are leveraging massive capital to continue competing in model performance and commercialization.

China’s influence is also growing. DeepSeek released nearly competitive new open-source models, expanding its influence. In the context of US tech giants’ dominance, the rapid catch-up of Chinese AI companies remains a market variable of ongoing concern. Additionally, with Cohere announcing a merger with Aleph Alpha, Europe and Canada are also showing signs of building independent camps.

On the other hand, AI proliferation also brings security risks. There are signs that Anthropic’s restrictive model “Mythos” has been accessed without authorization, raising concerns in cybersecurity about attackers potentially exploiting it to find vulnerabilities or launch automated attacks. As AI applications become more widespread, “control” and “governance” will become as critical as “performance” in competitive advantage.

AI-driven organizational and investment reforms… increasing lock-in concerns

The era of AI agents is also demanding changes in enterprise structures. Meta has begun large-scale layoffs, and Microsoft has launched its first voluntary resignation plan targeting some US employees. Whether this is a trend of AI directly replacing human labor or a reallocation of costs to expand AI investment remains to be further observed. However, considering Meta’s multi-billion-dollar AI infrastructure contracts with Amazon Web Services, the current trend of capital flowing into equipment and infrastructure rather than labor costs is more evident.

The importance of storage and data management is also rising. Vast Data, with an enterprise valuation of about $30 billion (roughly 44.325 trillion KRW), successfully raised $1 billion (about 1.4775 trillion KRW). This indicates that investors strongly agree that the more widespread “agent-based AI” becomes, the more essential a fast and efficient data access system is.

However, this trend also raises new “lock-in” concerns. Google, Amazon Web Services, and Microsoft are all strengthening integrated solutions from models to cloud platforms, data, and productivity tools. Once a company chooses a vendor’s stack, migrating later becomes extremely difficult. When work emails, file storage, databases, and security systems are bundled together, switching costs will increase further.

Ultimately, the message from this week’s market is very clear. “Agent-based AI” is no longer in the proof-of-concept stage but is turning into a competition that simultaneously changes infrastructure, capital, and software order. Google explicitly expressed its desire to become a core player at this event, but the final victory depends on which platform enterprise clients choose as the standard. The upcoming earnings season is likely to serve as an early indicator of this choice.

TP AI Notice: This article uses a language model based on TokenPost.ai for summarization. The main content may be incomplete or inconsistent with facts.