The conference focuses on three main lines: the iteration of the Gemini new model, the construction of the AI agent ecosystem, and the creation of a commercial closed loop for "agent checkout" based on the protocol. Although AI search may weaken traditional advertising clicks, self-developed TPU exports and strong cloud business are becoming new growth engines, with deep ties to Anthropic also seen as a strategic hedge at the infrastructure level.

Written by: Zhang Yaqi

Source: Wall Street Journal

After a 140% cumulative increase in stock price over the past year, Alphabet faces its most important annual investor narrative test—the Google I/O Developer Conference kicked off on Tuesday, where the market will gauge whether the company's AI full-stack layout represents real product competitiveness or an already fully priced market narrative.

The most anticipated highlights of this conference focus on three main lines: whether the new Gemini 4.0 model will be released, whether AI agents can move from concept to commercially viable product architecture, and whether Google can integrate search, shopping, and payment into an end-to-end "agent checkout" experience through the Universal Commerce Protocol. Mizuho analysts indicate that the release of Gemini 4 will put Google back at the forefront of model leadership, while merely releasing a 3.x update would feel more like a catch-up move.

For the market, the spillover effects of the conference are also worth attention. According to Mizuho, Google's deep push into the agent commerce space could create valuation pressure on platform companies like Booking Holdings, Expedia, DoorDash, Zillow, and Instacart, a sentiment that analysts believe is partially reflected in the recent weak performance of these stocks.

Wall Street has already bought into Alphabet's AI revival narrative, but investors are looking for a clearer product roadmap at this I/O—a complete proof chain across search, cloud computing, Android, chips, and enterprise software. Plexo Capital founding partner Lo Toney stated, "Google may be the company most capable of monetizing AI at scale since it essentially controls every layer of the tech stack."

Gemini New Model: Strategic Significance Behind Version Numbers

Before the conference, there was a clear division in market expectations regarding the new version of the Gemini model. According to Citigroup analysis, given that Google only released Gemini 3.1 Pro in February this year, following a release rhythm of three to four months, the likelihood of Gemini 3.2 or 3.5 being released is higher than a complete generational leap, making Gemini 4 not the baseline expectation of analysts.

Nevertheless, the meaning of the version number has transcended technical aspects. Mizuho wrote that the release of Gemini 4 would give Google a clearer differentiated advantage in its competition with OpenAI and Anthropic, while a routine 3.x update would be interpreted by the market as catching up rather than leading.

Analysts are also looking at broader updates to the Gemini ecosystem, including progress on Project Astra (Google's universal AI assistant), deep capabilities of Gemini Live, native tool calls across Search/Gmail/Calendar/Maps, and updates to the open source model series Gemma and Gemini Robotics.

From user data, the foundation has clearly expanded before the conference. According to Citigroup data, the monthly active users of the paid version of Gemini Enterprise increased by 40% quarter-on-quarter in Q1, while the monthly active users of the Gemini app in April in the U.S. grew 127% year-on-year. The token consumption reported by Google Cloud Next has reached 16 billion times per minute.

Additionally, according to TechRadar, a new type of Gemini agent named "Remy" is under development—this is a "always-on" assistant capable of running in the background around the clock, autonomously completing tasks across third-party applications, reportedly able to handle shopping, communication with third parties, and proactively finish various tasks without waiting for user commands.

AI Agent: From Chatbot to Operational Layer

If there is a theme that runs through all agendas of this I/O, it is agents. Google has arranged multiple themed discussions covering agent programming workflows, multimodal tools, media generation, robotics, and AI agents, aiming to reposition Gemini from a chatbot to the "operational layer" of Google's entire product ecosystem—capable of understanding context and taking proactive actions.

Lo Toney stated, "The real big market lies in who can win the office AI assistant market. If the bigger opportunity lies in AI agents and their orchestration—inference infrastructure, multimodal workflows, enterprise search—that is precisely where Google has a huge opportunity to drive Alphabet's future growth."

Agent programming is an important direction for Google to position Gemini as a competitor to Anthropic Claude Code and OpenAI Codex. This category has become one of the clearest proof areas of AI business value, especially at the enterprise software level.

Gene Munster, managing partner of Deepwater Asset Management, believes that the advantage of full-stack control lies not only in scale but also in speed. "Building products on self-developed custom chips means a speed advantage. When you have energy acquisition channels, you can build data centers faster—this is crucial in the AI competition."

AI Shopping: From Search Entry to Transaction Closed Loop

Monetization may represent the biggest implicit opportunity of this I/O. Google already possesses search, shopping, autofill, and payment capabilities; now it hopes Gemini will connect these abilities into an end-to-end "agent checkout" experience, where users receive not just shopping suggestions, but transactions are completed directly by Gemini.

Google has recently continued to expand its Universal Commerce Protocol, with newly added partners including Meta, Microsoft, Stripe, Klarna, and Affirm. Google Android ecosystem president Sameer Samat described a specific scenario: having Gemini plan a barbecue, create a menu, open Instacart, add ingredients to the Safeway shopping cart, and send notifications when tasks are completed. "Layering such scenarios on top of every day, you save a lot of time over a week," he said, "features like this are what people are really excited about, it's tangible."

Lo Toney believes that Google’s experience in multimodality gives it a structural advantage. "Enterprise workflows increasingly involve video, voice, images, and code; Google has accumulated a wealth of multimodal system experience on its largest applications like YouTube, Android, Maps, Search, DeepMind, along with TPU, forming a complete closed loop."

However, the advancement of agent shopping also poses challenges to Google's own monetization model. Citigroup data shows that AI-enabled campaigns currently account for over 30% of search ad spending. However, Mizuho estimates that about 93% of AI Mode searches end without external clicks, and the natural click-through rate for AI Overview queries has dropped by around 15%—which makes how to build new advertising products within AI Mode one of the top concerns for investors. Gene Munster mentioned he will focus on Google's explanation of the monetization path for agent commerce.

Cloud Computing and TPU: New Growth Engine

For investors, the most meaningful announcements at this I/O may come from the cloud computing and infrastructure sector.

Google Cloud has become one of Alphabet's strongest growth pillars. Q1 revenue grew 63% year-on-year, outpacing Azure and AWS. The cloud business backlog reached $462 billion, up about 90% quarter-on-quarter, half of which is expected to be confirmed in the next 24 months. Revenue from generative AI products grew approximately 800% year-on-year. CEO Sundar Pichai stated in the Q1 earnings call that the number of billion-dollar plus contracts signed in 2025 has exceeded the total of the past three years, and existing customers' actual usage exceeds initial commitments by over 30%.

A new variable is external TPU sales. Google revealed in Q1 that it will begin delivering custom AI chips to external customers in the second half of 2026 and plans to further scale it by 2027. Mizuho stated that investors will closely monitor the revenue recognition method for external TPU sales (whether based on total sales or licensing revenue), profit margin structure, and their inclusion in backlog orders.

Lo Toney noted that TPU is the most undervalued part of Alphabet's investment logic, believing that in-house chips enable Google to build a highly integrated AI infrastructure covering Gemini, Cloud, YouTube, and Android. Gene Munster estimates that the entire AI chip market is currently around $500 billion annually, and even if Google only gains a small share increase, it will have a significant financial contribution to Alphabet.

Anthropic Relationship: Concentrated Risk or Strategic Hedge

Entering the I/O conference, no relationship has garnered more market attention than Google's binding with Anthropic. Alphabet holds a significant stake in this AI startup, and if the reported $200 billion commitment to cloud computing is true, it could account for a substantial portion of Google's contracted future cloud revenue. Google has committed to a total investment of up to $40 billion in Anthropic, creating a funds cycle of "capital injection into Anthropic, then flowing back to Google through cloud and TPU consumption."

This concentration risk has precedent in the market: Oracle's stock price once surged after disclosing large backlog orders related to OpenAI, only to fall due to investor concerns over client concentration; Microsoft also faces similar discussions around its relationship with OpenAI.

Lo Toney holds a different view, considering Google’s relationship with Anthropic more of a hedge than a weakness. He stated that even if enterprise customers choose Claude over Gemini, Google can still benefit from the infrastructure demands behind it—"all those Claude activities have to run somewhere. With its TPU, Google is still the winner."