1. Google I/O 2026: The Declaration of Agentic AI — A New Google That Devours Search, Shopping, and Productivity
• Core Source
“AI Overviews MAU surpassed 2.5 billion”
“AI Mode MAU surpassed 1 billion”
“Monthly tokens processed across Google services exceeded 3,200 trillion — a 7x increase year-over-year”
“You may have heard the anecdote about companies using up their annual AI token budget before May, saying that a company using 1 trillion tokens per day could save more than $1 billion annually by shifting 80% of its workload to the Gemini 3.5 Flash model”
“Switching from daily prompt limits to a ‘compute-used’ model — billing based on prompt complexity, features used, and conversation length. AI Ultra plan at $100/month.”
• Expected Impact
The core of what Google demonstrated at I/O 2026 is that AI services have transitioned to a usage-based billing model, making the path to monetization clear. The figures — monthly token processing of 3,200 trillion, AI Overviews MAU of 2.5 billion, and AI Mode MAU of 1 billion — suggest that Google has succeeded not merely in launching AI models, but in deeply embedding AI across the everyday search, shopping, and work flows of billions of people.
From an investment perspective, two changes stand out. First, the shift in billing structure. By moving from daily prompt limits to usage-proportional billing and launching the AI Ultra plan at $100/month, a structure is now in place where growing AI usage directly increases Google’s revenue. Second, the agentic commerce expansion through Universal Cart and Search Agents. By combining Google’s Shopping Graph — which holds over 60 billion product listings — with agent technology, a significant portion of the shopping journey can now be completed entirely within Google, directly threatening Amazon’s commerce moat while diversifying Google’s ad revenue model toward transaction-based revenue.
In the near term, the market concern is that the AI-driven search overhaul could reduce outbound clicks and pressure existing search ad revenue — a worry reflected in Alphabet’s stock decline on the day of I/O. However, over the medium to long term, the three pillars of agent transition + commerce integration + usage-based billing are assessed to have entered full operational mode for Google’s AI monetization model. The Gemini app’s MAU more than doubling in a year to exceed 900 million is the key data point showing this trend is accelerating.
2. NVIDIA Vera Rubin CPU: First Shipments — The Dawn of the Next-Generation AI Platform Era
• Core Source
“Key recipients: Anthropic, OpenAI, SpaceX AI, and Oracle among leading North American cloud and AI companies, with deliveries beginning sequentially on May 15 and 18”
“Architecture and process: Designed on an Arm architecture and manufactured using TSMC’s 3-nanometer (nm) process”
“88 compute cores / NVLink high-speed interconnect at 1.8TB/s / 1.5TB system memory support and up to 1.2TB/s SOCAMM LPDDR5X”
“Quanta leased a new factory in Fremont, California at a total contract value of $61.71 million (approximately NT$1.974 billion) to aggressively expand its AI server production base”
• Expected Impact
The Vera Rubin platform shipment is more than a new product delivery — it is the official announcement that the AI server market has entered a new cycle. The sequential deliveries of the Vera CPU on May 15 and 18 effectively dispelled market concerns about Rubin platform delays, and Evercore ISI confirmed through channel checks that “concerns about Vera Rubin delays are overblown.”
The supply chain impact is immediate. Taiwan’s three major ODMs — Foxconn, Quanta, and Wistron — have entered a Vera Rubin-based AI server order surge, with Quanta already committing to a $61.71 million U.S. factory lease. TSMC’s leading-edge process utilization is also expected to rise as the 3nm process is applied to the Vera CPU.
From an investment perspective, the structural significance is that the CPU shipped ahead of the GPU. With the packaging process for the Vera CPU being simpler than for the GPU, it reached customers first — raising the likelihood that full platform replacement demand will concentrate at the point of Rubin GPU shipment. HSBC raised its NVIDIA price target from $295 to $325, forecasting FY27 Q2 revenue at $91.1 billion — above the consensus of $85.6 billion. HSBC also raised its CoWoS production allocation from 900,000 to 1.1 million wafers, projecting FY28 data center revenue of $528 billion, well above the consensus of $465.3 billion.
3. MLCC at a Historic Inflection Point — AI Demand Reverses Pricing Power in Passive Components
• Core Source
“The company clearly stated that the growth driver comes from ‘exploding AI data center demand,’ forecasting that revenue from this segment will grow YoY +84%”
“Murata’s Q4 overall production capacity utilization has already risen to approximately 95%, essentially full capacity, and the company stated it will continue operating at this limit through the new fiscal year (~March 2027)”
“Taiyo Yuden recently announced price increases for consumer low-capacity and automotive MLCCs, with increases of approximately 6–13%”
“TrendForce recently found that following price negotiations between ODM manufacturers and suppliers, the average MLCC price decline has already narrowed to below 0.5% — the lowest level in the past three years”
• Expected Impact
The structural shift in the MLCC market shows that AI demand is completely inverting the existing consumer electronics-driven cycle. Murata, the industry leader, officially guiding for YoY +84% growth in AI data center revenue and maintaining 95% utilization means that demand continues to grow in an environment where additional supply is virtually impossible.
The key for investors is the transmission mechanism by which price increases are spreading from “high-end AI-dedicated products” to “mass-market consumer products.” As major Japanese and Korean suppliers shift production capacity toward high-value AI applications, supply elasticity for consumer MLCCs is shrinking quarter by quarter. Taiyo Yuden’s 6–13% increase, Samsung Electro-Mechanics’ simultaneous notification of similar increases to distributors, and TrendForce’s confirmation that the average MLCC price decline has narrowed to below 0.5% — these together prove numerically that pricing power has shifted to the supply side after years of decline.
Meritz Securities raised its Samsung Electro-Mechanics price target from KRW 1.02 million to KRW 1.6 million (+56.9%), noting that the 800VDC adoption in 2027 is expected to open the high-voltage MLCC market in earnest, driving additional ASP improvement through product mix. MLCCs are a critical passive component with thousands installed per AI server, and while relatively unfamiliar to general investors, the fact that AI infrastructure expansion is spreading across all passive component categories — beyond semiconductor equipment and materials — makes this a noteworthy issue.
4. Tesla FSD China Launch Signals Strengthen — Will Autonomous Driving Monetization Begin in the World’s Largest EV Market?
• Core Source
“Tesla China recently posted numerous job listings related to smart driving tests, and industry observers say FSD could soon be introduced to the Chinese market in earnest”
“The posted positions include smart driving test (on-road test) technicians, smart driving test engineers, and test site personnel, with locations spanning nine cities including Beijing, Shanghai, Guangzhou, Shenzhen, Suzhou, and Wuhan”
“The customer service representative replied: ‘Current customer vehicles do not yet support this feature,’ adding that ‘we are actively pursuing the approval process as required by national regulations, and will deploy to Chinese customers as quickly as possible once approval is complete'”
• Expected Impact
An FSD launch in China carries significant potential weight as a single event capable of meaningfully re-rating Tesla’s financial outlook. Tesla has already sold the 64,000 yuan smart driving package in China, but actual feature activation via OTA remains pending regulatory approval.
China is one of Tesla’s largest single markets and its most fiercely competitive EV arena. With local players like Li Auto, Huawei, and BYD positioning smart driving as a core differentiator, FSD’s delayed launch in China had been weakening Tesla’s premium positioning. The simultaneous recruitment of test personnel across nine cities and the official customer service response confirming that “approval procedures are being actively pursued” suggest the regulatory process may be entering its final stage.
If FSD is activated in China, the two main investment implications are as follows. First, structural expansion of software revenue. FSD is a software subscription model rather than hardware, representing significant leverage for improving Tesla’s margin structure. Second, recovery of competitive advantage in China. It has the potential to widen the autonomous driving technology gap again, partially offsetting the demand pressure from the Model Y price increase ($500–$1,000). That said, uncertainty in China’s autonomous driving regulatory environment and the established position of local players like Baidu remain as risk factors.
5. NVIDIA’s $90 Billion Investment Over 16 Months Cements AI Ecosystem — From Selling Shovels to Owning the Mine
• Core Source
“Over the past 16 months, Jensen Huang has committed a cumulative $90 billion and invested in 145 companies — approximately 7 times the investment scale of Alphabet”
“This is assessed not as simple financial investment, but as a sophisticated ecosystem lock-in strategy”
“TSMC’s projection that AI revenue will grow at a CAGR in the ‘high-50s%’ over the five years from 2024 implies that NVIDIA’s data center compute revenue could reach more than $700 billion by 2029”
“Assuming 50% of FCF is returned, NVIDIA’s TTM capital return could grow from $40 billion to approximately $175 billion by end-2027”
• Expected Impact
NVIDIA’s $90 billion investment strategy is a structural move to transition from a GPU-selling company to a dominant shareholder across the entire AI industry. The figure of 145 companies invested in over 16 months signals not simple financial return-seeking, but a strategy to vertically integrate each layer of the AI ecosystem centered on NVIDIA chips — at approximately 7 times Alphabet’s investment scale over the same period.
Evercore ISI used this as a basis to note that NVIDIA currently trades at approximately a 30% discount to its 10-year average P/E, and derived the implication that reverse-engineering TSMC’s ‘high-50s%’ AI revenue CAGR points to NVIDIA’s data center compute revenue potentially exceeding $700 billion by 2029. On investor sentiment, the share of investors selecting NVIDIA as the top Logic/AI name falling from 70% to 35% over the past year suggests that the pullback in expectations has already been substantially priced in, and Evercore views this as a buying opportunity.
Assuming 50% FCF return, capital return could expand from $40 billion to $175 billion by end-2027 — structurally similar to Apple’s case where buyback expansion lifted NTM P/E from 10x to 40x. The combination of ecosystem investment deepening the moat alongside aggressive capital return is identified as the key catalyst for a potential valuation re-rating.
6. OpenAI Launches ‘Guaranteed Capacity’ — AI Compute Enters the Era of Reservations
• Core Source
“Launch of a long-term compute reservation product for enterprise customers. Guaranteed access to OpenAI compute on 1–3 year commitment terms”
“Customers increasingly want certainty about stable capacity. As models continue to improve, I expect the world to be in a state of compute capacity shortage for some time”
“Of the 14–15 million chips Anthropic has committed to, approximately 1.4 million have currently been deployed”
“Severe compute shortage in ’26–’27 is beyond doubt”
• Expected Impact
The launch of OpenAI’s Guaranteed Capacity demonstrates that AI computing resources have been elevated from a simple API service to a ‘strategic asset to be secured in advance’. The 1–3 year commitment-based compute reservation model resembles cloud infrastructure long-term contracts, but the difference is the severity of supply-side scarcity. As confirmed by CLSA’s analysis of the Anthropic case, of the 14–15 million chips committed to, only approximately 1.4 million have been deployed — meaning supply against demand commitments is at roughly one-tenth the level, making the shortage structural.
For enterprise customers, the significance of Guaranteed Capacity is that securing compute in advance is the only way to achieve predictability in operating AI agents and production services. CLSA confirmed that Ampere GPU rental prices have risen +9% since the start of the year, with Hopper and Blackwell commanding 18–20% higher premiums. In an environment where compute prices themselves are rising, long-term commitment discounts become a hedging instrument for enterprises.
From OpenAI’s perspective, long-term contracts provide revenue visibility and enable infrastructure pre-investment planning — strengthening the recurring revenue base that underpins OpenAI’s IPO valuation. As Wedbush noted, the core uncertainty of the Musk lawsuit ($134 billion in risk) has been effectively resolved by the jury verdict. Combined with this, Guaranteed Capacity can be interpreted as the first institutionalization of a recurring revenue model to support that IPO path.
Comment [0]