Analysis for informational purposes only. Capital at risk.
The Memory Ceiling in the AI Industry: Google’s decision to cap Gemini tokens signals an AI hardware bottleneck, from HBM to server DRAM. The era of unmetered compute access is ending.
The Ecosystem Toll: By embedding agentic AI across Search, Gmail, and YouTube, Google transforms its product suite into an ecosystem. A “base quota + pay-as-you-go” model protects margins on the downside and captures agent-driven token volume on the upside.
The Super App Readiness: China’s WeChat, Alibaba, and ByteDance already operate the integrated environments Google is racing to build. For investors, the critical question is which ecosystem model captures the agentic era’s economics.
Google latest annual product showcase (I/O 2026) wasn’t just a regular product rollout.
It signals a structural shift in the AI industry: the era of unmetered compute is ending, and ecosystem integration is the new monetisation lever.
The “Memory Ceiling” in the AI Industry
Google put token usage caps on Gemini subscribers. The move signals a supply chain constraint in the AI industry, not a product strategy.
Global memory shortage: High-Bandwidth Memory (HBM) and server DRAM contract prices have risen sharply. Samsung’s HBM capacity is sold out through end-2026. SK Hynix signals demand will outpace supply for three years — new factory capacity does not come online until late 2027.
Alphabet’s cloud contract backlog increased by 5x within one year. Its CEO admitted cloud revenue would have been higher if the company could meet demand.
Chinese solution: Chinese models such as DeepSeek have optimised for compute efficiency using Mixture-of-Experts architecture, model distillation, and quantisation. The result is two distinct AI markets: Western hyperscalers optimise for raw throughput, while Chinese platforms maximise yield per compute unit.
Memory supply for China: ChangXin Memory Technologies (CXMT), China’s leading DRAM maker, is preparing to list in Shanghai in 2H26. It currently produces consumer DDR4 and DDR5 but targets HBM production by end-2026, potentially alleviating China’s memory supply constraint in coming years.
China’s architecture efficiency gains and its memory build-out address different parts of the AI supply chain.
Usage-Based Pricing Captures the Token Economy
Google moved subscriptions from flat-rate access to “base quota + pay-as-you-go.”
Subscribers exceeding their quota are throttled to a lighter model or charged for additional tokens. Enterprise cloud has operated on consumption-based pricing for a decade. Porting this model to consumers is a logical next step.
Downside protection, volume capture: The base quota protects against margin pressure from uncapped compute exposure. Pay-as-you-go captures the volume upside as agentic workflows push token consumption 10-50x higher per query.
The token economy potential: The total daily output tokens generated by all AI models is projected to exceed the combined spoken and written output of the world’s population by June 2026. Google Cloud’s first-party models alone now process 23 trillion tokens daily, up from 14.4 trillion last quarter. OpenRouter’s weekly token volume rose by 6x YTD.
The fastest-growing use case is agentic inference: multi-step workflows where models plan, retrieve context, revise, and iterate. A simple chatbot requires around 30,000 tokens for a summary task; an AI agent executing a comparable programming task can consume up to 20 million tokens.
Tokens consumption as a corporate KPI: Token consumption is becoming an internal performance metric at the largest tech companies.
Meta’s 85,000 employees burned through 60 trillion tokens in 30 days via an internal AI leaderboard called “Claudeonomics”. Meta’s CTO said one top engineer spends the equivalent of his salary on tokens and reportedly 10x’d his output. Amazon and Microsoft have followed suit, with employee performance reviews now incorporating AI usage impact metrics.
The token demand is so high that budgets are breaking. Uber exhausted its entire 2026 AI budget by April after Claude Code adoption jumped from 32% to 84% of its engineers. Monthly API costs per engineer ran $500 to $2,000. By spring, 70% of Uber’s committed code originated from AI tools.
China’s spot market assault: Chinese LLMs captured about 40% of OpenRouter token volume currently, up from 10% in January, while the combined share of US hyperscalers on the same platform dropped from 70% to 40% over the same period.
Aggregators attract cost-sensitive developers and SMEs who use dynamic routing: routine tasks go to the cheapest capable model, while complex queries are reserved for premium Western models. Chinese LLMs have captured the bulk of this routine compute by delivering 80–90% of Western flagship performance at roughly 20% of the cost.
The Agentic Bundle: Dismantling the App Layer
Google integrates agentic capabilities across its entire product line, translating its product portfolio into an ecosystem.
Multiple features, one strategy:
- The AI Search Box lets users find and book hotels natively, bundling items into a unified “Universal Cart” that bypasses individual merchant apps and carries a potential take-rate for every transaction.
- Gmail Live turns the inbox into a voice assistant.
- Ask YouTube answers technical questions by jumping to the exact frame containing the solution.
- Daily Brief aggregates Workspace data overnight into a personalised morning summary.
Dismantling the app layer: By offering a one-stop, seamless agentic ecosystem to users, Google essentially disintermediates individual apps and captures the corresponding monetisation opportunities.
The Super App Advantage: China’s Head Start
Google’s products are currently fragmented: browser (Chrome), search, email (Gmail), video (YouTube), documents (Docs), cloud (Drive), payments (Google Pay), messaging (Messages), and meetings (Meet). Each requires a separate interface and workflow.
Google is trying to integrate these silos through agentic AI — a single assistant that moves across them on the user’s behalf.
In China, the super app already is the OS.
Tencent: WeChat handles communication, payments, commerce, document management, ride-hailing, travel booking, and financial services — all without the user ever leaving the app.
Alibaba: Alibaba’s ecosystem runs e-commerce, cloud infrastructure, enterprise workflows, and payments under one roof.
ByteDance: Embeds AI directly into content, commerce, and creation inside Douyin. The user does not switch contexts. The context lives inside the app.
Baidu — the exception: Baidu has a search engine comparable to Google Search, but lacks the commerce, communication, and enterprise workflow breadth of its peers. Without the ecosystem, its Ernie Bot has struggled for traction.
The App-Only Trap
The agentic era creates a structural problem for standalone apps. A standalone app cannot offer agentic workflows spanning email, search, calendar, and payments. A pure-play AI chatbot cannot execute transactions across disconnected platforms.
The structural threat: Take Booking.com as an example. Its business model takes a 15-20% commission on every hotel booking and flight. Google’s Universal Cart lets users search, compare, and book hotels directly from the AI Search Box, bypassing Booking’s entire intermediation layer.
A three-way dilemma: Integrate with Google’s cart and give up margin. Build its own AI travel agent and burn capital against Google. Or do nothing and watch transactions migrate to Google’s checkout.
Meituan — the China example: Meituan has deep local commerce capabilities but its major competitor, Alibaba, has integrated local commerce businesses (Ele.me) into the agentic workflow, potentially threatening Meituan’s market position.
The next phase of the AI race is shifting from model quality to workflow ownership. Google has signalled the pricing model for that race. China’s super apps have already built the roads.
This article is a “periodical publication” for information only and is not investment advice or a solicitation to buy or sell securities. This article does not constitute a “personal recommendation” or “investment advice” under UK FCA regulations. Investing in equities involves significant risk. The author holds NO position in the securities mentioned. There is no warranty as to completeness or correctness. Please do your own due diligence or consult a licensed financial adviser. Please read the Full Disclaimer before acting on any information. Images created with the assistance of AI.