Silicon Valley Tech Briefing: AGI by 2030, Nvidia Vera Rubin, and OpenAI's Super App Strategy

1. Demis Hassabis warns: AGI Could ARRive by 2030
Google DeepMind CEO Demis Hassabis predicted at Google I/O that artificial general intelligence (AGI) will arrive around 2030 (give or take a year), likening it to a technological "singularity"—a critical point of no return after a breakthrough. He believes AGI will trigger a social revolution "ten times faster and ten times more impACTful" than the Industrial Revolution, stating that humanity is currently "at the foothills of the singularity." Hassabis emphASIzed that AGI's core value lies in scientific discovery, not just chatbots; Google is advancing AI APPlications in drug discovery and weather forecasting through its "Gemini for Science" initiative.

2. Jensen Huang Confirms Vera Rubin Chip Mass Production: HBM4 Memory Costs Surge
NVIDIA's Vera Rubin platform utilizes sixth-generation HBM4 mEMOry, featuring a single stack capacity of 36GB and bandwidth of 2TB/s, representing over a 60% improvement over HBM3E while reducing power consumption by 15%. However, memory's share of the bill of materials has jumped from 9% to 26%, bringing the material cost for a single rack to $7.8 million. The HBM market is dominated by SK Hynix (over 50% share), Samsung (47%-49%), and Micron (10%-16%), a high concentration that introduces systemic risk. SK Hynix plans to double wafer capacity over the next five years, following a 193% year-over-year increase in HBM bit shipments in 2024. In Q1 2024, global AI server shipments grew 109.6% YoY, with HBM-equipped servers accounting for over 70% of memory servers, driving up data center hardware costs.

3. OpenAI Plans to Upgrade ChatGPT into a Super App
Facing pressure from its IPO and $14 billion in losses, openai (valued at $852 billion) plans to transform chatgpt into a "Super App" that integrates programming tools (Codex) and AI Agents, aiming to increase enterprise customer revenue share from 40% to 50%. The core of the upgrade is a shift from a Q&A tool to a task execution system, enhancing complex instruction understanding, task planning, tool invocation, and self-correction capabilities. As anthropic's enterprise adoption rate reached 34.4% in May 2026, surpassing OpenAI's 32.3% for the first time, OpenAI needs to defend its high-margin customer base in finance and tech through product differentiation.

4. Anthropic's Unreleased "Oceanus" Model Appears on API Proxy, Priced Three Times Higher Than Claude Opus
Anthropic's unreleased model, codenamed "Oceanus" (an upgrade from Claude Mythos), appeared on API proxy services on June 3, 2026, priced at $16 per million input tokens and $80 per million ouTPUt tokens—roughly three times the cost of its flagship claude Opus. The leak Prompted Anthropic to immediately halt a red teaming project and launch an internal investigation, potentially impacting the company's listing plans. Oceanus was previously accessible only through the "Project Glasswing" SECurity initiative to around 50 partners for code security scanning, where it had discovered over 10,000 high-risk vulnerabilities.

5. google Develops AI Memory Compression Tech: 10 Million Document Vector StoRAGe Shrinks from 31GB to 4GB
Google introduced the TurboQuant compression algorithm, which uses vector quantization to compress LLM KV caches from 32-bit to 3-bit, reducing memory usage by a factor of six (a query on 1 million tokens drops from 24GB to 4GB) with zero accuracy loss. This Technology increases AI Inference speed by 8x and lowers per-token costs by 30%-50%, achieving search speeds that surpass FAISS. This significantly lowers the barrier for local deployment of large models, enabling models like Gemma to run on edge devices such as phones and cars, while also causing volatility in the memory chip market (SK Hynix's stock price dropped by 5.7%).

6. UK's NHS Rolls Out Microsoft 365 Copilot to 505,000 Employees
The UK's National Health Service (NHS) is providing Microsoft 365 Copilot access to 505,000 CLInical and support staff to streamline administrative workflows, helping healthcare workers dedicate more time to patient care. The rollout will be phased, expanding to 200,000 users within the first six months. According to data from the world's largest Healthcare AI pilot, the tool saves 43 minutes per person per day in administrative time (equivalent to approximately 5 weeks per year). AI is being deployed across multiple scenarios, including ward clerks handling discharge processes, medical secretaries creating meeting minutes, and HR and finance transactions. This initiative is expected to reduce the need for 380,000 staff hires by the mid-2030s.

7. Tesla AI infrastructure VP Raj Jeganathan Officially Departs
Tesla's Vice President of AI Infrastructure, Raj Jeganathan, has officially left the company after 13 years. He was responsible for building the underlying neural network architecture for Full Self-Driving (FSD) and world-class GPU clusters that support the training of data spanning billions of miles. Jeganathan's departure is part of a recent wave of executive exits as Tesla pivots toward robotics as its core business. He will join the Cloud Infrastructure provider Chronoscale as CTO. The ability of his successor to maintain Tesla's edge in AI infrastructure is now a key focus.

8. OpenAI Chip Veteran Clive Chan Jumps to Anthropic
Clive Chan, a core member of OpenAI's custom chip project, has defected to Anthropic. This follows the departure of co-founder Andrej Karpathy, highlighting intensifying talent competition between the two leading AI firms. OpenAI is partnering with Broadcom to build a 10GW custom AI accelerator system, with the first racks to be delivered in the second half of 2026 and the project running until the end of 2029, aiming to reduce its dependence on Nvidia. Meanwhile, Anthropic has completed a $65 billion Series H funding round, reaching a Valuation of $965 billion and overtaking OpenAI, while building a diveRSIfied computing ecosystem through a hybrid approach (Google TPUs + Amazon Trainium + NVIDIA GPUs).

9. Notion Abruptly Re-routes Traffic Due to Anthropic Model Performance Degradation
On June 7, 2026, Notion AI temporarily disabled all Anthropic models and re-routed requests to other providers after a performance drop was observed in the Opus 4.7 and 4.8 models. The incident lasted approximately 18 minutes (04:25-04:43). The failure led to increased request failure rates for users selecting related models, making exclusive features dependent on Anthropic's models temporarily unavailable. Industry observers noted that as AI becomes part of the production environment, continuous and stable service output is a company's true "moat."

10. Google Gemini Pseudo-Context Alignment Vulnerability Exposed
A "pseudo-context alignment" vulnerability in Google Gemini bypassed deferred confirmation mechanisms through multilingual obfuscation and silent hyperlinks, potentially enabling risks like remote smart home manipulation and contact list tampering. Google patched this vulnerability in November 2025 by upgrading its content classifiers. OpenAI released a GPT-5 security update on June 10, 2026, adding a new multilingual malicious instruction detection module, while Anthropic launched a context alignment verification tool. According to Sina Finance, 73% of AI systems are susceptible to prompt injection attacks, AI-driven attacks have increased by 89% year-over-year, and the average cost per shadow AI incident is $4.63 million.

11. Microsoft Discloses Claude Code GitHub automation Vulnerability
On April 29, 2026, a Microsoft Research team discovered a high-risk Prompt Injection vulnerability in Claude Code within GitHub automation workflows. Attackers could use malicious tickets to trick the AI into reading sensitive system files (such as API keys in the /proc directory), threatening CI/CD environment credential security. Anthropic released Claude Code version 2.1.128 on May 5, adding a sensitive path access whitelist mechanism that strictly limits read permissions for critical directories like /proc and /etc. The security design flAWS in AI coding assistants have prompted multiple tech giants to strengthen sandboxing mechanisms and dynamic permission management.

12. AMD RDNA 5 Graphics Cards Delayed to Mid-2027 or Later
The launch of AMD's next-generation RDNA 5 discrete graphics cards has been postponed from the expected late 2026 window to Q2-Q3 2027 or even later. Nvidia's GeForce RTX 60 series (Rubin architecture) may also be delayed until after mid-2027. In Q1 2026, global PC GPU shipments totaled 70.3 million units, a 7.5% quarterly decline. JPR forecasts an overall GPU compound annual growth rate (CAGR) of -3% from 2025 to 2029. With the product cycles of both giants lengthening, consumer upgrade cycles may extend from 2-3 years to 3-4 years. future breakthroughs may rely more on new scenarios integrating AI and gaming.

13. Apple WWDC Opens Monday: Deep Integration of Custom Gemini AI Model
Apple and Google have entered a multi-year AI strategic partnership under which Apple pays approximately $1 billion annually for access to a custom Gemini model. This model features a parameter scale of 1.2 trillion, vastly exceeding Apple's in-house 150-billion-parameter cloud model. At WWDC 2026, Apple will showcase iOS 27, macOS 27, and other updates deeply integrating the Gemini model, with key upgrades to Siri and all Apple Intelligence features. Google's Gemini now spans both iOS and Android mobile operating systems, reaching over 5 billion terminal devices, while Google's parent company, Alphabet, has seen its market capitalization surpass $4 trillion.

Open Source Trend
14. Supabase Secures $500 Million Series F, valuation Soars to $10 Billion
The open-source Backend-as-a-Service (BaaS) platform Supabase has completed a $500 million Series F funding round, reaching a post-money valuation of $10.5 billion. This doubles its $5 billion valuation from its Series E round in 2025, representing a more than fivefold increase in value over two years. More than 60% of new databases are now automatically created by AI agents, making the platform critical infrastructure for AI-native application development. Its user base has surged from roughly 5 million to nearly 10 million, and the number of database deployments has grown by 600% year-over-year. Built on PostgreSQL, the platform offers full-stack backend capabilities like real-time databases and authentication and has garnered over 70,000 stars on GitHub.