U.S. Approves 10 Chinese Firms to Purchase H200; BTC ETF Sees Nearly $1 Billion Outflow in One Week - CoinResearch Weekly Report 0515 ⚠️ Key Events This Week NVIDIA × Hermes × Open Source Model Ecosystem @nvidia has published deployment documentation and publicly endorsed @NousResearch’s Hermes framework, positioning it as a 24/7 local AI architecture designed specifically for NVIDIA hardware. Hermes, paired with Alibaba’s Qwen 3.6 small model, runs on high-end RTX GPUs or DGX Spark personal supercomputers; its sub-agent design isolates minor tasks to enhance reliability. 📌 What can be inferred from the list of U.S. business executives accompanying Trump on his 2026 visit to China? 📌 One infographic: Understanding the AI Compute Funding Chain — 3 Major Demand Sides × 5 Layers of Flow Predicted Hot Topics for Market Attention 🔭 What will Trump say during his bilateral meeting with Xi Jinping? 🔭 Will the Clarity Act be signed into law in 2026? 🔭 LCK 2026: Gen.G vs T1 — The Traffic Battle Content Highlight — The HTTP-402 Payment Protocol in the Age of AI Agents by @paultimofeev_ Credit to @hank95007 The latest edition of CoinResearch Weekly Report — keep reading 👇

On founder control and no secondary market When Jensen Huang bet on CUDA in 2006, he couldn’t possibly have known it would be useful for deep learning—because deep learning didn’t even exist at the time. What he bet on was just a vague proposition: parallel computing would always be useful for something. This means that NVIDIA’s most critical success factors were not a single strategic vision, but the combo of three things: 1. Betting on a sufficiently general direction (parallel computing) 2. Wait for a super-large and clear PMF application to come in (deep learning) 3. During the more than ten years he was waiting, the direction and he as the CEO wasn’t replaced In 2008 and 2018, the stock price was halved twice, analysts scolded him for not doing his proper job, and wall st gives selling rating. If it had been a professional CEO under pressure from short-term shareholders, CUDA would very likely have been cut in one of the cost reduction rounds, or slashed down to just a skeleton. As a founder, Huang had enough control that the board couldn’t touch him, so CUDA was able to endure until the AlexNet moment in 2012.​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​ 1. Maintain company control 2. Early IPO/TGE pre PMF is against long-term growth

Mark your calendar 🗓️ The Q2 @maplefinance Investor Call is coming up on July the 8th at 11 AM ET This is a great way to understand what the founders of the project have in mind.

This is actually insane: SBF had 8% stake in Anthropic (Claude) Due to FTX bancruptcy, 8% were sold for $1.3B Now Anthropic is raising at the HIGHEST ever Pre-IPO valuation To be precise - at $950,000,000,000 valuation. 8% were sold at $16.25B valuation. SBF and FTX lost 98% or $74.7B And if the trend continues, this loss may grow to $150B< If debtors waited 3 more years, Sam Bankman-Fried would be the richest guy in Web3. Insane

Cerebras Systems Deep Dive: The Giant-Slayer or Just a Giant Chip? The Titan of Compute: Cerebras Deep Dive—Revolution or Illusion? Preface: If NVIDIA is building a wall of compute by carefully arranging thousands of “精致 bricks,” then Cerebras’s logic is: “I’ll just bring an entire Himalayan mountain.” In May 2026, following its successful NASDAQ listing and market valuation surpassing $50 billion, global investors turned their eyes to this “chip bigger than a face.” I. Company Profile: The Epic of the Compute Colossus Simplified Understanding: Imagine traditional chipmakers (like NVIDIA or TSMC) assembling puzzles—cutting out tiny pieces the size of a fingernail and connecting them with wires. Cerebras rejects this approach entirely. Their flagship product, the WSE-3 (Wafer-Scale Engine), is a single, uninterrupted 8.5-inch silicon wafer. Its Appearance: Other companies measure chips in millimeters; Cerebras measures them in cafeteria trays. Its Logic: Since communication delays between chips are the root cause of AI slowdowns, why cut them at all? Instead, let every core grow directly on one plate—face-to-face collaboration. II. The Spear and the Shield: Advantages and Disadvantages 1. Advantages (The Spear): Dimensional Efficiency Bandwidth Dominance: With no need for data to travel between separate chips, its memory bandwidth is 7,000 times greater than NVIDIA’s top-tier GPUs. When processing ultra-large models (e.g., GPT-5 level), inference speeds reach thousands of tokens per second—so fast the generation process becomes invisible to the naked eye. Simplified Programming: Previously, training large models required hundreds of engineers tuning cluster communications. For Cerebras, the entire machine is one giant chip—coding feels as simple as running on a single machine. 2. Disadvantages (The Shield): Backlash from Physical Laws Yield Challenges: A single defect on the entire wafer can render the whole board useless. Although Cerebras incorporates redundant cores, manufacturing demands extreme precision and comes at an exorbitant cost. Heat Sink Black Hole: Running this massive board generates heat equivalent to multiple electric stoves operating at full power. You’re not buying a server—you’re buying a beast with a fiery temper, requiring prohibitively expensive liquid cooling systems. III. Commercial Progress: From Lab to Compute Factory 2026 Latest Updates: OpenAI’s Secret Weapon: In early 2026, Cerebras signed a $10 billion agreement with OpenAI. OpenAI discovered that for certain specialized reasoning tasks, Cerebras’s architecture consumes significantly less energy than NVIDIA’s clusters. Condor Galaxy 3 Cluster: The supercomputer cluster co-developed by Cerebras and Abu Dhabi’s G42 Group has gone live in Dubai—marking its evolution from selling individual units to offering compute-as-a-service. IV. Financial Status and Outlook: Bread or Circus? Revenue Status: 2025 fiscal year revenue tripled compared to 2024, surpassing $1.5 billion for the first time. Though still dwarfed by NVIDIA, its gross margins are exceptionally high due to proprietary architecture and exclusive supply. Future Outlook: In late 2026, as enterprise AI (private deployment) demand surges, mid-sized giants unwilling to manage complex clusters are beginning to order Cerebras systems. Analysts project 2027 revenue could reach $5 billion. V. Overall Valuation and Competitive Dynamics 1. Valuation Logic: At its current $50 billion market cap, the price-to-sales ratio (P/S) is extremely high. Investors aren’t betting on current profits—they’re betting on “What if NVIDIA’s moat (CUDA) cracks?” 2. Competitive Landscape: NVIDIA: Still the undisputed ruler, with an unrivaled software ecosystem (CUDA). In-House Builders (Google/Amazon): Giants are all developing their own TPUs. Cerebras’s Moat: It is the only company pushing boundaries in “physical architecture.” If AI’s ultimate bottleneck lies in physical latency, Cerebras is the only solution. VI. Investment Risks: Dancing on a Razor’s Edge Supplier Dependence: It relies heavily on TSMC’s specialized packaging technology. If TSMC faces capacity constraints, Cerebras cannot secure even a single wafer. Software Ecosystem Anxiety: Engineers are deeply accustomed to NVIDIA’s ecosystem. Changing programmer habits is harder than changing physical architecture. Big Tech Counterattack: If NVIDIA reduces latency sufficiently through NVLink technology, Cerebras’s architectural advantage will be diluted. Conclusion: Chief Content Officer’s Final Words Cerebras is the “brutal aesthetics” of compute. It refuses to patch old systems—it overturns the table entirely, solving AI’s hunger with the most hardcore, direct physical approach possible. In one sentence: If you believe AI’s future lies in “extreme speed” and “extreme simplicity,” then Cerebras is the lone hero daring to charge against NVIDIA. But if you believe “ecosystem is everything,” then it remains merely an expensive, solitary scientific oddity. In 2026, Woodie has already bet—what about you?

🚨 Dan Ives targets $30,000 for Nasdaq within 6–9 months: AI IPO boom and chip demand are driving the narrative. The message is clear: “The AI cycle is still in the 3rd inning.” Momentum is strong, but expectations are now crowded on the trade side. 🟢 The AI cycle is still in its early stages 🟢 Chip supply < demand (strong pricing power) 🟢 The IPO wave is increasing liquidity 🟢 “Parabolic re-rating” scenario for Nasdaq

Sam Altman made a profit from Cerebras' IPO 🚀 The head of OpenAI earned about $30 millions after the AI chip manufacturer Cerebras went public. The company's shares soared from $185 to $350 on the first day of trading. The success of the partner is also beneficial for OpenAI itself: the company has a $10 billions deal and the right to purchase equipment until 2030. #AI | #OpenAI | #Cerebras | #IPO

The promoters have been around since 1893. They are so conservative that their debt is almost zero. While that’s good for safety, it makes you wonder if they are too scared to take the big leaps required to become a multi-billion dollar player. Also, promoter holding has dropped from 50.59% to 46.52% in the last year. If the captains are trimming their stakes, why should the sailors stay calm? Search for Amrutanjan Health Care Q4 FY26 on website