News

Tech Updates #2 , OpenAI Chip, Claude Theft, HL2 in Browser, Deno 2.9, RubyLLM

OpenAI makes its own chip (Jalapeño), Anthropic accuses Alibaba of stealing Claude, Half-Life 2 runs in browser, Deno 2.9 with Desktop API, RubyLLM 4.1K stars, and IBM 0.7nm chip. Tech Updates #2 , June 25, 2026.

Muhammad Amien·Jun 25, 2026·6 min read

Tech Updates #2 , OpenAI Chip, Claude Theft, HL2 in Browser, Deno 2.9, RubyLLM

Two days in a row OpenAI makes big headlines. Yesterday they announced their first custom chip , Jalapeño. Today? Anthropic's accusation against Alibaba regarding the theft of the Claude model. On the other hand, someone managed to run Half-Life 2 in a browser, Deno released version 2.9 with a desktop API, and IBM made history with a 0.7 nanometer chip. Tech Updates #2 , let's go.

OpenAI "Jalapeño" , First Custom Chip, Built with Broadcom

From TechCrunch , OpenAI finally has its own chip. Its name is Jalapeño, built in partnership with Broadcom, specifically designed for AI inference workloads. Until now, OpenAI was fully dependent on Nvidia GPUs , this chip is the first step towards hardware independence.

Why this is super important: (1) inference costs will drop significantly, (2) GPT model performance will be more optimal because the chip is tailored, (3) reducing dependence on Nvidia whose supply is always a bottleneck. Custom chips are a strategic moat , Google has TPU, Amazon has Trainium, and now OpenAI has Jalapeño.

HN: **788 points**, 451 comments , the hottest thread today.

Status: ✅ Officially announced , the chip is still in the production preparation stage.

Anthropic Accuses Alibaba of Stealing Claude Model , The Largest IP Theft Case

From Reuters , Anthropic formally accuses Alibaba of illicit extraction of Claude model capabilities. This is the largest AI model IP theft case so far , and it happens amid increasingly heated US-China tensions.

The accusation is serious: Alibaba is suspected of extracting Claude through suspicious API access patterns, then using it to strengthen their internal models. If proven, this will be a huge precedent: frontier model security, export controls, and international AI regulation.

The HN thread exploded with 1,119 comments , debates about open-source safety, model security, and AI ethics. For developers: this is a reminder that API access doesn't mean safe from model extraction. Rate limiting and usage monitoring become increasingly crucial.

HN: **687 points**. Status: ⏳ Investigation ongoing.

Half-Life 2 Runs in Browser , WebAssembly + WebGPU Maturity

This is crazy. Someone managed to port Half-Life 2 , Valve's legendary AAA game , to the browser. Open hl2.slqnt.dev, and play right away. No download, no install needed. Everything runs on WebAssembly + WebGPU.

Technically impressive: WebAssembly compiles native C++ code to browser bytecode, WebGPU handles 3D graphics rendering that used to only be possible via native OpenGL/DirectX. This is concrete proof that the web platform is mature for applications that were previously impossible without native code.

The implications are huge: game preservation (old games can be played forever in the browser), cloud gaming without proprietary tech, and heavy application distribution via the web. Imagine: you send a link to a friend, and they instantly play an AAA game.

HN: **531 points**. Status: ✅ Live , can be tried now.

Deno 2.9 , Desktop API, CSS Imports, and Supply Chain Security

From deno.com/blog/v2.9 , Deno 2.9 arrives with features that make it even more serious as a universal runtime:

Deno Desktop, native desktop API: window, tray, menu, notifications. Deno can now replace Electron/Tauri for cross-platform desktop apps.
CSS Module Imports, import styles from './styles.css' directly in JS/TS. Frontend devs who are used to Vite will be familiar with this.
Supply Chain Security, minimum dependency age enforced by default, no-downgrade trust policy. The most secure in its class.
Snapshot Testing, built-in in deno test. No external library needed.

What makes Deno even more compelling: input-based task caching performance, concurrency control, and increasingly solid Node.js compatibility (Web Crypto API). If you haven't tried Deno yet, version 2.9 is the right moment.

HN: **60 points**. Status: ✅ Release , deno upgrade now.

RubyLLM , One Ruby Framework for All AI Providers, 4.1K Stars

From rubyllm.com , the Ruby/Rails ecosystem finally has a serious unified AI framework. RubyLLM supports all major providers: OpenAI, Anthropic, Gemini, Bedrock, DeepSeek, Mistral, Ollama, OpenRouter, Perplexity, xAI , and only 3 dependencies.

Got **4,180 stars** in a short time , a clear indication that there is a huge demand that has been underserved. Its features are complete: chat, image generation, embeddings, agents, streaming, structured output, audio transcription, moderation, and Rails integration via Active Job.

Created by Carmine Paolino. For Indonesian developers who still use Rails (and there are many!), this is a game-changer. You don't need to switch to Python just for AI integration.

HN: **439 points**. GitHub: **4,180** ⭐. Status: ✅ v1.16.0 , production-ready.

IBM Makes History , The World's First 0.7 Nanometer Chip

From IBM Newsroom , IBM demonstrated the world's first sub-1 nanometer chip technology: 0.7 nanometers, or 7 angstroms. It uses a revolutionary transistor architecture called "Nanostack".

Context: so far the physical limit of silicon has been considered to be around 1nm. IBM broke that barrier. Moore's Law is still alive , at least for now. The implications directly affect AI acceleration, mobile devices, data centers. Smaller chips = more power-efficient = faster.

This article is still underreported (only 12 HN points), but its significance is huge. This is news you will see again 5 years from now as a milestone.

HN: **12 points** (rising). Status: 🔬 Research demonstration , not yet in production.

Hello-Agents + Awesome-Evals , Trending AI Agent Resources

Two trending GitHub repos on the topic of AI Agents:

Hello-Agents(131 ⭐) , a comprehensive tutorial for building AI Agents from fundamentals to production-grade multi-agent systems. Covers deep learning, RAG, LangChain, OpenAI. Suitable for beginners.
Awesome-Evals(258 ⭐) , a curated library of the best resources for building & evaluating AI agents: papers, blogs, talks, tools, benchmarks. Maintained by BenchFlow. Evaluating agents is the hardest problem in AI 2026 , this is a must-have reference.

Both are fresh , Hello-Agents was just uploaded on June 23, immediately trending. A clear indication that AI agent education is gaining popularity.

Status: ✅ Open source , accessible now.

Others Worth Checking Out

Some news that didn't make the main feature but are important:

Cloudflare OAuth for All(302 HN pts) , OAuth becomes a standard feature, not enterprise-only.
Ford Rehires 350 Engineers After AI Fails(476 HN pts) , AI quality inspection failed, Ford brings back human engineers. A reminder that AI is not a silver bullet.
LuaJIT 3.0 Syntax Extensions(208 HN pts) , Proposal for modern syntax extension for Lua.
Apple Price Hike 20%(229 HN pts) , The impact of US-China trade tariffs on consumer tech prices
Herculaneum Scroll First Full Read, AI successfully reads the entire ancient scroll from Pompeii for the first time.

Bottom Line

Two big themes today: AI hardware independence (OpenAI chip, IBM sub-1nm) and AI security (Anthropic vs Alibaba). Both show that 2026 is not just about models getting smarter , but about who owns the infrastructure and how to secure it.

On the developer side: Deno 2.9 makes desktop apps more accessible, RubyLLM bridges the Rails-AI gap, and WebAssembly/WebGPU is mature enough to run AAA games in the browser. If you're a developer, there's no reason not to explore all three.

See you in Tech Updates #3. Stay curious.