OpenAI's chip beats Nvidia's best

Hola Decoder😎

If someone forwarded this to you and you want to Decode the power of AI and be limitless, then subscribe now and Join Decode alongside 30k+ code-breakers untangling AI.

🔥 OpenAI’s custom AI chip beats Nvidia in its first tests

OpenAI has published the first benchmark results for Jalapeño, its custom inference chip built with Broadcom. The chip is designed specifically for running AI models rather than training them, and OpenAI says its early tests show significant speed and efficiency gains over Nvidia hardware.

The Decode:

Jalapeño beat Nvidia on speed - OpenAI says its 700-watt chip delivered responses up to 3.6x faster than Nvidia systems rated at 1,200 watts in its benchmark tests.

It also used power more efficiently - Jalapeño reportedly delivered up to 1.9x more work per watt, an important advantage as inference becomes a growing part of AI infrastructure costs.

AI helped design the AI chip - OpenAI says it used its upcoming Astra model and Codex during development, helping take Jalapeño from its initial design to manufacturing-ready silicon in just nine months.

OpenAI is keeping every chip - Jalapeño won't be sold externally because OpenAI says its internal demand is already enormous. Two more generations are being developed, with deployment beginning later this year and production ramping through 2027.

The bigger story is OpenAI starting to control the hardware underneath its models. Nvidia remains critical for training, but if OpenAI can move a growing share of inference onto faster, more efficient custom silicon, it could lower one of the highest costs of serving AI at scale

Together with Tatari

When Meta hits a ceiling, the smartest brands look beyond it.

Scaling digital used to be straightforward. Now brands hit a Meta Ceiling where reach stalls and CAC skyrockets. 

The brands still compounding aren't fixing it inside Meta. They're diversifying. 

Tatari is a TV buying platform that lets you run across linear TV, streaming, and direct publisher buys, all from one place.

You see exactly where every ad aired, so nothing is hidden, and nothing is wasted. Every dollar ties back to actual site visits, signups, and revenue not impressions that look good and prove nothing.  

• Aroma360 cut CPA by 80% and doubled ROAS from a $14K linear pilot
• Fiverr retargeted Super Bowl viewers on streaming and slashed CAC by 75%
• Saatva saw a 40% lift in branded search from their very first Tatari test

Every week you wait is a week someone in your category is pulling customers you could have reached first.

Book a free demo and get a launch-ready TV plan built around your numbers.

P.S. In NYC Oct. 29? Join Tatari at Forward for an afternoon on what’s next in advertising with Reddit CEO Steve Huffman, and marketers from Liquid IV, MANSCAPED, and more.

✨How to audit and fix your LinkedIn profile using Claude

Step 1: Paste your full profile - Copy everything, headline, about section, experience, skills, and paste it into Claude.

Step 2: Run the audit - "Audit this LinkedIn profile critically. What's weak, what's missing, what's generic, what would make a recruiter or potential client scroll past? Be specific."

Step 3: Rewrite section by section - "Rewrite my headline to position me as [role] who helps [target audience] achieve [outcome]. No buzzwords. Under 220 characters."

Then repeat for the About section, each experience entry, and your skills list.

Step 4: Generate connection request templates - "Write 5 personalized LinkedIn connection request templates for reaching [target person type]. Each under 300 characters. Sound human, not salesy."

Profile audited, rewritten, and ready to attract the right people in one session.

🖥️ Perplexity brings its AI agent onto your computer

Perplexity and Nvidia have launched Portable Computer, a local version of Perplexity’s Computer agent that can work with files directly on your machine. The pitch is simple: run capable AI agents without constantly sending your private data to the cloud.

The Decode:

The AI can run completely locally - Users can choose Qwen 3.8 27B or Perplexity’s PPLX 27B, keeping on-device tasks and files on their own hardware.

Local agent work is free to run - On-device tasks consume no Perplexity credits, and the agent can connect with popular apps and access cloud models only when the user approves it.

You can still call bigger cloud models - Portable Computer can reach more than 15 cloud models when local models aren't enough, giving users a mix of privacy and additional computing power.

It starts on Nvidia’s DGX Spark - The agent installs with one click on Nvidia’s $4,699 desktop supercomputer, with support for certain PCs expected later.

AI agents are usually tied to someone else’s cloud, which creates costs and privacy tradeoffs every time they work with your data. Running the agent locally changes that equation. As capable models get smaller, personal computers could increasingly become where autonomous AI actually lives and works.

Together with Playbook Pro

Your Next 1,000 Customers Are Already Here

Pouring more money into ads won’t fix flat growth. The real leverage is hiding inside the customers you already have, and every day you wait, that momentum leaks away.

Networked Growth Engines is the 26-page field guide built from 200+ hours of research and real brand wins. It gives you ready-to-use systems you can plug in this week, no giant team or budget required. With it, you can:

🔥 Spark self-propelling communities using the Community Building Playbook, exactly how Nyfter built a 500,000-member Discord that drives product innovation.

📈 Design referral programs that fund themselves with the Referral ROI Calculator, the same loop that powered 34% of Blue Apron’s customer acquisition and delivered a 17.3× ROI for Latico Leathers.

🤝 Form trust-based partnerships fast using the Partnership MoU and 50-template Outreach Library to unlock new audiences without paid ads.

Your current customers are the growth engine. 

Download the guide today and start compounding growth before another week of ad spend vanishes.

🏆 Tools you Cannot Miss:

🔥 Flare – A graph-first IDE that maps your codebase visually, helping developers understand relationships and work with AI coding agents more effectively.

📊 Jotform AI Data Assistant – Turns collected form data into AI-powered insights and actions, letting you analyze responses without manually digging through submissions.

🛡️ Hacktron Automations – Finds real software vulnerabilities, validates them, and automatically generates tested fixes your engineering team can review.

🎧 Assistly – Provides real-time AI guidance during meetings through a private, bot-free overlay that doesn't appear in screen shares or recordings.

🧪 Agnost AI – Tests AI agents beyond standard evals to uncover real-world failures and behaviors that traditional evaluation pipelines can miss.

🚀 Quick Hits

⚡ US data centers are driving a gas power boom. Proposed gas-fired power capacity tied to data centers nearly doubled in the first half of 2026, with Texas accounting for almost a third of the planned capacity. 

🧠 Claude cowork can now remember your chats. Cowork now shares Claude’s existing memory, letting it carry context across tasks without repeated explanations, while sensitive information stays excluded unless users explicitly opt in. 

🏗️ OpenAI loses its head of data centers. Chris Malone has left after roughly 17 months, adding to a wave of senior departures as OpenAI reorganizes the infrastructure team responsible for its massive data center buildout. 

💰 Anthropic could pitch a $30 trillion revenue opportunity. The Claude maker is reportedly expected to tell investors its potential revenue opportunities exceed $30 trillion, surpassing the $28.5 trillion figure SpaceX presented to investors. 

📰 Google’s official blog covers the company’s latest launches, product updates, AI developments, research, and company news across products including Search, Gemini, Android, Chrome, Pixel, and more. 

💃 What's trending on socials 

💧 Ninja Launches a Smarter Brita Rival - The $199 HydraSense filters 80+ contaminants, dispenses water six times faster than a standard Brita, and automatically tracks usage and filter life.

🚀 Monaco Launches Its AI Sales Platform Publicly - After onboarding hundreds of startups in beta, Monaco says customers added 16 percentage points to monthly revenue growth and tripled their meeting volume on average. 

✂️ This Free AI Tool Could Replace Paid Video Clippers - OpenShorts clips YouTube or local videos with no watermarks or limits, is self-hosted and open source, and supports AI models like Claude.

🧩 Kimi K3 Turns 2D Schematics Into Interactive 3D - This coded app transforms a flat schematic into a 3D visualization, with the generated tank easily replaceable with entirely different objects. 

🎬 Why Polished AI Anime Still Feels Wrong - A frame-by-frame critique of a Seedance 2.5 clip highlights inconsistent construction, drifting details, continuity errors, and movement that lacks convincing weight.

🤳AI Nugget of the Day

Thanks for Decoding with us🥳

Your feedback is the key to our code! Help us elevate your Decode experience by hitting reply and sharing your input on our content and style.

Keep deciphering the AI enigma, and we'll be back with more coded mysteries unraveled just for you!