• AI Valley
  • Posts
  • OpenAI's first AI chip outperforms Nvidia

OpenAI's first AI chip outperforms Nvidia

PLUS: Anthropic thinks AI is a $30T market

Author

Barsee
26 Aug

Together with

howdy, it’s Barsee again.

happy wednesday, AI family, and welcome back to AI Valley.

here are the biggest things worth knowing today:

  • OpenAI’s custom chip is beating Nvidia in early tests

  • Anthropic thinks AI is a $30T market

  • Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

VOICES

Courtesy: Voices

If you're fine-tuning a speech model, you've probably hit this wall: DNSMOS gives you a score, but it doesn't tell you whether the data behind that score is actually right for your model. Treat it as a pass/fail gate and you'll end up training on audio that looks clean on paper but drags down real-world performance while good source data gets tossed for no reason.

Voices' CTO DJ Jalali (with the team's senior audio and voice data engineers) just published a breakdown of the four-step calibration framework they use internally to set model-specific quality thresholds instead of trusting the raw DNSMOS number. It also covers where DNSMOS breaks down and how Voices validates audio for custom datasets at scale.

*This is sponsored

THROUGH THE VALLEY

1/ OpenAI’s custom chip is beating Nvidia in early tests - OpenAI published the first benchmarks for Jalapeño, its custom 700W inference chip built with Broadcom. In OpenAI’s InferenceX tests, it delivered 1.5–1.9x more AI work per watt and 1.7–3.6x lower end-to-end latency than Nvidia’s GB200 and GB300 systems across GPT-OSS 120B, DeepSeek R1 670B and Kimi K2.5 1T.

The interesting thing is OpenAI built it in just nine months, using its own models and Codex during development. It’s not trying to replace Nvidia either. OpenAI still needs Nvidia for training, and Jalapeño is mainly about making inference cheaper and faster.

Two more generations are already in the works, with Jalapeño starting to roll into OpenAI’s data centers later this year and ramping through 2027. So yeah, OpenAI is slowly becoming a chip company too.

Some interesting stuff related to this:

  • Why is every AI lab suddenly trying to build their own chips?

2/ Anthropic thinks AI is a $30T market - Anthropic is reportedly telling IPO investors that its total addressable market is more than $30T, based on the amount of work AI models could theoretically do. That’s even bigger than SpaceX’s $28.5T TAM from its IPO filing.

Anthropic’s own numbers are a bit more interesting. Its revenue run rate has already passed $65B, and it’s projecting $190B-$200B in revenue for 2028.

Obviously, TAM numbers are mostly a way to tell investors “this can get really, really big.” But $30T is a pretty insane number. For context, the entire tech sector in the S&P 1500 made about $2.4T last year.

So either AI really is about to eat a huge chunk of the economy, or Anthropic has found the biggest TAM slide ever.

TRENDING TOOLS

  • ChatGPT Work - it can now use its computer and browser to sign in to websites on web and mobile, without ChatGPT ever seeing your username or password

  • Perplexity and NVIDIA launched Portable Computer - a fully local AI agent (software that completes tasks, not just chats) with zero token costs

  • OpenWorker - an open source agent that doesn't just chat but completes tasks on your laptop

  • RenameClick - an AI-powered desktop app that intelligently renames and organizes files by understanding their real content

  • Claude's memory - now works the same across chat and Claude Cowork

  • Theojaffee - visual explorer of every company in the world worth more than $10 billion

  • S1 - a new foundation model that learns from one example

  • Icon - pivots from AI admaker to human-created ads with $30M raise

  • Grok Voice Think Fast 2.0 - Now #1 on the Artificial Analysis Speech-to-Speech Index

  • Soloop - hands you an AI founding team that plans, builds, and markets your product

  • Higgsfield Ad Multiplier - Take your top-performing creative from Meta and create endless versions: new characters, locations, outfits, text and dubbing

  • Stickers in ChatGPT - build sticker packs from photos to send on iMessage

  • Claude Code Templates - a massive new open-source library with 100+ ready-made agents, skills, and templates you can install instantly

  • Twin - an AI-powered company builder that helps you go from idea to execution faster

  • Manus Skills - turns Manus from a generalist into any expert you need, on demand

  • Kimi K2.5 - Moonshot AI’s powerful open-source model built specifically for agent workflows

  • Wan 2.5 AI - an open-source video generation model that turns text or images into high-quality videos with synchronized audio and smooth frame-level motion

WHAT I'M CONSUMING

THE VALLEY GEMS

What’s trending on social today:

THAT’S ALL FOR TODAY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.