Skip to main content
Back

SSI secures $5B NVIDIA deal , Kimi K3 ships 2.8T params , OptMem add

NVIDIA backs SSI compute push, Kimi K3 ships with corp lockout, MCP server cuts token costs 95%, 3B model beats Qwen 9B

July 28, 2026 6 min read

Hey,

Today's theme is simple: compute is the new currency, and everyone is racing to print more of it.

SSI just broke two years of silence, not with a product or a paper, but with a $5B NVIDIA deal and a 10x compute upgrade. Meanwhile Moonshot ships FlashKDA, squeezing 2.2x more speed out of existing GPUs. Two very different bets, same obsession.

More power, faster inference, better agents. We're not building software anymore. We're building engines.
In Partnership with · partner_image
Author photo
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
Summary

Read time: 6 min 34 sec

Top News

▸ NVIDIA invests in SSI to 10x its compute within 12 months

WorkOS

▸ Manage your auth platform from any AI agent

Top Model

▸ Moonshot AI ships Kimi K3 with a non-commercial license that locks out big companies

Anyscale

▸ Ray Summit + vLLM Conference

Top Repo

▸ Developer ships plug-and-play infinite memory prompt for any AI agent

Signals

1. Kimi open-sources attention kernel that cuts prefill time by up to 2.2x
2. Augment Code: Your agents don't share memory, context, or work. Augment Code's Agentic SDLC shows how to fix that
3. This open-source MCP server cuts AI coding token costs by 95% across 48,000 developers
4. LangGraph paper shows how graph engineering replaces messy prompt control flow
5. New 3B model beats Qwen 9B and Gemma 12B on agentic tasks
6. Kimi open-sources the distributed agent training system behind Kimi K3
Top News
NVIDIA invests in SSI to 10x its compute within 12 months
12,169 Likes
NVIDIA invests in SSI to 10x its compute within 12

Ilya Sutskever's Safe Superintelligence (SSI) just ended two years of total silence. No product. No published research. Just a team quietly working on one thing: building a safe superintelligence.

Now NVIDIA is writing a massive check. The investment stretches into multiple billions, with Bloomberg reporting a figure of $5 billion.

But the money is almost secondary. Here's what actually matters:

  • SSI gets access to NVIDIA's next-gen Vera Rubin GPU platform, letting them increase their compute by an order of magnitude, meaning 10x more raw processing power.
  • Both companies will collaborate on advancing NVIDIA's current and future hardware, using SSI's research insights.
  • This adds an NVIDIA GPU path on top of SSI's earlier Google Cloud TPU setup.

SSI has spent two years quietly pursuing a new research direction for powerful, aligned AI, and NVIDIA only joined after getting rare access to that closely guarded work. More compute means that research can now actually scale.

Read more
forward →
Presented by WorkOS
Manage your auth platform from any AI agent

Debugging SSO, managing users, adjusting auth policies, configuring branding: every configuration task has lived behind a UI that only a human can drive.

The WorkOS MCP server gives agents the same access as your dashboard login. Hundreds of operations, discoverable at runtime. Connect in one command via OAuth, with scoped tokens instead of a master API key.

Pass a screenshot of your marketing site and ask your agent to match the login page. If a human had to do it before, an agent can do it now.

Connect your agent
partner with us →
Top Model
Moonshot AI ships Kimi K3 with a non-commercial license that locks out big companies
3,235 Likes
Moonshot AI ships Kimi K3 with a non-commercial li

Moonshot AI just dropped Kimi K3, a massive open-weight model with 2.8 trillion parameters that you can actually run yourself. Think of it like getting access to a frontier-level AI without paying per API call.

Here is what makes it interesting for you:

  • It reads text, images, and video natively, no extra setup needed
  • 1 million token context window, meaning you can feed it entire codebases or long documents at once
  • Scores 88.3 on Terminal-Bench, competing directly with Claude and GPT-5
  • Works with vLLM, SGLang, Ollama, and Docker out of the box

Now the catch: the license looks open but has teeth. Free for personal projects and small companies. But if your company earns over $20M per year, you need a commercial deal. Big platforms with 100M+ users must also display the Kimi K3 branding. Read the license before shipping to production.

Read more
forward →
Presented by Anyscale
Ray Summit + vLLM Conference

Catch sessions and network with the teams building at the frontier of AI:

  • Get trained on Ray, vLLM, DeepSpeed on other top oss frameworks
  • 60+ sessions on RL, Physical AI, Foundation models for life sciences & more
  • Keynotes from CTOs and Heads of AI across industries

Get 50% off with code RS26SUMMITMKTG50

Get your ticket
partner with us →
Top Repo
Developer ships plug-and-play infinite memory prompt for any AI agent
3,024 Likes
Developer ships plug-and-play infinite memory prom

AI agents forget everything when a session ends. That's the core problem. OptMem is a plug-and-play memory system that fixes this with just two text files and six shell commands.

Here's the clever part. Every memory gets appended to LOG.txt and never deleted. But reading all of it would be too slow. So OptMem builds a tree of compressed summaries, where old memories get squished together and recent ones stay verbatim. You always read a fixed 208-line document, no matter how many memories exist.

What you can do with it:

  • Drop it into any agent harness: Claude Code, Codex, or a plain shell
  • Run memo wake at session start to reload the agent's full identity
  • Use memo note, memo sleep, memo recall, and memo forget to manage memories
  • Search raw history anytime with regex, even after compression

Install is three lines. Identity persists across models and vendors. Nothing is ever lost.

Read more
forward →
Signals
1
Kimi open-sources attention kernel that cuts prefill time by up to 2.2x
2,858 Likes
2
Your agents don't share memory, context, or work. Augment Code's Agentic SDLC shows how to fix that
Presented by Augment Code
3
This open-source MCP server cuts AI coding token costs by 95% across 48,000 developers
2,230 Stars
4
LangGraph paper shows how graph engineering replaces messy prompt control flow
742 Likes
5
New 3B model beats Qwen 9B and Gemma 12B on agentic tasks
228 Likes
6
Kimi open-sources the distributed agent training system behind Kimi K3
2,690 Likes
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 300,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.

Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws.
Looking to promote your company, product, service, or event to 300,000+ AI developers?
Work with us