Hi, it’s Marc.
Welcome to +485 new readers since our last update. A lot has happened in the last two weeks. Last week, we held X spaces with some of the leading projects & VCs on the state of the market / crypto x AI. Today we cover:
DeepSeek
Top reads
Top quotes
Alpha
Let’s jump in.
DeepSeek R1: AI's "Sputnik Moment"
Last weekend, DeepSeek R1 dropped. While Western giants pour billions into bloated AI orgs and GPU clusters, an under-the-radar Chinese team built a world-class model for less than the salary of one Meta AI exec.
Let that marinate:
Trained on $5.5M (yes, million) while competitors burn billions. DeepSeek-R1 delivers OpenAI-o1 level intelligence at 90% less cost.
Outperforms GPT-4, Claude 3.5, and even Llama 4 on key benchmarks.
Open-source and commercially licensed—threatening every closed, overpriced model out there.
In short: DeepSeek AI questions Western AI dominance because of how efficiently it achieves comparable results.💥
3 take-aways
1. Massive Malinvestments in the U.S./West:
Throwing money at GPUs isn't innovation. The West wasted billions on brute-force scaling & ignoring efficiency, while DeepSeek-R1 proves smarter algorithms > bigger budgets.
The US poured billions into funding closed-source models (@OpenAI & @Anthropic have raised almost $20 billion) which are suddenly outcompeted by $5M open-source models.
If that's not all: Most startups are wrappers for GPT-4 or Claude. R1’s capabilities will commoditize 80% of them overnight.
2. Panic at big tech:
Meta’s GenAI team is scrambling. Llama 4 is already obsolete. Engineers are dissecting DeepSeek to copy whatever they can.
Leadership is terrified—how do they justify billion-dollar orgs when DeepSeek’s entire training budget was less than some of their execs’ annual salaries?
3. Nvidia is a house of cards:
The West bet billions on brute-force scaling. But DeepSeek-R1 proves smarter algorithms > bigger budgets.
When 5M models beat $500M ones, Jensen’s “compute is all that matters” mantra starts sounding like joke.
Marc Andreessen calls it “AI's Sputnik moment".
DeepSeek isn’t just a model—it’s a wake-up call.
More links on DeepSeek:
Top AI Agents by mindshare
Some agents you want to keep an eye on:
$AIXBT ($658M): A crypto-savvy AI that scans markets and social media to deliver trading insights and alpha opportunities.
$GRIFFAIN ($508M): Creates AI tools that make AI agents smarter and more efficient, like enabling them to trade on DeFi protocols.
$ ARC($372M): A modular toolkit for creating AI bots that are highly customizable, especially for crypto ecosystems.
$SWARMS ($171M): Focuses on teams of AI bots (“swarms”) that work together to tackle big challenges.
$ALCH ($97M): Simplifies building AI apps and tools, letting developers create cool things quickly.
$POLY ($15M): A nimble, lightweight AI for straightforward trading and asset management
Top Quotes
“We're not just building more chatbots. we're looking for ideas that explore how AI can make web3 more intelligent.” – AI Rig Complex
“Agents are the new SaaS” – @AtomSilverman (AgentOPsAI)
“Search is UI on the web, and since AI is eating UI, AI will eat search.” – Naval
Top Reads
Google’s “Chain-of-Agents” paper. Link
The ultimate crypto AI primer. Link
How to position yourself in this agentic cycle. Link
DEFAI is the new DEFI. Link
A deep dive into ARC Rig complex. Link
11 Hidden Trends Reshaping the AI Agent Ecosystem. Link
Other Alpha:
@aicceleratedao: The project faced backlash after 100% of its tokens went to 245 insiders, some of whom cashed out after a price spike—triggering accusations of unfair practices and rug-pull fears. To rebuild trust, Aiccelerate promised a vesting plan for allocations. Still, doubts linger over its governance, tokenomics, and transparency, raising bigger questions for the DeFi space.
Virtuals started to buy back tokens:
The ai agent platform plans to use $VIRTUAL to buy and burn over 10k agent tokens from post-bonding trading revenue. A higher buyback pressure can lead to supply shock.
Meanwhile, the platform’s revenue imploded. Down 99% from ATH.
AI Rig Complex is gaining traction: arc emerged the second most active web3 Ai project on GitHub after Eliza (ai16z) and reached 2440 stars, 205 forks and 212 contributions. It also ramped up partnerships with five projects including Abstract and Arbitrum to bring EVM compatibility to the framework. arc project is currently hand-selecting projects to join its ecosystem via the Handshake program
$swarms is marching forward: The project did launch its roadmap tagged ” Building the Foundation of the Agentic economy.” It plans to integrate the $swarms coin into its marketplace, launch $swarms cloud to host agent decentrally and enable agent tokenization and investment amongst others.
DeFAI gained a lot of traction. In general, DeFAI landscape can be categorized into 4 major sectors:
AI Abstraction
Autonomous Portfolio Management & Yield Optimization
Market Analysis & Prediction
DeFAI Infrastructure & Platforms (Wallet etc.)
Holoworld's flagship AI agent launchpad is now live on Solana.
That’s it for now.
– Marc
Thaks Marc!