
GM {{first_name}} !
If you ask me, we’re in the “Yahoo Directory” phase of AI. It’s early, messy, and full of potential. It’s all greenfield.
Doesn’t this bring back memories?

In crypto AI, the early wins won’t come from going directly against the incumbents. They’ll come from fixing what’s already broken (just like stablecoins vs traditional banks).
Do you really want to build another social app, ChatGPT clone, or cloud service? Or do you want to build where no one else is looking?
A great quote we saw from Aurelius:
"the bottleneck is no longer access. It is perception. It is knowing what matters and when.
The edge no longer belongs to the most connected. It belongs to the most perceptive. The person who can reduce complexity into a punchline of truth"
AI levels the field. And now the real edge still comes from human intuition. The ability to filter chaos, add context, and see what others miss.
In other news: SoftBank’s Masayoshi Son is pitching his boldest vision yet: a $1 trillion plan to turn Arizona into a global AI and robotics hub.
SoftBank plans for a $1 trillion AI and robotics complex in Arizona with TSMC tie-up, marking its largest industrial AI push yet.
Project Crystal Land in Arizona.
This is in addition to SoftBank’s $19 billion stake in the $500 billion Stargate AI project in USA.
according to
— #Rohan Paul (#@rohanpaul_ai)
7:45 PM • Jun 20, 2025
The project, dubbed “Crystal Land,” would be a tech manufacturing zone filled with AI-powered factories and SoftBank-backed startups like Agile Robots. Son wants it to be America’s answer to Shenzhen.
He’s trying to rally support from TSMC, Samsung, and U.S. officials, asking for tax incentives and political backing. So far, there’s no official buy-in, and TSMC isn’t exactly jumping at the offer.
To make it happen, Son is tapping into infrastructure-style financing while deploying capital elsewhere: $30B into OpenAI, $6.5B to acquire chipmaker Ampere, and billions into AI data centers around the world.
It’s a massive bet. If it works, the U.S. could become the heart of global AI hardware. If not, well he’s just going to burn a lot of cash.
Hope everyone who made it to Permissionless had a good one.
We’ve got a big week coming up at Chain of Thought. Some exciting stuff is on the way. Don’t want to spoil anything but watch for it!


☀️ Sahara AI just launched $SAHARA. Community claims opened June 26.
🤝 Warden Protocol is teaming up with Venice to bring censorship-resistant, verifiable AI agents on-chain, spanning more than 100 blockchains.
🚀 CESS Network’s $CESS token is now live and trading on major platforms including Binance Alpha, Bitget, and KuCoin.
Meanwhile, the global AI training data market is projected to explode from $4.85B in 2024 to $34B by 2033. That’s a 24% annual growth rate, driven by the unrelenting demand for smarter models.
What’s fueling it? AI needs fuel. High-quality, labeled data is non-negotiable, whether the task is diagnosing cancer, predicting markets, or navigating traffic.
Multimodal datasets, those blending text, image, and audio, are gaining traction fast. They're key to training more capable, context-aware AI.

Source: x.com/josh_benaron
Asia-Pacific is leading the charge. China and India are scaling national AI infrastructure with government backing, while tech giants like Microsoft, Google, Amazon, Scale AI, and Appen compete to control the data pipelines.
Whoever owns the best datasets wins the arms race.

🚗 NATIX Hits 250K+ Registered Drivers

NATIX is building a decentralized map by turning everyday cameras—phones, dashcams, CCTVs—into real-time sensors. These devices collect traffic, road, and parking data locally, using edge AI to strip out personal info before anything leaves the hardware.
No billion-dollar satellites, no Google-level infrastructure. Just a swarm of drivers feeding environmental intel directly from the streets.
So far, over 256,000 drivers have contributed. Together, they’ve mapped more than 184 million kilometers and logged over 971 million data points. The network spans 171 countries, and >300M $NATIX tokens have been burned.

Source: docs.natix.network
To join, just mount your phone on the dash and run the Drive& app. Tesla owners can plug in the VX360 to mine their own dashcam data. That data powers a Dynamic Map usable by devs, cities, and apps that need geospatial awareness.
NATIX is also running a Bittensor subnet called StreetVision. It trains decentralized AI models on the crowd-collected video to detect road work, traffic jams, and driving behavior. Think navigation upgrades, smarter AVs, and an open mapping protocol powered by humans, not megacorps.
It’s a crowdsourced, privacy-first data economy, hiding in plain sight.

🧊 Nexus Launches Their Testnet III
Earlier this week, Nexus kicked off Testnet III with an initial rollout across eight high-activity regions, including the U.S., India, and Nigeria. A broader global expansion is set to follow in the coming weeks.
Testnet III has begun.
Today, we're rolling out Testnet III across the United States, Vietnam, Nigeria, Russia, India, Indonesia, the United Kingdom, Hong Kong SAR and Mainland China.
Tomorrow, the world.
Learn more: nexus.xyz/testnet?utm_so…
— #Nexus (#@NexusLabs)
5:34 PM • Jun 23, 2025
Nexus is building a Layer 1 chain engineered for the AI-native internet.
It wants to be a global supercomputer, not just another smart contract platform.
Anyone can contribute compute using Nexus OS (a one-click browser experience) or the CLI for more advanced setups. In return, users generate zero-knowledge proofs and earn NEX Testnet III Points. These points will convert into tokens, making it easy for anyone with spare compute to participate and earn.

Source: nexus.xyz
The core innovation is the Universal Proof. It compresses massive-scale computation into a single, cryptographically verifiable artifact. This allows AI agents to transact and coordinate on-chain without relying on centralized gatekeepers.
Nexus uses a three-layer architecture: Execution for compute, Consensus for coordination, and Storage for verifiable data. It’s designed to handle billions of secure, auditable AI interactions per second.
The bigger vision is a verifiable internet. One where every AI action is provable, traceable, and secure by default.


Fundraises
Inference Labs pulled in $6.3M across four rounds to build verification infrastructure for decentralized intelligence markets. Their most recent round, led by Native Capital, sold out in under a week.
[Web2] OpenRouter just secured $40M from a16z and Menlo to scale its LLM routing stack. The platform already routes across 450+ models for over a million developers and has processed more than $100M in GMV.
[Web2] Delphi raised $16M from Sequoia to transform expert knowledge into AI-native minds. These adaptive chatbots are already helping coaches, creators, and educators run always-on, personalized audience engagement.
Project Updates
Prime Intellect just dropped SYNTHETIC-2, a massive open reasoning dataset powered by its decentralized inference stack. Anyone with a GPU can now join in and push forward verifiable, permissionless AI training.
Gensyn unveiled GenRL, the modular core behind RL Swarm. It delivers fully distributed, fault-tolerant reinforcement learning without a central controller and supports 100+ environments via Reasoning Gym.
Story Protocol and OKX Ventures launched a $10M fund to back teams turning intellectual property into on-chain, programmable assets, treating IP as the next big real-world asset class.
0G Labs opened validator access on its Galileo testnet. Node operators can now help secure the network, process transactions, and shape the infrastructure for decentralized AI.
ChainOpera AI is live with a full-stack platform: ChatGPT terminal, agent builder, GPU layer, and L1 protocol. It’s already racked up 1M DAUs and 150K paying users.
OpenGradient released MemSync, a portable AI memory layer for instant personalization across apps. It’s open-source, user-owned, and crushing benchmarks.
Worldcoin launched Deep Face, a real-time deepfake detector powered by World ID. It lets users instantly confirm they’re talking to a human, not an AI clone, with no personal data leaked.
MOONBERG Terminal is live, offering AI-powered trading tools with 700+ data points, advanced screeners, and full UI customization. Early users earn perks ahead of the token launch.
Gradient launched Parallax, the first fully distributed inference engine for LLMs. It serves models peer-to-peer across GPUs and Apple Silicon, turning AI compute into a global, verifiable network.
Rayon Labs added a new free tier for Chutes with a 200-request cap. Paid plans now start at just $0.30 per million tokens, matching OpenRouter’s floor.
AI Agents
DashX just launched Dobby AI, the first AI portfolio agent built for Web3, in collaboration with Gaia. It manages wallets, trades, and adapts in real time using high-precision models and live data feeds.
Project Zero unveiled Zero Chat, an AI copilot that lets users track wallets, analyze DeFi flows, and automate on-chain actions.
Alchemist is expanding its ecosystem with $ACID, a new product-layer token powering its AI Browser Assistant for automating both Web2 and Web3 workflows. 20% of the $ACID supply is reserved for $ALCH holders.
Hive Intelligence went live on Sei Network, offering developers an API for real-time, LLM-ready blockchain data. It’s a major unlock for building AI-native apps within the Sei ecosystem.
Wander launched Agents, a dashboard-native tool that automates AO yield conversions and LiquidOps deposits. Set it once and let it optimize your earnings.
Web2 AI
Warp has launched Warp 2.0, the first all-in-one platform for agentic software development, featuring top-ranked coding agents, multi-threaded tasking, and deep context workflows for devs.
Mistral has released Small 3.2, a lightweight update to Small 3.1 that improves instruction following, reduces repetition errors, and enhances function calling reliability.
Google launches Gemini CLI, an open-source terminal agent with a 1M-token context window, real-time search, MCP support, and the biggest free tier in AI coding tools.
🔥 Our Weekly Top 5
Tesla launches the Robotaxi
Super congratulations to the @Tesla_AI software & chip design teams on a successful @robotaxi launch!!
Culmination of a decade of hard work.
Both the AI chip and software teams were built from scratch within Tesla.
— #Elon Musk (#@elonmusk)
7:57 PM • Jun 22, 2025
ChatGPT downloads on the App Store = TikTok, Facebook & Instagram COMBINED.
Worldwide iPhone App Store downloads over the last 28 days:
ChatGPT: 29,551,174
TikTok + Facebook + Instagram + X: 32,859,208— #Similarweb (#@Similarweb)
6:54 AM • Jun 24, 2025
A temporary 128-subnet cap is introduced on Bittensor
128 Subnet cap introduced
⚪️ "This means registering new subnets will not be possible until the cap is increased"
⚪️ "Like previous limits we've applied during earlier stages of #Bittensor’s growth, this cap is intended to reduce noise and improve the quality of technologies
— #DREAD BONGO (#@DreadBong0)
12:41 PM • Jun 25, 2025
The OpenAI Files. Better than The X-Files, even without Gillian Anderson. Grab your popcorn.
Somebody just released the OpenAI Files.
10,000+ words on Sam Altman, employee exits, secret deals, and whether he was ever actually CEO of Y Combinator...
We read it all so you don’t have to 👇
— #big machines (#@bigmachinesAI)
2:34 AM • Jun 20, 2025
The future of work: Many YC-funded AI companies target tasks in the “Red Light” Zone (low worker desire).
Future of Work with AI Agents
Stanford's new report analyzes what 1500 workers think about working with AI Agents.
What types of AI Agents should we build?
A few surprises!
Let's take a closer look:
— #elvis (#@omarsar0)
6:51 PM • Jun 20, 2025
That’s a wrap for this week! Got thoughts, feedback, or something cool to share? Just hit reply. We read it all.
Cheers,
Teng Yan & Issy
Did you like this week's edition?
This newsletter is intended solely for educational purposes and does not constitute financial advice. It is not an endorsement to buy or sell assets or make financial decisions. Always conduct your own research and exercise caution when making investment choices.