Joe W (@JoeDataCap) profile picture

CMO @ Aethir. Building and researching AI infrastructure.

Cloud
Joined November 2017
Photo by Joe W (@JoeDataCap)Photo by Joe W (@JoeDataCap)Photo by Joe W (@JoeDataCap)Photo by Joe W (@JoeDataCap)Photo by Joe W (@JoeDataCap)Photo by Joe W (@JoeDataCap)Photo by Joe W (@JoeDataCap)Photo by Joe W (@JoeDataCap)Photo by Joe W (@JoeDataCap)Photo by Joe W (@JoeDataCap)
Aethir gets into GPU datacenters ACCELERATE is building its footprint across 10 sites in the US + EU, all already secured by @AethirCloud The key: they're powered shells — existing buildings, power grid connected. This allows for faster deployment. These sites are the scarcest asset in AI right now
Introducing Aethir ACCELERATE: Aethir is making its largest move into physical AI infrastructure to date. Access secured to 10 initial sites, with up to 20MW of total capacity. Optimized for NVIDIA B300 and GB300 clusters of 64 to 256 nodes. Sites are located across the US and Europe. The numbers: up to $700 million in contracts projected by the end of 2026. Over $2 billion in total contract value once all 10 sites are fully built out. For four years, the decentralized GPU cloud model was defined by what it didn't own. That ends here. Aethir’s network is now entering the data center business with its own GPU capacity. The timeline is in months, instead of years. The full breakdown is in our latest blog👇
4
13
15
5,669
Just how big is Binance’s reported $6B hack? Looking at only the top 10 largest hacks. This one is bigger than the other nine hacks on this chart combined at $5.49B. And 4.2x the size of second biggest Bybit’s $1.5B breach. Here’s how it compares with crypto’s largest hacks. 🧵
1
21
Who are the hackers? Three hacks are confirmed to North Korea account for $8.4B, or 71% of this top 10 total. Two others suspected North Korea cases brings the count to 5 of top 10
1
4
82% vs 0% The more interesting stat is the breakdown, centralized exchanges account for 82% of the amount in the top 10 largest hacks, while decentralized finance is 0%. Bridge hacks make up 18%. Crypto custody and secruity still key. Stay safe
3
NYC folks, join me and Aethir tonight for a small intimate chat on AI infra Drinks and conversation with people building and running AI infrastructure. Would love to hear what you're working on. Will share more about @AethirCloud AI plans 6–9 PM ET, Manhattan, NYC. Hudson Yards. RSVP a spot: luma.com/mz1z6wkc – at New York, USA
2
1
21
132,564
Not all neoclouds are the same. The useful question is which neoclouds can fund the gap between buying GPUs and getting paid Prepayments, financing, and delivery timing matter more than a headline backlog numbers. Same AI demand. Very different risks
Replying to @chamath
Pretty much. As of the past few weeks, debt markets have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centers. SpaceX closed $40B of debt for GPUs, but we’re investment grade and have Starlink & launch revenue. That means the 200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many.
1
1
7
229
Timing to bring sites live is one of the most key metric for neoclouds Good to see Axe working with an experienced partner like Penguin to enable faster deployment and dependable operations @AxeCompute @WeAre_Penguin
Demand for Axe Compute Build is surging. Today AGPU announced a multi-year partnership with Penguin Solutions (NASDAQ: PENG) to expand capacity and accelerate delivery to meet it. Penguin will support multiple sites as we work to bring $3.45B+ in contracted AI infrastructure projects to production globally. #AIInfrastructure $AGPU $PENG #AIInfrastructure @WeAre_Penguin
1
53
Congrats to the Axe Compute team on ringing the Nasdaq bell A moment to celebrate the team building behind the scenes. Looking forward to seeing the next chapter of AI infrastructure @Nasdaq @AxeCompute
Last week Axe Compute rang the @NasdaqExchange Opening Bell in Times Square. A milestone for the team, and a marker on the way to something bigger. The building is just beginning, and we’re here to define the next generation of AI infrastructure. We’re just getting started. $AGPU #AIInfrastructure
48
A big step for our team This part I care most about is connecting compute activity to ATH burns Starting in November, a share of Aethir’s fees will fund daily burns. And it will be verifiable onchain
Major ATH tokenomics update: we’re introducing daily ATH burns, funded by a share of network fees. Every burn goes to a public burn address, where tokens are permanently removed from supply. The mechanism goes live in November. We’re also replacing the fixed platform fee with a floating network fee, set by a published formula that reads market conditions. When the market weakens, the fee falls and the share that gets burned rises. Cloud Hosts are protected by design: the fee can never exceed the value of Cloud Host rewards. The underlying host emissions schedule remains unchanged. The update changes the fee mechanism applied around those rewards. A dashboard tracking daily, monthly and cumulative ATH burned will be available in November. Burns that follow real network activity, and you can check every one of them. Read the full announcement below👇
3
70
Two things on his mind @AnthropicAI Building a better model How to get more compute
2
81
At SF Tech Week: everyone agrees the compute demand is crazy A question popping up at multiple meetings within the infra circle is if the compute shortage due to real capacity, or hoarding and bad matching and utilization? Shortage means H100 contracts up 40% this year, Blackwell up ~50%. And hearing most datacenters have a degree of under utilization. Compute can be genuinely scarce and still badly underutilized at the same time. The question is what data would prove the difference over time.
1
1
3
189
In town for SF Tech Week! Excited for a full week of AI infra + founder meetings Only 1 day in, and already overwhelmed by amount of quality folks working on AI datacenters, compute pricing, and agents If you are working in AI infra or just building something interesting, hit me and @AethirCloud team up! @Techweek_
1
7
16,993
Great week in Seoul for KBW 🇰🇷 One of the highlights was bringing together AI builders, investors and partners at the Aethir AI Dinner. Korea continues to be one of the most important markets for AI and compute. Going forward, @AethirCloud will be working more closely with local partners. More coming from Korea soon
This year's @kbwofficial is a wrap, and Seoul delivered. Three days at Walkerhill, September 29 to October 1, with a program built around institutional finance, AI integration, and real-world adoption. Full halls, side events that ran late, and a builder community that came with real questions. What stood out to our team was how often the conversation turned to compute. Teams across Korea's Web3 and AI scenes want GPU capacity they can reach quickly and scale without a long procurement cycle. Thank you to everyone who spent time with the Aethir team in Seoul. We're already looking forward to next year.
2
145
Late to the game but marketing is on point
we are considering opening a muse merch store i am doing market research how excited you would be by muse merch, from a scale of "meh" to "you can have my right kidney"?
1
1
97
10Y treasuryat 5.04%, highest since 2007. Nebius guided 2026 capex up to $25B, from $5B in 2025. >90% of the old $16B capex was covered by cash; the rest is "additional financing." Getting power is hard, gettingn money got harder. $NBIS
1
1
1
187
The signal in Amazon x Qualcomm deal is the warrant, not the $60B headline. @amazon AWS can buy up to 25M $QCOM shares, or ~$4B worth. Amazon is not just investing into a supplier vendor. It is aligning @Qualcomm upside with its AI infra buildout. Hyperscalers buying optionality $AMZN $QCOM
QUALCOMM AND AMAZON PARTNER ON CUSTOM AI DATA CENTER CHIPS Qualcomm $QCOM and Amazon have announced a multi-generational collaboration to develop customized silicon for AWS AI infrastructure, with an initial focus on large-scale AI inference. The companies will also work on advanced optical connectivity for data centers, including links up to 1.6T, using Qualcomm’s SerDes and optical DSP technology. Qualcomm also plans to expand its use of AWS AI infrastructure, including Amazon Bedrock, for chip-design and EDA workloads as it looks to shorten development cycles.
2
1
275
AI infra is too constrained to treat critical vendors casually. Hyperscalers are building their own chips and using these warrents / commerical structures to help secure capacity with suppliers
1
61
The hype is pretty Astra demos and examples. Core support comes from actual users describing an agent that can take a vague goal, code and software, inspect its output, and fix issues. Basically agentic looping. Not polished finished work yet. But it is the first move from model output to verified task completion. @OpenAI
This impressed me: I asked GPT-6 Astra to turn Zork (the classic 1977 text adventure) into a full 3D action-adventure game It kept the original plot & puzzles, added fight scenes, and built all of the characters & environments directly in Threejs Play: zork-underground-empire.netl…
2
1
79
Astra isnt just raw intelligence. It is getting to agentic reliable execution
34
Nvidia guides $108B rev assuming 0 China input. China response: $295B, 5 years, 80% domestic chips China takes <10% of new compute today, and doesn't see as downside, Beijing see it as AI buildout opportunity. Exclusion vs substitution
1
1
50
Pricing frontier: 9¢ vs $1.04 per task. Same tasks, same day, public prices. China's open models are within ~2 points of GPT 5.6 Sol at ~9% of the cost Intelligence Index: GPT 5.6 Sol score: 59 (closed) @OpenAI GLM 5.3-Flash score: 57 (open) @ZhipuAI Qwen 3.8 Flash score: 56 (open) @Alibaba_Qwen Cost per task: GPT-5.6 Sol : $1.04 GLM-5.3 Flash : 9¢ Kimi K3 $0.94: above Sol @Kimi_Moonshot DeepSeek V4 Flash: $0.14/M tokens @deepseek_ai Models are commoditizing. More compute is needed
2
1
166
The moat moved up the stack: power and distribution. Whoever serves commoditized AI at the lowest marginal cost wins And compute demand goes up $AGPU
2
170