Hedici. Stock Investor. Inflection point. Collector. Arblocks, Brightmoments, fxhash. Some 1/1 that heals my heart. Banner: Parnassus/PFP: Edifice. @GrailersDAO
Wherever there's Art..
Joined January 2021
- Tweets4.6K
- Following2.3K
- Followers2.8K
- Likes103K
Pinned Tweet
I still love art 🖼️ — that hasn’t changed, and it isn’t going anywhere. But from here on, I’ll mostly be posting as an investor🧐: ideas worth thinking about, lines from conference calls that stuck with me, and the one sentence from a CEO or a fellow investor and so on.
$AMKR 3Q26 earnings call, Kevin Engel(CEO)
"I think that has highlighted even more that packaging is becoming one of the key bottlenecks. Customers are really working to make sure they secure capacity into 2027 and even into 2028. So that is where you start to see a lot of these advanced payment discussions.
...They are more focused on making sure they secure the capacity. So at this point, we feel very optimistic around the pricing environment related to these multi-year agreements."
3Q26 Meta Earnings CC, Mark Zuckerberg
"Part of the reason why we're not just charging everyone upfront to use this is we want to establish this muse network, and we think that it's going to be more valuable for people at scale when there are a lot of people on businesses using Muse."
Same logic as an MMORPG: nobody wants to log into an empty server.
FY3Q26 $SPCX earnins call
Elon Musk
"The value of that compute is proportionate to the useful intelligence per watt. That is why you see things like even old computers, like the H100s, renting for more than they rented when they were new. I think that is likely to continue. I think we are going to face a shortage of super intelligence compute, and that shortage is going to remain for a long time."
"I do not think we are going to have a demand issue here. I am not saying this will happen, it is not impossible that we will have a six-month payback period for compute next year."
Quick thought on how Amazon should navigate agentic commerce:
In the era of agentic AI, Amazon can’t protect every moat.
The winning play isn't guarding the front-end UI or ad space—it’s owning the catalog, real-time inventory, and fulfillment.
Let third-party agents (like Meta's Muse) handle discovery, while Amazon monetizes the back-end via catalog APIs and sponsored agent rankings. Concede the interface to own the infrastructure.
$AMZN, $META
I understand the first line of thinking that $AMZN doesn't like $META's Muse to shop around, and people starting to use AI agents as the entry point for commerce, but this is the future.
I don't believe $AMZN has a chance of having its own world-dominant personal AI assistant that people would fully transition to. Best case is $AMZN has an endpoint or their own agent that Muse and other personal agents talk to, and not crawl on their websites.
$AMZN's e-commerce business will still do well in the agentic era (but they have to lean into the agentic era, not go away), but their ad business (the part of it that is related to discovery on their website) will probably be worth a lot less as agents take over.
Crucial point at this juncture: Given their unbeatable distribution moat (cross-marketing & pricing), bridging the last gap to the frontier with Watermelon will let them capture the first-mover advantage.
the most rewarding part about muse is seeing it make a difference for people who aren't in tech and don't give a shit about AI
seeing muse click for people in my life outside of tech, and watching them use agents without having to know what a CoT or MCP or CLI are, feels like what we've been trying to build towards the whole damn time
love all the musers ❤️
$META
100M free tokens per user per week isn’t generosity. It’s a land grab.
Meta isn’t just trying to win AI usage. It’s trying to own the habit, the context, and eventually the personalization layer.
And if Meta only targets advertising, its TAM is capped by the world’s ad spend.
Zuckerberg seems to be thinking much bigger: AI could expand Meta beyond the limits of the advertising economy.
Mark Zuckerberg (@finkd) on Muse, Meta's personal AI agent
I recently sat down with Zuckerberg for episode two of the Sources podcast. We go deep on Muse, Meta's new (and, in my experience over the last few days, very capable) agent that comes with its own virtual machine and can do much more than a chatbot.
Zuck, in fact, is confident that Muse will pay for itself by making people money.
We also discuss his views on the data center backlash, his concerns about the concentration of power in AI, Meta's recent youth-safety settlement, privacy concerns about Meta's smart glasses, and more.
Timestamps:
00:00 Zuckerberg’s Vision for AI
08:34 Data Centers, Jobs and Local Communities
15:03 Why Meta Is Building Personal Superintelligence
19:14 How Zuckerberg Uses Muse
25:00 Muse Pricing and Business Model
30:31 Privacy and Security for Personal Agents
41:06 Who Will Win the Personal AI Agent Market?
44:31 Rebuilding Meta’s AI Lab
54:09 AI Safety and Working With Government
01:01:51 Smart Glasses and Privacy
01:05:50 Teen Safety and Social Media Limits
Thanks to the show's premier sponsors: @meetgranola, @mercury, and @Atlassian
OpenAI president Greg Brockman on Astra:
“Maybe the headline one is computer use. It’s really crossed the threshold for me.”“For the past two years, it’s been a connector era.”
“We now have the technology that’s almost this universal connector.”Coding agents were the moment AI could write software.
Astra looks like the moment AI can use a computer.
stratechery.com/2026/an-inte…
$IREN, FY26 Q4. Ignore the $684M loss — that's the mining wind-down. Look at price.
Three-year contracts above $20M/IT-MW, up ~125% since November. Implied November base: ~$8.9M. Nine months.
And it isn't spot — "3-to-5-year deals, not 2-year, not spot capacity." The near end of the term curve moved twice as fast as the far end.
Meanwhile: "Twelve months ago, GPU financing barely existed as an asset class." $6.5B raised in three months, both ends of the credit spectrum.
The bear case was that only IG counterparties could be funded. That constraint is dissolving.
GPUs being underwritten as collateral is, to me, one of the most consequential things in this quarter.
Two lines from the $NVDA call (FY27 Q2) that quietly reprice a lot of models.
On memory: "As you are already aware, we are experiencing extreme pricing conditions in memory. The magnitude of the price increase has exceeded our prior expectations and are headed even higher into next year."
➡️So much for the peak-memory call. This is the buyer of memory saying it.
On agents: "The amount of compute necessary for an agent versus a human using it is probably 15 - 100 times, depending on the type of problem you are trying to solve."
➡️Most agent frameworks I've seen were penciling in something closer to 10x.
What does this news mean for $OKTA?
It is a landmark, MCP-based milestone that establishes Identity Access Management (IAM) in the Agentic Era—transforming identity from mere human logins into a central governance plane for AI agent actions and tool access!
Supported by Asana, Atlassian, Canva, Datadog, Figma, Granola, Linear, Notion, Slack, and Supabase.
Developers can also add enterprise-managed auth to their own third-party MCP connectors in the Claude directory.
Blog:
claude.com/blog/enterprise-m…
Docs:
claude.com/docs/connectors/b…
Humans still type at roughly the same speed they did last year. Agents don’t. 14x in six months, and nothing about that curve says it’s done. (graph via @a16z )
$NET, Q2: “more than 50% of the traffic flowing across Cloudflare’s network was not human.” AI agent requests +1,700% YoY.
$FSLY, Q2: “Machines don’t browse the way people do.” AI traffic growing ~6.5x the rate of human traffic.
Meta just announced plans to sell excess AI compute via “Meta Compute” (hosted models + raw capacity), while simultaneously renting massive capacity from Nebius, CoreWeave, and others. Looks contradictory at first glance.
But it actually makes sense when you separate training vs inference:
• Large-scale training needs dedicated, tightly-coupled internal clusters (performance + security).
• Inference is bursty and benefits enormously from multi-tenancy. Pooling external demand significantly raises GPU utilization.
At current neocloud rental rates (~$2–3.5/hr for H100), even partial utilization on excess/older-gen GPUs delivers strong ROIC. Meta can use short-term rentals for urgent training while building its own infra long-term, and monetize Hopper-era capacity through hosted Llama inference or raw compute.
Skepticism on the timing is fair, but strategically this is a rational move — especially as inference will dominate long-term compute demand.
In March, Meta signed a $27B cloud services deal with Nebius. In April, Meta signed a $21B expansion of their cloud services agreement with CoreWeave.
Two weeks ago, Bloomberg reported that Meta signed up to rent 1.6GW of data center capacity from Crusoe. And just 3 days ago, the Financial Times reported that Google had capped Meta's use of Gemini because of capacity constraints.
And now all of a sudden, Meta actually has excess capacity?
In reality, I suspect Meta looked at SpaceX's valuation and recent deals to sell excess capacity.. And came to the conclusion that investors will look favorably upon any willingness to monetize their capacity externally.. Which will give their share price some relief, and allow them to raise capital.. Which they'll use to buy and build more compute capacity, lol.
$META $GOOG $NBIS $CRWV
Richleech retweeted
Transformer’s Attention mechanism has come a long way. We’d like to thank the researchers and the engineers in the open-source community for continuing to make high-performance AI accessible. Please celebrate with us by sharing this post, tagging more contributors, and sharing anecdotes to complete the open history of Attention! (1/8)🧵
Richleech retweeted
The most beautiful composition from yesterdays night - such a striking visual dialog 🥺
@sofiagarcia_io has an amazing intuition and impeccable taste 🙌
