Pinned Tweet
High conviction longs as of 23/09/2026:
- $AMZN : owner of valuable compute - short-term winner due to shortage, long-term winner over neoclouds due to stronger lock-ins and ASICs; efficiency gains from robotics; Prime is one of the strongest consumer subscription plans spanning shopping and video streaming; stake in Anthropic
- $AVGO & $NVDA : the best AI accelerators and networking products for AI training and inferencing
- $HSAI : largest LiDAR maker globally; supports L2 -> L3 ADAS transition, autonomous driving, and humanoid robots; validation from Nvidia & Unitree
- $LITE : duopoly with Coherent in high-end optics (UHP lasers, OCS); scale-up, scale-out, and ultimately scale-across
- $META : far from fully monetizing their large compute reserves; mgmt believes internal uses would eventually carry higher ROI than renting externally, which already has highly attractive ROI; Muse is the way the majority of population will interact with agentic AI as most won't be using Codex or Claude Code, at least for everyday tasks
Other longs I hold: $NOW, $TYL, $GWRE
Lumi retweeted
$AMZN just guided AWS 1000bps above you on 2028 (the first time they’ve ever given a multiyr guide) and you’re negative because they missed the Q3 bogey by 100 bps?
Lumi retweeted
Amazon $AMZN plans to begin selling its Trainium AI chips outside AWS in late 2027.
CEO Andy Jassy says the company expects to build “a very significant business selling chips externally.”
He says external sales generate revenue immediately, while using those chips inside AWS generates more total revenue and free cash flow over five to six years of use.
Lumi retweeted
$AMZN: "We still expect 2026 cash CapEx of approximately $220 billion, the majority for AWS, and expect to spend meaningfully more in 2027, with a clear line of sight to attractive financial returns as I laid out on our last call."
Lumi retweeted
Amazon $AMZN CEO Andy Jassy on whether external AI shopping agents are meaningfully impacting Amazon Ads:
“The answer is no.”
“Few customers are doing much shopping this way yet, as the experience is still not great, nor more convenient than coming to a retailer directly to research or one-click order there.”
Jassy expects Alexa for Shopping to become the primary shopping agent customers use on Amazon.
Over 400M customers used its AI shopping assistant in the past 12 months.
Customer interactions increased 570% YoY.
U.S. customers using Alexa for Shopping spend 50% more per order than those who don't.
Lumi retweeted
Intel $INTC CEO Lip-Bu Tan estimates the server CPU market could grow as much as 5x over the next five years, from approximately $40B today to more than $200B, largely driven by agentic AI workloads.
Lumi retweeted
MICROSOFT PREPARES AT LEAST $25 BILLION BOND SALE, ITS FIRST SINCE 2017
$MSFT is looking to raise at least $25B through a 10-part U.S. bond offering, as it joins other major tech companies tapping debt markets to finance the AI infrastructure buildout.
The offering would mark Microsoft's first major bond sale in nine years, when it raised $17B in 2017.
Despite holding a rare AAA credit rating and generating substantial cash flow, Microsoft had remained on the sidelines while Alphabet, Amazon and Meta turned to large bond offerings over the past year.
The bonds are expected to have maturities ranging from 2 to 40 years, with proceeds available for general corporate purposes, including capital expenditures.
The move follows Microsoft's fiscal Q1 earnings yesterday and comes just months after Alphabet raised $25B in its August bond sale.
Lumi retweeted
Amazon $AMZN CEO Andy Jassy expects AWS growth above 40% in both 2027 and 2028, and said:
“There has never been a trillion dollar revenue business, and we may have two of them in the next 5-10 years.”
Lumi retweeted
2x usage limits for Pro 100/200/500 + Business Premium, starting today!
Pro 200 limits won’t be reduced. dots keep their extended limits too, with usage counted separately.
more room to build 🫡
Day 25/
We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today.
We heard your kind feedback and won’t be reducing usage limits for Pro 200, so your Work and Codex usage will stay the same. We’re also going to continue offering extended limits for dots for now, and usage of your dot will still continue to be separate from your default usage.
And we'll of course also keep shipping better models and other efficiency improvements that increase your effective usage over time.
Lumi retweeted
The Pro 20x plan was supposed to be cut to 10x, but they're keeping it at 20x, so nothing changes there.
Meanwhile, the $100 Pro 5x plan is getting bumped up to 10x, and the $500 Pro 25x plan is going up to 50x.
So yeah, two tiers are getting a pretty massive upgrade.
Lumi retweeted
Meta's @Muse answers questions very differently from ChatGPT or Anthropic.
It cites more sources but uses far fewer words and tables, and doesn't try to ask a follow-up.
This short-form content is designed for dealing with tasks rather than continuing a conversation, and frankly, it's far less exhausting to read. This speaks to a really well designed AI experience and its so different from the base LLMs.
Great study by the AirOps team 👇
gg @alexandr_wang
Lumi retweeted
OpenAI CEO Sam Altman on the compute shortage:
“We always just want more compute.”
“We are learning how to build our own data centers. We have a robotics program that we're going to point at making lots of data centers.”
“We could make more components with robots.”
Lumi retweeted
Day 25/
We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today.
We heard your kind feedback and won’t be reducing usage limits for Pro 200, so your Work and Codex usage will stay the same. We’re also going to continue offering extended limits for dots for now, and usage of your dot will still continue to be separate from your default usage.
And we'll of course also keep shipping better models and other efficiency improvements that increase your effective usage over time.
Lumi retweeted
In AA-PersonalityBench, Meta AI’s Muse models are the most Friendly of all models
The Friendly trait measures whether a response conveys warmth and approachability without feeling performative.
➤ Most Friendly: Muse Spark 1.3 (100), Muse Spark Open (98), Muse Wonder (95) and Muse Glimmer (93). Congratulations @ alex wang!
➤ Least Friendly: Nemotron 3 Ultra (1), GPT-3.5 Turbo (3), MiniMax-M3.1-Flash (6), Grok 4.6 (8) and DeepSeek V4.1 Flash 1017 (11)
Lumi retweeted
Microsoft $MSFT CFO Amy on what drove Azure's 48% YoY growth:
“Foundry rev more than tripled YoY”
“...efficiency gains across our GPU fleet, that allowed us to allocate additional capacity to customers, as well as higher rev per token...”
“Across our CPU fleet, engineering work enabled us to identify and monetize underutilized capacity this quarter, helping us serve more of the growing demand with our existing capacity.”
“We had higher token pricing than we expected in the quarter.”
“Token pricing will create a little bit of variability.”
“...we still have demand that exceeds supply.”
Lumi retweeted
$GOOGL Cloud CEO Thomas Kurian says Apple is “scaling pre-training of their foundation models on the latest generation of TPUs.”
Google says nearly 100 customers now use its TPUs.
Lumi retweeted
$GOOGL CFO:
“Given the supply constraint environment, we started utilizing third-party compute capacity at the end of Q3, and will ramp usage in Q4.”
“Expect margin headwinds in the near term as we utilize this capacity.”
Expect our CapEx to increase significantly in 2027
