On one hand $meta Muse potentially creates more customer support enquiries / tickets, and on the other, Meta Business Agent helps businesses handle this new workload. Genius.
$META ACQUIRES SWEDISH AI STARTUP STILLA
Meta is buying Stilla AI to accelerate Meta Business Agent, its AI system for helping merchants handle customer conversations and transactions across WhatsApp, Messenger and Instagram.
Stilla built an agent that can retain company s across workplace software, with integrations spanning thousands of tools.
The acquisition plugs that technology into a business product Meta says is already used by more than 1M businesses, with over 1B active business conversations taking place across its messaging apps each day.
Stilla was founded in Stockholm in 2024 and raised $5M in pre-seed funding earlier this year. Meta also plans to expand its presence in Sweden following the deal.
4
1
25
3,685
Unimicron's chairman explains substrate demand in an accessible way:
"In recent years, each generation of TSMC’s advanced process technology has been accompanied by corresponding advances in packaging technology.
As package sizes increase and I/O interconnects become denser, substrate manufacturers must simultaneously make substantial improvements in factory cleanliness, equipment precision, and the number of substrate layers.
That means installing entirely new equipment and building new factories capable of meeting the requirements of each generation.
... virtually every new TSMC factory creates demand for a corresponding new substrate factory.
... TSMC said it would build 13 fabs in Taiwan and another five to six overseas over the next several years.
This effectively means that 18 new ABF substrate factories for AI applications will need to be built before 2030."
"In recent years, each generation of TSMC’s advanced process technology has been accompanied by corresponding advances in packaging technology.
As package sizes increase and I/O interconnects become denser, substrate manufacturers must simultaneously make substantial improvements in factory cleanliness, equipment precision, and the number of substrate layers.
That means installing entirely new equipment and building new factories capable of meeting the requirements of each generation.
... virtually every new TSMC factory creates demand for a corresponding new substrate factory.
... TSMC said it would build 13 fabs in Taiwan and another five to six overseas over the next several years.
This effectively means that 18 new ABF substrate factories for AI applications will need to be built before 2030."
5
3
30
4,474
Taiwanese substrate manufacturers have been dismissive of foreign attempts to break into the supply chain. However, as Tech in Taiwan explains in the post below, SEMCO proved them wrong.
They might be too dismissive of AT&S too...
cwnewsroom.substack.com/p/nvidia-turns…
They might be too dismissive of AT&S too...
cwnewsroom.substack.com/p/nvidia-turns…
0
0
4
1,033
Trapped retail investors in Kioxia have no one to bail them out.
Outstanding leveraged long positions in the stock totaled about ¥920 billion ($5.8 billion) as of Oct. 15, the largest among all TSE companies.
Meanwhile, trading activity in Kioxia has declined along with its ver dwindling by 84%.
Kioxia has been fairly good to its shareholders though, buying back ¥800 billion in just six trading days at an average price of ~¥ 16,500, when it previously said it would prioritise dividends.
This wouldn't be the first time they've suggested one thing and acted differently. In an interview in September, Kioxia's CEO said the company would avoid unsustainably high prices for data center customers.
However, Kioxia's LTA strategy suggests they are positioning to capture what they view as more pricing upside. Kioxia has said it thinks around 50% of US DC demand is unmet and expects that demand to carry forward into next year. Compared to peers, and counter to its customers' requests, the company has generally resisted locking up its capacity under long LTAs. This suggests that Kioxia is prioritising pricing over a gentler plateau.
With Meta, Google and SpaceX signalling that they are pressing ahead with capex, that strategy may pay off.
$kxiay $285.a
Outstanding leveraged long positions in the stock totaled about ¥920 billion ($5.8 billion) as of Oct. 15, the largest among all TSE companies.
Meanwhile, trading activity in Kioxia has declined along with its ver dwindling by 84%.
Kioxia has been fairly good to its shareholders though, buying back ¥800 billion in just six trading days at an average price of ~¥ 16,500, when it previously said it would prioritise dividends.
This wouldn't be the first time they've suggested one thing and acted differently. In an interview in September, Kioxia's CEO said the company would avoid unsustainably high prices for data center customers.
However, Kioxia's LTA strategy suggests they are positioning to capture what they view as more pricing upside. Kioxia has said it thinks around 50% of US DC demand is unmet and expects that demand to carry forward into next year. Compared to peers, and counter to its customers' requests, the company has generally resisted locking up its capacity under long LTAs. This suggests that Kioxia is prioritising pricing over a gentler plateau.
With Meta, Google and SpaceX signalling that they are pressing ahead with capex, that strategy may pay off.
$kxiay $285.a
2
2
43
6,889
Two pitches at Sohn Tokyo (20%) were SaaS companies.
Constellation Software now up year to date. $CSU.TO
1
0
8
3,662
I found being called an AI manic amusing - I think some bottleneck bros are misunderstood. I am bullish about shortages, exponential demand, and ASP increases, not necessarily frontier lab financials.
@tickerplus And way way below the bullish estimates many of the AI manics were pointing to suggesting ARR running as high as 95bln.
x.com/illyquid/statu…
4
0
25
4,606
Yesterday, cyber, data protection, cooling and power were really strong.
The boffins have been busy tinkering with @FundaAI's thematic rotation play. Its helpful to make sense of things if you don't have a terminal.
funda.ai/play/sector-ro…
The boffins have been busy tinkering with @FundaAI's thematic rotation play. Its helpful to make sense of things if you don't have a terminal.
funda.ai/play/sector-ro…
0
0
10
2,668
Embarrassed to find myself in such illustrious company here - it’s just been an extraordinary year of crazy market and supply chain conditions.
@__paleologoi@macrocephalopode@0xfdfs@moreproteinbars:@bennpeifertt@KrisAbdelmessihn@0xFaust12nt who has taught me the most. I don't come from a funda@gregoryblotnicky background myself and it was really really useful to see things from a (very) good pod guy. Also love his philosophy of life !
@illyquid : found an enormous amount of opportunities on the Asian and European markets this year, always learning and teaching new stuff, always humble and measured
@kevin__mak : probably the guy who has the most discernable edge and bet understanding (coming from poker) of fintwit. Stock pitches are amazing and bet/investing understanding is freaking good.
Tons of other people who are good theme/stock pickers with pitches based on simple and clear ideas - I would most of them fall under the category of "thematic"/narrative investing.
Which is interesting because it is not so amenable to systematic (or at least I can't see an easy way !), and probably not that pursued by HF (unless it falls under macro ?)
1
0
13
4,282
$nvda says Vera Rubin capacity is now “thousands of racks a week,” with a VR compute tray assembled in 1 minute against 90 minutes for GB200.
We think full-rack testing is emerging as a key determinant of output: fundaai.substack.com/p/researchpowe…
We think full-rack testing is emerging as a key determinant of output: fundaai.substack.com/p/researchpowe…
Jensen Huang says $NVDA Vera Rubin is “in full production” and now takes less than 5 minutes to assemble per node versus 2 hours for Grace Blackwell with zero cables and 100% liquid cooling.
That’s over 24x faster assembly which could mean faster deployments and better margins t generation of AI systems.
0
2
33
6,838
Rather than reporting that investors are "almost certainly" in an AI bubble, I think the fairer interpretation of Ken Griffin's comment is that there are parts of the AI trade that are in bubble valuation territory ("there's almost certainly bubbles within this moment in time.")
He then goes on to say that looking back at the dot com boom, some internet companies had pretty fanciful business models but he wishes he kept his Amazon stock. I.e. investors are being more discerning now and some companies will come out of this big winners.
He then goes on to say that looking back at the dot com boom, some internet companies had pretty fanciful business models but he wishes he kept his Amazon stock. I.e. investors are being more discerning now and some companies will come out of this big winners.
Citadel founder and CEO Ken Griffin says investors are “almost certainly” in the midst of an AI bubble during an interview with Bloomberg's Dani Burger at Citadel's annual Future of Global Markets Conference in New York youtu.be/mV1V0FXixDE
3
2
29
7,971
A former Google employee says that starting around mid-Q2, customers began swapping some Anthropic and OpenAI models out of their multi-agent setups for open-source ones.
That is consistent with an expert interview we did in August:
That is consistent with an expert interview we did in August:
Former $GOOGL employee on why data center capacity and electricity are the real bottlenecks, not GPU production itself, and how unreliable some of the neoclouds are ($AMZN, $GOOGL, $MSFT, $NVDA):
1. According to the expert, open-source models are closing the gap with frontier cted, with customers increasingly replacing Anthropic and OpenAI models in multi-agent architectures with open-source alternatives starting from Q2 of this year. The shift started with simple tasks like tool calling and MCP, but has moved up the stack to low-level orchestration.
2. The expert identifies capacity as the biggest constraint at the moment, with certain companies' traffic growing 5x while GPU availability, particularly in the U.S. where latency requirements are tightest, simply cannot keep up. Sourcing capacity overseas in Asia or Australia is technically possible but adds latency that most customers will not accept.
3. Hyperscalers offer reliability but are expensive, require multi-year commitments of three or more years, and have limited availability of the latest hardware, with customers already chasing B300s. AI-native companies tend to gravitate toward neoclouds due to hyperscaler pricing and contract rigidity, but neoclouds come with their own reliability issues, with GPUs dropping off unexpectedly and requiring engineering resources to manage fleet stability around the clock.
4. The third tier consists of smaller independent operators who have invested in GPU racks, seeing the opportunity to lease out capacity and recoup costs within a year or two, with the remainder being pure profit. The expert describes these as essentially unusable in practice, with operators who do nothing more than find land, install racks, connect electricity, and set up cooling, with no real cloud management, maintenance, or operational expertise behind them.
5. The expert expects overall GPU availability to increase meaningfully over the next year or two, with $NVDA continuing to ramp production. The real bottleneck is not GPUs themselves but data center capacity and electricity in the U.S. On the supply side, the expert also sees players $SPCX beginning to lease out spare capacity to the market, with more likely to follow.
6. The expert notes that U.S.-based neoclouds typically require 20-30% prepayment upfront, with the remainder paid monthly or quarterly over the contract term, and rebates negotiated at the end for any reliability incidents like traffic loss. APAC-based providers operate differently, with no prepayment required at all.
found on @AlphaSenseInc
6
7
86
17,921
Slightly alarmist version of the point that Neolabs spend a lot at Neoclouds.
This chart shows the key risk to the AI trade, and it's missing from the latest revenue announcements.
The AI boom is fundamentally driven by a top 1% of companies that drive 80% of spend. Within that top 1%, 80% are venture-backed.
That's all fine if the AI trade works out, nt investors say, everyone wins.
But if it doesn't work out, then we have a cascading effect of large customers simultaneously pulling back their AI spend.
0
1
9
2,553
Funda's power and liquid cooling takeaways from OCP draws on learnings from the show floor and 95 sessions, 22 of which we attended in person.
I thought the point about Vera Rubin’s bottleneck shifting toward full-rack testing and the data hall was interesting.
fundaai.substack.com/p/researchpowe…
I thought the point about Vera Rubin’s bottleneck shifting toward full-rack testing and the data hall was interesting.
fundaai.substack.com/p/researchpowe…
1
1
13
3,937