DanielRomero

vip
Active for: 5.8y
Peak Tier 5
No content yet
Samsung (005930.KS) challenges $TSM production with Panel-Level-Packaging
Samsung plans to bring its 415×510mm FOPLP platform into mass production around 2028, sticking with a significantly larger panel format than TSMC’s 310×310mm standard
The larger Samsung panel offers 2.2× the usable area of a 310×310mm panel, potentially providing meaningful unit-cost advantages. Samsung also already has high-volume PLP manufacturing experience from mobile and wearable chips, giving it existing infrastructure and expertise in warpage control
The challenge is moving the technology into AI server packaging,
post-image
  • Reward
  • Comment
  • Repost
  • Share
Powertech ( emerges as a key advanced-packaging supplier for $AMD and $AVGO
AMD and Broadcom have already booked most of Powertech’s planned FOPLP capacity, with mass production expected to begin by mid-2027. Capacity is projected to increase from 1,000 panels/month in 2026 to 2,000 by mid-2027 and 5,000 in 2028
Powertech expects total investment in the platform to reach NT$70B, while the initial pilot-line investment is expected to break even in 2027. Management expects FOPLP to become a major growth driver from 2027 onward
Its PiFO platform uses panel-level RDL and silicon bridges to connec
AMD-1.25%
AVGO-1.35%
post-image
  • Reward
  • Comment
  • Repost
  • Share
$IREN projected $25–30B in FY27 capex
· $14B is already secured through prepayments and GPU financing
· Another $8B is expected through the same means
· That leaves $3–8B to be financed
· The rest will come from cash flow, data center financing, and equity issuance
The plan isn’t crazy, especially if there’s further share-price appreciation
But so many people can’t read past the headline
IREN-13.70%
post-image
  • Reward
  • Comment
  • Repost
  • Share
$IREN brings out so many dumb takes
It’s painful to read or listen to
For some reason, this company attracts the most amateur critics with the shortest attention spans
IREN-13.70%
post-image
  • Reward
  • Comment
  • Repost
  • Share
$MU today on the supply-demand equilibrium:
“We think meaningful supply will only start to ramp in 2028, and even then, it will just be the initial ramp. It will gather momentum in the years that follow.”
“It will take some time for the industry to find a new equilibrium point. Exactly when that happens, we don’t have line of sight on yet.”
MU-0.54%
post-image
  • Reward
  • Comment
  • Repost
  • Share
$MU on Deep Memory Co-Design With Its Largest Customers:
“We are becoming embedded in the R&D roadmaps of many of our customers. They are bringing us increasingly ambitious ideas, and we are actively working on projects that stretch the capabilities of what memory can deliver.
These deep co-design relationships are difficult to replicate across multiple suppliers. Customers can typically partner this closely with only one or two companies.When you go down this path, the relationship starts to look much more like an ASIC engagement.
The customer takes greater ownership of the design, the suppl
MU-0.54%
post-image
  • Reward
  • Comment
  • Repost
  • Share
You’re sidelined? $NVDA says it’ll grow 70% while supply is constrained, and you’re sidelined?
NVDA-4.05%
post-image
  • Reward
  • Comment
  • Repost
  • Share
The implications of $NVDA projecting 70% growth in FY2028 are massive
That would put $NVDA at around $730B in revenue and $420B in operating profit
At that level, $NVDA would be trading at just 12x 2028 EBIT
How crazy does that sound?
NVDA-4.05%
post-image
  • Reward
  • Comment
  • Repost
  • Share
$NVDA could announce $2 trillion in operating profit and the majority purchase of Mainland China, and the stock would dip 2% after hours
NVDA-4.05%
  • Reward
  • Comment
  • Repost
  • Share
@mikealfred is getting nervous
Look at the post he just deleted
The reason? He got called out by @matthew_sigel for conflicts of interest
He probably won’t keep showing off the $2M ARR now
$IREN shouldn’t have a board member who runs a paid investor group
It’s simply absurd
IREN-13.70%
  • Reward
  • Comment
  • Repost
  • Share
One has to wonder:
If OpenAI says AI played a key role in designing Jalapeño, what is the new potential for other chip designers?
One would imagine that $NVDA or $AMD have the resources and talent to take this kind of AI co-design to the next level
I think we'll most likely see generation gaps shorten as far as chip design goes, but manufacturing will remain the bottleneck
Sadly, you can't solve hybrid bonding issues or poor yields as easily as you can design new architectures
CHIP2.90%
NVDA-4.05%
AMD-1.25%
  • Reward
  • Comment
  • Repost
  • Share
YMTC already has a 14% market share
China is really growing fast, wowΩ
post-image
  • Reward
  • Comment
  • Repost
  • Share
It’s crazy that OpenAI achieved these results with such an underdeveloped software stack
SemiAnalysis saw more than 2x throughput improvements at some operating points in under two weeks, and OpenAI enabled TP32 just eight days after earlier TP8 configurations
Jalapeño is already at:
· DeepSeek R1: 700+ tok/s/user
· GPT-OSS: ~1,400 tok/s/user
· Kimi K2.5: nearly 700 tok/s/user
And that’s without speculative decoding
Speculative decoding alone can potentially increase token generation speed by another 2–3x
OpenAI’s software stack is still in its infancy. They’re building it from zero
There’s so
DEEPSEEK-19.53%
  • Reward
  • Comment
  • Repost
  • Share
Samsung Foundry is getting a boost from Nvidia’s Groq 3 inference accelerator, manufactured on Samsung’s SF4X 4nm process
Groq 3 wafer input is estimated at 10,000 wafers per month, rising above 15,000 in 2027, implying at least 50% volume growth. Samsung’s 4nm yields have reportedly improved to above 80%, easing a major production constraint
Samsung’s 4nm capacity is already at full utilization, supported by Qualcomm, HBM base dies, and AI chips. Samsung also reportedly raised SF4 pricing by 5–15% in July, improving the economics of the node
Groq 3 LPX is now in full production, with Nebius a
post-image
post-image
  • Reward
  • Comment
  • Repost
  • Share
$NVDA created the neocloud industry from 0 by leveraging the GPU shortage
It took massive confidence in its moat, but NVIDIA knew it had to take advantage of its quasi-monopoly
$NVDA managed to get GPU-starved hyperscalers to strike huge deals with these smaller companies
Hyperscalers are effectively funding their future competition, lowering barriers to entry by helping neoclouds reach massive scale
That’s also what makes NVIDIA’s strategy so impressive. It deliberately created this neocloud tier from zero
$NVDA could’ve sold almost all of its capacity directly to hyperscalers and called it a
NVDA-4.05%
post-image
  • Reward
  • Comment
  • Repost
  • Share
$NVDA IN TALKS TO INVEST IN PERPLEXITY AT $30B+ VALUATION
According to The Information, Nvidia is reportedly in talks to invest in Perplexity as part of a funding round that would value the AI startup at more than $30 billion, up more than 50% from its $20 billion valuation last year
Perplexity’s annualized revenue has surged to more than $750 million, from less than $250 million at the start of 2026, partly driven by Perplexity Computer, its cloud-based AI agent for automating professional tasks
Earlier this year, Perplexity also signed a $750 million agreement with Microsoft to use Azure and
NVDA-4.05%
MSFT2.37%
post-image
  • Reward
  • Comment
  • Repost
  • Share
X had a period of enlightenment
For a few months, investors joined forces and dug into the semiconductor supply chain like never before
Granted, there was grift, sensationalism, and lies, but what cultural revolution didn’t leave a few bodies behind?
  • Reward
  • Comment
  • Repost
  • Share
$GOOG and $AMD are pushing toward specialized AI inference chips that reduce reliance on GPUs and HBM
Google’s Frozen v2, planned for 2028, reportedly delivers 6–10× higher token throughput than current TPUs by hardwiring frequently used Gemini inference functions into silicon
AMD is going further with Taalas, which embeds model weights directly into the chip, potentially eliminating much of the memory-transfer bottleneck and sharply reducing inference costs. The downside is that the hardware becomes tied to a specific model
AI inference could shift from general-purpose GPUs toward a mix of GP
GOOG1.46%
AMD-1.25%
post-image
  • Reward
  • Comment
  • Repost
  • Share
DIGITIMES: A MARKET SHIFT IN AI SPENDING COULD BE IMMINENT, PAID CONVERSION STUCK AT 3%
DIGITIMES deputy director Jason Tsai noted at the "AI on Chips: Semiconductor Industry Trends Forum", that in the US market AI adoption is high, but conversion to paid users remains low at 3%, with average spending hovering at just $20 per user
Furthermore, as enterprises increasingly deploy AI to optimize products and services, token fees paid to cloud service providers (CSPs) have become a significant financial drain. If the value generated fails to offset these rising token costs, a market shift could be
post-image
  • Reward
  • Comment
  • Repost
  • Share
Social media has made us afraid of being wrong
In real life, you’ll talk, say some incorrect stuff, and learn from each other through conversation
That’s being human
On social media, if you’re wrong, even slightly, you get insulted and mocked
There’s no place for being human
  • Reward
  • Comment
  • Repost
  • Share
  • Pinned