Memory is no longer a commodity
At least not completely
And this trend is only accelerating
Ten years from now, memory won't be considered any more of a commodity than what $TSM or $INTC sell
HBM today is the early-stage part of this trend, with China as proof
CXMT is only achieving a 25% yield on HBM3, versus 90%+ for SK Hynix
They're struggling to translate their DRAM manufacturing know-how into HBM because of its much higher complexity
That's even considering that CXMT uses 3,000 TSVs per HBM3 stack, compared with 8,000+ for SK Hynix, which simplifies manufacturing but can limit bandwidth
M

TSM-1.60%

INTC-5.60%

SK Hynix-0.16%

SKHY-5.29%

SKHYV-0.98%

DanielRomero 3 hours ago

$NVDA AI Demand Is Creating a New Bottleneck in High-End Chip Testing per Digitimes
AI chip demand is creating a shortage of high-end chip testing equipment, especially probe cards and test sockets
Nvidia is reportedly paying higher prices to secure a large share of available capacity, which is forcing TSMC and other chip companies to look for more suppliers
The problem is especially bad during the early stages of chip development, when companies are still testing prototypes and checking whether the design works properly. At that point they may only need 5–100 custom test substrates, but tradi

NVDA-2.25%

TSM-1.60%

DanielRomero 18 hours ago

DIGITIMES reports that AI server demand remains very strong, but component mismatches are worsening, with supply gaps reaching 15–20% across memory, CPUs, PCBs and MLCCs
The issue is expected to persist through September
Vera Rubin has already begun production shipments, with server volumes expected to ramp further in Q4 2026 and become a major revenue contributor in 2027. Demand also remains strong for Nvidia H- and GB-series systems, ASIC servers and general-purpose servers
ODM results reflect that strength. Quanta August revenue rose 177.5% YoY, Inventec 35.8%, and MiTAC 119.9%. Quanta and

NVDA-2.25%

DanielRomero 18 hours ago

$AAPL built a $5T company by taking existing products and making them 10% better
They haven’t done anything truly revolutionary since the iPhone, but making consumers feel safe spending their hard-earned cash on Apple products has been more than enough

AAPL+3.55%

2
1
2
2.67k

DanielRomero a day ago

$AMD at the Citi Global TMT Conference
“We haven't given a ton of details yet, but we do have some of our own internal silicon ambitions for ultra-low-latency inference that would fit into our architecture via chiplets.”
The Taalas acquisition and the Cerebras partnership are just $AMD’s baby steps into ultra-low-latency inference
The goal is to develop this silicon in-house going forward, and the trend we’ll likely see is increasingly specialized ASICs designed for specific workloads

AMD-3.36%

1
3
2.5k

DanielRomero a day ago

How in this world did $GOOG fall so far behind in the AI race?
We’ve heard about the cultural problems within the company for a long time, but they seem to be getting worse after the latest AI leadership shakeup
They had the head start, the most resources, and some of the earliest leading models
It’s insane that $META surpassing them is even possible

GOOG+0.63%

META-1.38%

1
1.56k

DanielRomero a day ago

2Q26 Branded DRAM Supplier Market Share
1. Samsung — 39.4% market share, revenue up 63.4% QoQ
2. $SKHY — 24.9% market share, revenue up 37.9% QoQ
3. $MU — 23.3% market share, revenue up 65.5% QoQ
4. CXMT — 9.5% market share, revenue up 99.3% QoQ
Source: TrendForce

SKHY-5.29%

MU-4.62%

2
2
2.46k

DanielRomero 2 days ago

$AMD Says MI500 Will Offer Both NPO and Copper Options
AMD’s MI500 series, set to be introduced in the second half of 2027 and become its primary product in 2028, will feature scale-up domains larger than the current 72-GPU configuration
For scale-up connectivity, AMD says it will offer both copper-based and near-package-optics-based options, with the two technologies expected to coexist for multiple generations

AMD-3.36%

1.22k

DanielRomero 2 days ago

Fujitsu aims to make practical quantum computing possible by 2030–2035
The company plans to scale its superconducting systems from 256 qubits to 1,024 and eventually 10,000+ physical qubits, while targeting 250 logical qubits by 2030 and 1,000 by 2035

DanielRomero 2 days ago

You can only go all in on one industry and hold for 10 years
Which industry are you choosing, and which stocks are you picking?

DanielRomero 2 days ago

With all due respect...
Beware of people who act like they can comment on every stock
Trust those who focus on a particular industry and are reluctant to step outside of it
Too many people act as if they can analyze every company
You can’t.
It’s not just about revenue, profit, and the balance sheet. You need to understand the fundamental aspects of each industry and even each company
It goes well beyond the basics

DanielRomero 2 days ago

Let’s do a thought experiment
Every year, there are stocks that 10x
If I told you to guess which stock will 10x, not in 1 year, but within the next 5, what would it be?
Filter out all the “good” or “decent” stocks
What’s the actual 10x opportunity that comes to mind?

DanielRomero 2026-09-07

Europe has signed a $451 million contract for LUMI-AI, a new AI supercomputer in Finland that is set to be deployed in H2 2027
It will use $AMD Instinct MI430X GPUs and 6th-gen EPYC CPUs, making it a major sovereign compute win for AMD
LUMI-AI is expected to deliver 10x the AI capacity and nearly 2x the HPC throughput of the current LUMI system
It will also use Bull’s European-designed BXI interconnect instead of NVIDIA NVLink, reducing dependence on NVIDIA’s ecosystem

AMD-3.36%

NVDA-2.25%

BULL-1.37%

DanielRomero 2026-09-07

GPT-6 Astra could become a major new demand driver for CPUs
Astra isn't just a better chatbot. It's designed to actually operate computers
OpenAI says it can autonomously install and test software, troubleshoot applications, run frontend QA, interact with websites, fill forms, update CRMs, and execute complex multi-step workflows
The improvement is very significant
On OSWorld 2.0, a key benchmark for measuring how well AI agents can operate computers and complete real multi-step tasks, Astra reaches 72.6% while finishing tasks in 40 minutes, versus 65.7% and 75 minutes for GPT-5.6 Sol
Combined

AMD-3.36%

INTC-5.60%

DanielRomero 2026-09-07

China is trying to make 3D DRAM possible without EUV
Naura Technology has demonstrated a key etching process for 64-layer 3D DRAM, potentially giving China a way to keep scaling DRAM density despite lacking access to EUV lithography
Instead of making transistors smaller in 2D, 3D DRAM stacks them vertically, similar to what NAND did years ago. Naura’s process selectively removes silicon-germanium layers from a 64-layer Si/SiGe stack while preserving the silicon structure
The reported performance is strong: SiGe-to-Si selectivity above 500:1, silicon loss below 1 nm per cycle, and more than 95%

DanielRomero 2026-09-05

Rosenblatt Securities on $CIFR:
- CIFR is down 25% since Texas froze large-load interconnection approvals
- The issue is time to power, not a lack of demand
- CIFR plans up to 2.5 GW of behind-the-meter power generation across 10 Texas sites
- This could bypass ERCOT delays and accelerate data center construction
- It should not require major CIFR-funded CapEx
- West Texas has very cheap, sometimes negative, natural gas prices, making on-site generation attractive
- Rosenblatt believes CIFR could turn its Texas exposure from a weakness into an advantage
- BUY, $30 price target, based o

CIFR-5.79%

DanielRomero 2026-09-05

Ping Yeh, a software engineering manager at $GOOG Quantum AI, believes useful quantum computing is more than five years away
Hartmut Neven, who heads Google Quantum AI, has said recent results leave him confident that useful applications only quantum computers can handle will arrive within five years
“He is very confident,” Yeh said at the SEMICON forum. “I am a little skeptical. I want to be more conservative.”