IREN $IREN partners with ElectraNet to secure high voltage grid connections for a new 800MW AI campus in South Australia.

The strategic infrastructure project leverages direct submarine fiber links to connect global markets.

Morgan Stanley doubles the price target on Micron $MU to $1,050 while maintaining an Overweight rating.

Analysts predict a severe, industry wide AI memory shortage will persist for at least 2 to 3 years.

USA Rare Earth $USAR finalizes a massive $1.6B federal funding package under the U.S. CHIPS Program.

The strategic capital injection accelerates the development of a core heavy rare earth mining project in Texas.

Benchmark reiterates a Buy rating on USA Rare Earth $USAR and maintains a $45 price target.

Analysts note that planned capacity expansions are already factored into a model assuming 10,000 tons of annual magnet output by 2030.

$USAR/USD H2 #Indicator #Signal #Trading #GlobalOneAI
Globalone.ch #Prompt #AI

Backtesting and feature engineering for quantitative finance (free on GitHub):

github.com/hudson-and-thames/mlfinlab

Here's what's inside:

Backtest Overfitting Tools
Data Structures
Labeling
Sampling
Feature Engineering
Models
Clustering
Cross-Validation
Hyper-Parameter Tuning
Feature Importance
Bet Sizing
Synthetic Data Generation
Networks
Measures of Codependence
Useful Financial Features

🚨UN researchers say data centers could double power and water consumption by 2030 as AI demand accelerates.

The numbers are huge:

2024 electricity use: 448 TWh
2030 projection: 945 TWh

2024 water use: 4.5 trillion liters
2030 projection: 9.3 trillion liters

QuantMind turns every paper, news article, SEC filing, and blog post in quant finance into a queryable knowledge graph.

Hedge funds pay analysts six figures to do this manually.

github.com/LLMQuant/quant-mind

🎞️ DGX Spark Super Gallery at Computex. Product breakdown 👇

Before the RTX Spark there was the DGX Spark. Not to confuse the two, one runs Windows and the other Linux. Here is a breakdown of the GB10 mini-desktops.

1. NVIDIA DGX Spark (Founders Edition)
🖤 NVIDIA Official
💰 $4,699
🧠 GB10 Grace Blackwell Superchip
💾 128GB Unified LPDDR5X
📀 4TB NVMe SSD (self-encrypting)
⚡ 1 PFLOP FP4 AI Performance
Perfect for local 200B parameter models. 🔥

2. MSI DGX Spark
💰 ~$4,599–$4,699
🧠 GB10 Superchip
💾 128GB Unified Memory
📀 4TB NVMe SSD
Compact & premium design with strong cooling. Great for developers. ⚙️

3. ASUS DGX Spark
💰 ~$3,999–$4,699
🧠 GB10 Grace Blackwell
💾 128GB LPDDR5X Unified
📀 4TB Storage
Excellent build quality and connectivity. Popular with creators. 💻

4. Acer Veriton GN100 DGX Spark
💰 ~$4,100
🧠 GB10 Grace Blackwell Superchip
💾 128GB Unified LPDDR5X
📀 4TB NVMe SSD
Strong thermal performance in a compact chassis. Excellent value option. 🟢

5. Gigabyte AI TOP ATOM DGX Spark
💰 ~$4,699
🧠 GB10 Superchip
💾 128GB Unified Memory
📀 4TB NVMe SSD + ConnectX-7 SmartNIC
Often praised for best-in-class performance and cooling. ⚡

6. HP DGX Spark
💰 ~$6,000–$6,599
🧠 GB10 Superchip
💾 128GB Unified Memory
📀 4TB NVMe
Enterprise-grade reliability with HP’s signature design. 🏢

7. Lenovo ThinkStation DGX Spark
💰 ~$6,000
🧠 GB10 Grace Blackwell
💾 128GB Unified RAM
📀 4TB Storage
Built for professional workstations and heavy AI workloads. 💼

8. Dell Pro Max DGX Spark
💰 ~$6,300 (config dependent)
🧠 GB10 Superchip
💾 128GB LPDDR5X
📀 4TB NVMe SSD
Strong enterprise integration and support. Ideal for businesses. 🏭

WHICHLLM scans your hardware and ranks the best local models from HuggingFace that will actually run on your system.

> Evidence-based ranking from merged benchmarks, not just VRAM size
> Simulate any GPU before buying with a single flag
> Recency-aware scoring so stale leaderboard entries don't mislead you

Stop guessing which model fits your machine.

Github: github.com/Andyyyy64/whichllm

⏰ Oh yeah super excited about this! 🚀 A new local 12B dense model from Google! You have to try this out.

💥 Google just dropped Gemma 4 12B → a powerful multimodal open model that actually runs well on laptops (and cards).

Run on 👉 ✦ RTX-4070 TI ✦ RTX-5060 TI ✦ RTX-5070 TI ✦ RTX-5080

Key Highlights:
🧠 Encoder-free architecture (35M projector + native LLM handling)
💻 Runs smoothly on 16GB VRAM / unified memory machines
📸 Strong vision + audio understanding
⚡ Excellent reasoning performance for its size
✅ Fully open source (Apache 2.0)

🔥 Finally, a capable multimodal model you can actually run locally without a datacenter. This changes the game for on-device agents.

💪Wow! New open source Computer Use model shows strong local performance on 📊 LLM Leaderboard using a single DGX Spark!

This model can handle your agents 💻 Computer Use workload fast!

🆕 Holo-3.1-35B-A3B-NVFP4 (vLLM + NVFP4)

» 101 tokens/sec at low concurrency
» Scales up to ~195 tokens/sec at higher concurrency (c10)

🎯 A 35B-class model delivering nearly 200 tokens per second on a single DGX Spark node for 💻 Computer Use is impressive. Great efficiency for a local agentic model. 🔥

🤔 how Mac/StrixHalo/RTX Card would do?

spark-arena.com/benchmark/2bf2209e-0827-4de6-9df9-8ad70ae1b618

🚨Copper just hit $14,000/ton

Goldman, Citi and HSBC all published bullish notes today.

And Codelco, the world's largest copper producer, just fired its CEO over inflated production figures.

The supply crisis just got a leadership crisis on top. ⛏️

3 things happened simultaneously today that every copper investor needs to understand:

1️⃣The banks officially joined the trade:

→ Goldman Sachs: lifted end-2026 target +10% to $13,735 cut global mine supply estimates by 350,000 tons

→ Citi: $14,500 this month. $15,000 within a year.

→ HSBC: overlaid the Hormuz "super-squeeze" thesis on top of the structural deficit

Ex-US copper deficit projected to balloon 10x to 640,000 tons. 💰

2️⃣The supply problem is structural:

→ Grasberg (world's 2nd copper mine): still recovering from September 2025 underground flood. Force majeure still active. Full capacity: not before 2028.

→ Kamoa-Kakula: cut 2026 guidance by 22% after seismic disruption. Same timeline.

2 of the world's most important copper mines simultaneously impaired for multiple years.

On the demand side: nothing has softened.

Grid expansion, EVs, AI data centers, defence all absorbing copper faster than the market can supply⚡

3️⃣CODELCO – Corporación Nacional del Cobre de Chile the world's largest copper producer just changed CEO:

Jorge Gomez named to replace Rubén Alvarado who exits amid a scandal over inflated production figures.

This is Chile's state copper giant.

397,000 tonnes in Q4 2025.

Second only to BHP globally.

When the world's largest copper producer has a production credibility problem in the middle of a structural supply squeeze that is is signal. 🌍

Alphabet is about to do the largest equity sale in history.
Bigger than SpaceX's IPO.

Bigger than Petrobras in 2010.
Nearly 3x Aramco's 2019 IPO.

$80 billion, in a single equity raise.

The previous record holders were both oil companies at the peak of their cycles.

🇧🇷Petrobras in 2010: $65 billion to drill the pre salt ultra deepwater fields off Brazil the biggest oil discovery in decades. The market couldn't get enough.

🇸🇦Aramco in 2019: $25 billion to monetise the world's largest oil reserve. The most profitable company on earth going public. Still the largest IPO ever at the time.
Both were energy. Both were oil. Both represented the dominant capital deployment thesis of their era.

Now Alphabet an AI company, is about to raise more than both combined.
Not to drill. Not to pump. To build compute.

The capital cycle has rotated completely.

2010s: the world's largest equity raises funded oil megaprojects deepwater, pre-salt, sovereign oil reserves.

2026: the world's largest equity raises fund AI infrastructure data centers, GPUs, power.

🚨Private data center construction just crossed public transportation infrastructure.

$50 billion.

It took roads and bridges 20 years.

Data centers did it in 2.

$50 billion in construction per year needs electrons.

The grid took the same 20 years to build its current capacity.

It cannot absorb a vertical demand curve on a 2 year timeline.

Grid interconnects: 5-7 year queue.

Data center construction: vertical.

That gap is where the private gas turbine market lives.

Hyperscalers aren't waiting for the grid.

They're building behind the meter mobile turbines on site in 12-18 months.

The binding constraint isn't capital.

It is gas & It's turbines.

How to build insanely powerful agent finance skills with Hermes.

Hermes is the best AI agent ever built.

And one of its best use cases is for deep financial research.

If you inject this prompt into your agent, it builds custom agentic finance skills.

You'll want to use this

Broadcom is extremely undervalued and two things just happened today that make the case impossible to ignore (Save this).

Broadcom reports earnings tonight and Google just raised $85 billion to pour directly into the infrastructure Broadcom builds.

Sundar Pichai announced this morning that Alphabet's equity offering was oversubscribed and upsized from $80 billion to $84.75 billion in 48 hours.

Berkshire Hathaway put in $10 billion directly.

The stated purpose is to fund AI compute infrastructure because demand from enterprises and consumers is already exceeding available supply.

Google Cloud just posted $20 billion in quarterly revenue up 63% year over year, the fastest growth since the segment began reporting separately.

Enterprise AI solutions grew 800% in the same quarter and the backlog nearly doubled quarter over quarter to $462 billion in signed customer commitments.

Every dollar of that $462 billion backlog requires physical infrastructure to deliver, TPUs, networking silicon, custom accelerators and the design and supply chain for all of it runs through Broadcom.

Sundar Pichai just raised $85 billion to pour into the exact infrastructure Broadcom builds and services through 2031 under contract.

Now here is why Broadcom specifically is the cleanest expression of this trade.

Broadcom does not sell merchant silicon into a commoditizing market but rather it co designs proprietary custom accelerators, XPUs directly with hyperscalers who cannot source these chips anywhere else.

And their customer list reads like a directory of the most consequential AI companies on earth, Google, Meta, Anthropic, OpenAI, Fujitsu, ByteDance.

Google locked in Broadcom through 2031 for TPU design and supply and then immediately raised $85 billion to accelerate spending under that same contract.

Anthropic is receiving 1 gigawatt of TPU capacity this year, scaling to 3 gigawatts in 2027 and OpenAI's first custom XPU ships in 2027 with over 1 gigawatt of committed capacity.

The networking side of the business is just as important and even more overlooked.

AI networking revenue grew 60% year over year last quarter and represents a third of total AI revenue.

Broadcom's Tomahawk 6 switch and 200G SerDes components are the connective tissue inside every hyperscaler's GPU cluster and the order backlog for AI switches alone exceeds $10 billion.

Now here is the earnings setup for tonight.

Analysts are expecting $22 billion in revenue, 47% year over year growth and $10.7 billion in AI chip revenue alone, up 140% from a year ago.

Tonight's earnings are the moment the market catches up to what the balance sheet already knows.

Uber's CEO said it himself that they blew through their entire 2026 AI budget in a single quarter (Save this).

And today they cut 23% of the people who used to do what AI now does.

These two facts are not separate stories but rather the same story.

Earlier this year, Dara Khosrowshahi explained exactly what was happening inside Uber's engineering org.

The company introduced Anthropic's Claude Code to engineers in late 2025, adoption reached 32% by February, 84% were classified as agentic coding users by March and by April 95% of engineers were using AI tools monthly.

Roughly 70% of committed code was AI generated and round 11% of real time backend updates were being deployed by autonomous agents with no human in the loop.

The annual AI budget was gone in four months and Dara's response was not to slow down adoption but rather to slow down hiring.

When each engineer is producing materially more output per hour, you need fewer engineers to hit the same targets.

And when AI handles the work that used to require people to hire, onboard, and manage those engineers, the HR and people operations layer beneath them, that layer contracts too.

Today's announcement is that 23% of Uber's People and Places division is gone.

Uber's spokesperson said the cuts are unrelated to AI but that framing does not hold up to the sequence of events.

IBM understood this two years ago, when 94% of typical HR questions are answered by an AI agent, the HR Business Partner role collapses for everyone except the most senior strategic functions.

The budget that used to fund that team gets reallocated to engineering and sales, the two functions that remain bottlenecked by human judgment rather than human volume.

This is the pattern that will repeat across every major enterprise over the next 18 months.