Pinned Tweet
Is DePIN Dying? No, No, No. You Should Check Out the Most Unmissable DePIN Industry Opportunities in 2026. @arkreen_network
Levi Strauss & Co. was founded in San Francisco in 1853; Strauss and tailor Jacob Davis patented riveted pocket openings on May 20, 1873.
The 173-year-old denim maker is enjoying a comeback as shoppers rediscover jeans with low rises and loose fits. on.wsj.com/3VL1iV5
58
Heinrich Zimmermann held the first World Animal Day in Berlin on March 24, 1925; a 1931 congress in Florence adopted October 4 as its date.
28
The cook-off is held in Marlinton, West Virginia during its Autumn Harvest Festival; the first one ran in 1991.
Appalachian cuisine prizes thrift and resourcefulness: hunting, drying and pickling all feature heavily. For 35 years Pocahontas County has honoured those traditions with an annual “Roadkill Cook-Off” economist.com/culture/2026/1…
28
Europe alone sold about 3 million heat pumps in 2023; the IEA's 53 bcm figure covers Europe, Japan and China together.
Heat pump deployment can help strengthen energy security & resilience by reducing countries’ reliance on fuel imports for heating.
In 2025, heat pump installations avoided around 53 bcm of natural gas use for heating across Europe, Japan & China.
More 👉 iea.li/3TUE7Hr
21
Per Citadel, MSFT, NVDA, AAPL and META added about 300 S&P points in Q3, while the other ~499 stocks took away roughly 150.
MSFT, NVDA, AAPL and META alone added about 300 points to the S&P in Q3, more than 200% of the index's total gain. The other 499-ish stocks together subtracted about 150 points: Citadel
23
Karl Ernst Claus isolated ruthenium at Kazan in 1844 and named it for Russia; annual output was only about 35.5 tonnes in 2017.
Ruthenium has risen from about $560 an ounce in early 2025 to dealer quotes near $1,950 this September. It is a key material in hard drives, with new use cases also emerging. (1/5)🧵
24
Columbia's Stijn Van Nieuwerburgh puts the bar at about $3.5 trillion a year of AI services by 2032, roughly 8.8% of US GDP.
AI BOOM MAY REQUIRE U.S. SPENDING EQUAL TO 9% OF GDP
America may need to spend roughly $3.5 trillion annually on AI services by 2032 — 8.8% of GDP — to justify today’s massive data-center investment, according to Columbia professor Stijn Van Nieuwerburgh.
The risk is that expanding compute capacity drives AI prices and margins sharply lower, leaving revenues unable to support current investment.
AI could still deliver major productivity gains — but economic benefits don’t necessarily translate into investor returns.
32
The AI buildout has a physical side: Flexport's CEO puts data-center parts at 8% of Asia-to-US air freight, 30 chartered freighters a day.
8% of the air freight entering the US from Asia is components for data centers.
30 full freighters per day just in chartered volume, a lot more shipping less than plane loads at a time.
20
HASTE is open source on GitHub and has backed over 30 disaster responses since 2023, turning satellite images into per-building damage maps.
When disaster strikes, the speed and accuracy of a response can make all the difference. HASTE helps responders assess damage and pinpoint the hardest-hit areas within hours instead of weeks, helping get aid where it’s needed most.msft.it/6002acswm
16
MI355X: CDNA 4 on TSMC 3nm, 288GB HBM3E, 8TB/s, shipping since June 2025. Vendor-hosted RAG test, so the 42% stays a claim until replicated.
More users shouldn't have to mean slower AI.
In @Signal_65 RAG testing hosted via @TensorWave, AMD Instinct MI355X delivered ~42% lower p99 latency and roughly 2x the concurrent-user headroom before breaching the evaluated SLA: bit.ly/45hAvl3
1
45
462,000 would sit about 7% under the 497,099 record of Q3 2025, a quarter pulled forward by the $7,500 US credit that ended that Sept 30.
Tesla has just released their company compiled Q3 analyst delivery consensus.
Analysts expect @Tesla to deliver 462,000 vehicles in Q3 2026, and deploy 15.9 GWh of energy storage this quarter.
12
Germany logged 457 hours of negative wholesale prices in 2024, up from 301 in 2023: 5% of the year with nowhere to put the surplus.
Negative electricity prices have become a common occurrence in many markets.
Their rising frequency highlights the need for greater system flexibility – including battery storage and demand response – to support reliable, efficient power systems 👉 buff.ly/kYN16Pr
10
Porsche has sold the Taycan since late 2019 and an electric Macan since 2024; the only line it will not cross is the 911.
Porsche boss rejects EV sceptic label as ‘unfair’ — just don’t expect an electric 911 ft.trib.al/flju7wc
18
Eight real agent sessions and 168 turns replayed on DGX Spark, RTX 5090, Ryzen AI Halo and M5 Pro: local agent speed, measured the same way.
Announcing AA-AgentPerf-Local, our open-source inference testing tool for local AI models - test how fast agentic AI can run on your own laptop or workstation, and browse our list of serving configurations to plan your next agent setup
Key points:
➤ We’re open sourcing AA-AgentPerf-Local, which replays real agent trajectories on laptop & workstation hardware to test inference performance
➤ We’re releasing initial results for NVIDIA DGX Spark, NVIDIA GeForce RTX 5090, AMD Ryzen AI Halo, and MacBook Pro M5 Pro
➤ The tool and leaderboard will soon expand to cover more hardware, frameworks and models, and will stay updated over time as new releases launch
We already benchmark inference performance on mobile phones and datacenter-scale hardware, and are expanding our coverage to include laptops and workstations. Alongside hosting a leaderboard displaying results from popular model and hardware combinations, we are open-sourcing all code and data required to run AA-AgentPerf-Local to ensure that individuals and companies are able to run their own trials, informing their local AI serving decisions.
AA-AgentPerf-Local replays real agent sessions. Our default workload is 8 recorded agentic tasks, spanning 168 model turns. Each request carries the full conversation so far, as a real agent's would, so context grows to ~56K tokens. Every turn generates exactly its recorded number of tokens, so every system does identical work. Tool execution is skipped by default to isolate inference speed, but when benchmarking your own system, you can also replay the real recorded tool delays or run tool calls live on your CPU.
At launch, we are focusing on the performance a single agent can achieve when able to utilize the entire system; we plan to expand this coverage over time to cover multi-agent systems and scenarios where an agent must run alongside other regular processes.
The initial set of hardware covered on our official page is: NVIDIA DGX Spark (128 GB), AMD Ryzen AI Halo (128 GB), MacBook Pro M5 Pro (64 GB), and NVIDIA GeForce RTX 5090 (32 GB). These have been selected to cover a range of platforms (CUDA, ROCm, Vulkan, Metal), memory capacities, and bandwidths. We will be expanding the featured hardware to include x86 (and other) laptops, AI-focused graphics cards such as the RTX PRO 6000 Blackwell, and more, to give consumers a well-rounded view of performance across different hardware types.
The models featured at launch are: Qwen3.5-9B, Qwen3.8-27B, Qwen3.6-35B-A3B, and Ling 3.0 Flash (124B / 5B active). These initial models span a range of memory requirements and dense/MoE architectures, and are each benchmarked at 4-bit quantizations to reflect realistic serving conditions. The core set of models we feature will shift over time as new open weights models are released. Beyond the featured models, AA-AgentPerf-Local is able to benchmark performance of any OpenAI-compatible inference server the user runs, meaning that any model and config can be tested locally.
Choice of serving configuration is important, with the runtime, quantization and speculative decoding changing results substantially. Where an official off-the-shelf config was available for a system and model pair, we used the published config. We developed our own configs for all other cases. Every config uses speculative decoding (MTP, DFlash or DSpark), and all 14 are published in the repo and on the configs page on our website.
Initial results:
➤ Completion time mapped most closely to each model’s active parameter count: Qwen3.6-35B-A3B (3B active) was the fastest model on every system, e.g. 2.5-3.3x faster than the dense Qwen3.8-27B. However, active parameters are not the whole story, with Ling 3.0 Flash (124B total, 5B active) still finishing behind Qwen3.5-9B (nearly double the active parameter count) on all hardware that can support it.
➤ The GeForce RTX 5090 was the fastest system for every model that fits in its 32 GB, achieving completion times >3.5x faster than the other systems. Single-user decoding is heavily influenced by memory bandwidth, and the GeForce RTX 5090 has 1,792 GB/s against 256-307 GB/s for the unified-memory systems.
➤ The DGX Spark and Ryzen AI Halo are overall similar systems, with the same amount of unified memory, comparable memory bandwidth, and the same launch MSRP of $4,000. On our default trajectory set, the DGX Spark was 1.4-1.7x faster on three of four models, with the Ryzen AI Halo tying it on Qwen3.5-9B. The gap is far larger than their 7% bandwidth difference - the Spark has greater low-precision compute than the Ryzen AI Halo, enabling it to prefill faster, and its more mature CUDA software likely plays a role in enabling better MoE and speculative decoding performance.
➤ The MacBook Pro (M5 Pro, 64 GB, 20-core GPU) is the only laptop tested so far, and exhibited competitive results, finishing within 2–9% of the Ryzen AI Halo on Qwen3.6-35B-A3B and Qwen3.8-27B (though 21% slower on Qwen3.5-9B). It has the most memory bandwidth of the three unified-memory systems (307 GB/s) and its current price of $3,700 is the lowest of the systems tested so far. Its results are likely held back by software maturity and compute available for prefill.
➤ Despite the agentic trajectories serving 73-93% of prompt tokens from the KV cache, simply reading each turn’s new input (during prefill) used up substantial proportions of the end-to-end completion time, e.g. 22-41% for Qwen3.8-27B. This was especially impactful on systems with low compute FLOP/s relative to their memory bandwidth, such as the Ryzen AI Halo and potentially the MacBook Pro (MacBook FLOP/s are unpublished).
➤ Speculative decoding was implemented on all of the most successful configs so far, e.g. raising Qwen3.8-27B decode speeds ~30-120% above the bandwidth-constraint roofline.
133






























