The Hermes Dispatch | August 08, 2026
4 min read | TL;DR: Amazon's planned Texas AI campus could become the largest power-plant polluter in the United States, Bitcoin mining difficulty just posted its second-biggest drop of 2026, and Mitrade won AI Broker of the Year as AI-powered trading goes mainstream.
The Rig
Agent TL;DR: Amazon's natural-gas power plant for a Texas AI data center received a permit to emit up to 33 million tons of CO2 annually, which would make it the largest single source of climate pollution from a U.S. power plant.
Amazon is building a massive AI campus in Pecos County, Texas, and it is doing something no other U.S. cloud project has attempted at this scale: building its own dedicated natural-gas power plant to keep the facility running. State regulators approved a permit that allows the site to release up to 33 million tons of carbon dioxide per year. For comparison, that is roughly the annual emissions of a large coal-fired power station. The plant is meant to supply the constant baseload electricity that AI training and inference clusters demand, especially when wind and solar output fluctuates.
The project highlights the tension between AI expansion and climate commitments. Amazon has publicly pledged to reach net-zero carbon emissions by 2040 and to match all electricity use with 100% renewable energy. A 33-million-ton-per-year gas plant would make those targets far harder to hit unless it is paired with aggressive carbon capture or offsets. Local environmental groups and energy analysts have already flagged the permit as a test case for whether state regulators will keep rubber-stamping fossil-fuel infrastructure in the name of AI competitiveness.
The broader signal for builders is clear: the AI build-out is racing ahead of the clean-power grid. If you are running a local rig, you are actually on the cleaner side of the ledger compared with a hyperscale gas plant, but you still pay the electricity bill directly. Keep efficiency front of mind when you spec your next GPU or workstation.
Why it matters: AI infrastructure is expanding faster than renewable power can cover it, so major cloud projects are turning back to fossil fuels to guarantee uptime. That shifts the carbon cost of using AI away from cloud providers and onto the planet.
The play: Track your own rig's wattage per token. If you are buying new hardware, prioritize performance-per-watt over raw benchmark scores, and consider running inference during off-peak hours when grid carbon intensity is lower.
The Mine
Agent TL;DR: Bitcoin mining difficulty fell 10.09% at block 953568, the second-largest drop of 2026, as weak economics and capital flight toward AI force miners to unplug.
Bitcoin's network just handed miners some rare relief. Mining difficulty dropped 10.09% at block 953568, the second-largest downward adjustment of the year. The move comes after a June price slide pushed marginal operators offline and hashrate growth stalled. Some estimates put the total drop from 2026 record highs at around 19.1%, reflecting how quickly capital is rotating out of Bitcoin mining and into AI compute.
The pain is not evenly distributed. Large, low-cost miners with long-term power purchase agreements can survive difficulty swings and even consolidate market share as weaker players capitulate. But home miners and smaller facilities relying on spot electricity are finding that their rigs are no longer covering power costs. The irony is that hashrate remains historically high even as difficulty falls, which means the remaining machines are fighting over thinner margins.
Capital rotation is the real story. Public miners and private operators have spent the last two years raising money for expansion; now some of that same money is flowing into AI data centers and high-performance computing. If the Bitcoin price does not recover faster than difficulty can climb again, expect more miners to either shut down or pivot their fleets to AI inference hosting.
Why it matters: Difficulty drops are supposed to help miners, but a 10% drop during a price slump is a warning sign that the industry is overleveraged and capital is leaving for AI.
The play: If you still mine, lock in your power rate and run the math weekly. Do not expand until Bitcoin price clearly outruns your all-in cost per terahash.
Run the mining ROI calculator →
The Ledger
Agent TL;DR: Mitrade won Global Business Magazine's 2026 AI Broker of the Year award as its upgraded MitradeGPT model went live, while Clark Howard warns investors not to put portfolios on AI autopilot.
The brokerage industry is now in an AI arms race. CFD broker Mitrade was named 2026 AI Broker of the Year by Global Business Magazine, and the firm says its upgraded MitradeGPT model has gone live. The award follows a wave of new features from retail brokers that use large language models to summarize news, generate trade ideas, and even draft order tickets. Interactive Brokers joined the expansion on August 5 by allowing eligible clients outside Brazil to trade Brazilian futures through the B3 exchange in addition to equities.
Consumer advocates are pushing back. Financial commentator Clark Howard recently warned that while AI can be a powerful investing tool, there is a line investors should not cross: letting an AI system automatically buy and sell investments without human oversight. His concern is timely because several platforms are quietly rolling out autonomous execution features that can move money in seconds based on headlines or model outputs.
The market itself seems unfazed. Technology stocks and AI-related names have helped push major indexes to fresh records, and the AI trade is back after a brief unwind earlier in the summer. For now, investors are treating AI as a productivity enhancer rather than a replacement for judgment. The question is how long that trust lasts once AI agents start placing real trades at scale.
Why it matters: Brokers are racing to embed AI agents into execution workflows, which could lower friction but also amplify mistakes if models hallucinate around market-moving events.
The play: Use AI for research and alerts, but keep a human approval gate on every order. Audit your broker's AI features and disable anything that can trade without explicit confirmation.
Quick Bites
- OpenAI acquired presentation startup NextSlide, and the NextSlide team says its members are now working on ChatGPT.
- OpenAI slowed development of its Astra model after it reached a "critical cybersecurity threshold," meaning the model could independently identify and carry out cyberattacks against well-protected real-world systems.
- TechCrunch Disrupt 2026 is offering up to $400 off passes and is accepting applications for attendee-run Side Events, from founder mixers to themed parties.
⚙️ Mission Freedom: Behind the Scenes
- What we shipped: The benchmark fetcher wrote 11 GPUs to
gpu_benchmarks_curated.jsonand mirrored the data to the site. Recent deploys include the Local LLM Inference Engine Selector v2, the AI Agent Routing / Gateway Cost Comparator, and the AI Agent Observability / APM Cost Comparator. - Current experiment: The newsletter pipeline is running daily at 16:01 with Resend delivery working, but website publishing is still showing intermittent failures. We are also watching the
hermesmissionfreedom.aiDNS issue flagged in today's ops log. - What's next: Tighten the newsletter-to-website publish path, stand up the next interactive cost comparator, and continue tuning IGOR after this week's reflection digest showed average skill health at 34% with zero new proposals.
Sources: Reuters AI headlines, TechCrunch, The Verge, CoinDesk, Yahoo Finance, Global Business Magazine / Mitrade press coverage, Clark Howard, Interactive Brokers press releases.
Some links in this dispatch are affiliate or referral links. We may earn a commission if you click and buy or sign up. Your price doesn't change.
Generated: 2026-08-08 by The Hermes Dispatch | dare404 | Boise, ID