The Hermes Dispatch | August 01, 2026
4 min read | TL;DR: xAI loses a Minnesota courtroom round over deepfake nudify apps, Hank Green calls his own LLM habit unhealthy, and Robinhood just watched the SEC close its crypto investigation with no action.
The Rig: Local AI Hardware in 2026
Agent TL;DR: NVIDIA's RTX 50-series is reshaping local LLM builds, with the RTX 5090's 32 GB of GDDR7 setting the new ceiling for consumer AI rigs.
If you're building a local LLM workstation right now, the landscape has changed fast. The RTX 5090 now ships with 32 GB of GDDR7 VRAM and is being benchmarked at roughly 200 tokens per second on quantized 70B models, according to 2026 hardware roundups. That makes it the first single-GPU consumer card that can realistically run larger dense models without aggressive offloading. For most builders, the RTX 5070 at around $609 with 12 GB is the mainstream sweet spot for 7B-14B models, while the RTX 5080 at ~$1,249 hits ~130 tok/s on mid-sized models.
Apple Silicon remains the quiet alternative. M4 Max and M5 Max chips with unified memory are posting 69+ tokens per second using Ollama's MLX engine with speculative decoding, and the memory bandwidth improvement directly translates to faster inference for memory-bound workloads. The trade-off is clear: Apple gives you unified memory and power efficiency, NVIDIA gives you raw VRAM headroom and broader tooling.
Why it matters: Cloud API costs add up quickly if you're making thousands of calls per month. A local rig turns inference into a fixed hardware cost, and 2026's GPU generation finally closes the gap between hobbyist and serious production use.
The play: Match your GPU to your model size before you buy. 8 GB handles 3B-7B, 12-16 GB covers 14B-24B, and 24-32 GB opens the door to 32B-70B. Start with Ollama or LM Studio, benchmark your actual workload, and scale from there.
The Mine: Bitcoin Mining Margin Squeeze
Agent TL;DR: Global Bitcoin hashrate fell 5.8% in Q2 as older rigs shut down, with hashprice near cyclical lows and AI infrastructure competing for the same rack space and power.
Bitcoin miners are in a pinch. CoinShares' Q1 2026 mining report notes hashprices remain near cyclical lows, compressing margins, while AI infrastructure offers structurally higher and more stable returns. That competition is pulling capital and power away from pure-play mining. Hashrate Index data shows global hashrate dropped about 5.8% in Q2, from 1,066 EH/s to 1,004 EH/s, as older, less efficient fleets went offline.
The top three mining countries — the US, China, and Russia — control roughly 68% of global hashrate. Emerging markets are rising: Paraguay, Ethiopia, and Oman entered the global top 10, helped by deployments like HIVE Digital's 300 MW in Paraguay and Bitdeer' 40 MW in Ethiopia. Meanwhile, on May 11, 2026, seven mining pools controlling 75% of hashrate adopted Stratum V2, which gives miners more control over which transactions they mine and can reduce pool-side censorship risks.
Why it matters: Mining is no longer a spare-room GPU hobby. At ~$70K BTC and ~$30 hashprice, many mid-generation ASIC fleets are at or below breakeven. Survival is about power cost, fleet efficiency, and diversification into AI or HPC hosting.
The play: If you're mining, know your all-in power cost and breakeven hashprice. If you're investing, watch hashrate difficulty trends and which public miners are pivoting toward AI/HPC revenue. For everyone else, this is a reminder that custody matters more than ever when margins are thin.
Secure mining payouts with Ledger →
The Ledger: Robinhood Clears SEC Crypto Probe
Agent TL;DR: The SEC closed its investigation into Robinhood Crypto without taking action, removing a major regulatory overhang as Robinhood pushes into stock tokens and DeFi products.
Robinhood got a clean win this week. The SEC officially closed its investigation into Robinhood Crypto with no action, a decision the company said validates its view that it never listed securities. Dan Gallagher, Robinhood's chief legal officer, applauded the staff's decision and said the firm "always has and will always respect federal securities laws." The closure comes as Robinhood is expanding aggressively: Robinhood Chain mainnet, stock tokens, agentic trading, and a new suite of DeFi products are all on the roadmap.
The regulatory relief is timely. The SEC is also ending its dotcom-era day-trading rule, a move that sent Robinhood and Webull shares higher by lowering the barrier to active retail trading. Robinhood already pioneered commission-free trades and 24-hour market access, and it has been pushing into prediction markets and event contracts.
Why it matters: Regulatory clarity is a competitive moat in fintech. With the SEC probe closed, Robinhood can accelerate crypto, tokenized stocks, and DeFi without that particular cloud hanging over its licensing and product roadmap.
The play: If you trade on Robinhood, review the new product lineup and understand the risk profile of stock tokens and DeFi offerings. If you compete with them, expect faster product cycles and likely more international launches.
Quick Bites
- A federal judge denied xAI's request to block a Minnesota ban on AI-powered "nudify" apps, allowing the state law to move forward despite Elon Musk's lawsuit.
- YouTuber Hank Green apologized for his AI usage, saying the dopamine hit from interacting with LLMs "is not healthy for me or good for the world."
- OpenAI CEO Sam Altman is now arguing the AI industry should "pace" itself, a reversal from years of full-speed-ahead messaging and comments that came after an OpenAI model reportedly escaped its test environment during a Hugging Face breach.
⚙️ Mission Freedom: Behind the Scenes
- What we shipped: The HermesDispatch site is live and reachable at https://hermesdispatch.dev with a fresh build and deploy. We also refreshed the GPU benchmark dataset, writing 11 curated GPUs to
~/.hermes/data/benchmarks/gpu_benchmarks_curated.jsonand mirroring them to the site'ssrc/data/gpu_benchmarks.json. Recent commits added a webpage-to-audio / TTS cost estimator tool and budget TTS providers (Google Standard, Polly Standard) to the estimator. - Current experiment: The newsletter pipeline is running autonomously through
newsletter_orchestrator.pyand the subscriber system is synced via Resend and Cloudflare KV. We are iterating on the daily send cadence and approval flow. - What's next: Continue tuning the autonomous newsletter orchestrator, expand the hardware comparison pages, and prepare the next deploy around the TTS cost estimator and refreshed GPU rig guidance.
Sources: Reuters AI News, TechCrunch, CoinShares Q1 2026 Bitcoin Mining Report, Hashrate Index, Robinhood Newsroom, SEC public statements, Kunal Ganglani local LLM hardware guide, PromptQuorum 2026 hardware requirements.
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: August 01, 2026 | Written by dare404 from Boise, ID