Pondero AI logo

Pondero AI

Archives
Log in
Subscribe
August 14, 2026

Pondero Brief: Anthropic buys 20 years of compute from a bitcoin miner

Pondero Brief - AUGUST 14TH, 2026

Plus GPT-5.6 Sol hits 750 tokens/sec and DeepSeek V4-Pro edges past Opus.
pondero. BRIEF · AUG 14

Anthropic locks in $9.1 billion, 20-year compute deal with Riot Platforms

Riot holds the grid permits that take years to build, so Anthropic skips the construction queue.

Riot Platforms disclosed a $9.1 billion, 20-year data-center lease alongside Q2 earnings, with Bloomberg naming Anthropic as the customer. The deal covers 191 megawatts at Riot's Rockdale, Texas campus, with 96 MW online by December 2027 and the rest by June 2028, worth up to $16.1 billion with extensions.

Also in today's brief

  • GPT-5.6 Sol previewed at 750 tokens/sec on Cerebras silicon
  • DeepSeek V4-Pro ships GA and open-sources agent harness
  • ChatGPT now serves ads on 1 in 4 commercial queries
  • Copilot Agent Store gated behind IT admin approval
  • Cursor's recipe for reliable agent environments
  • Four enterprise agent guides now live on Pondero
 
Models & Releases

OpenAI previews GPT-5.6 Sol Ultrafast at 750 tokens per second

Running on Cerebras wafer-scale chips that keep the full model weights in 44 GB of on-chip SRAM, the preview clears the memory-bandwidth wall to hit 14x standard Sol speed with the same intelligence, per OpenAI's announcement. It is OpenAI's first production-preview run on non-Nvidia silicon, aimed at real-time voice and incident response. Limited API access with a waitlist.

 

DeepSeek V4-Pro ships GA with open-source agent harness and 1M-token context

TerminalBench 2.1 climbed from 72.1 to 87.9 and V4-Pro placed above Claude Opus 4.8 on several agent benchmarks in DeepSeek's own comparison table, with a 1M-token context. The MIT-licensed DeepSeek Harness ships alongside as a provider-agnostic answer to framework lock-in. Prices rise August 16 to $0.435 per million input tokens, still far under Opus.

 
Tools & How-To

ChatGPT now shows ads on one in four commercial queries

An SE Ranking study of 50,006 prompts found paid placements on 25.94 percent of commercial queries, rising to 28.69 percent in healthcare, versus 2.64 percent for Google's AI Mode in the same category. OpenAI just added shopping carousels and conversion-optimized CPC. If you run GEO for health or finance, audit your organic citation rate this week, because paid slots now compete on the same surface.

 

Microsoft 365 Copilot puts its Agent Store behind IT admin approval

Custom agents no longer reach end users until an admin approves them, closing the governance gap that stalled enterprise extensibility rollouts. The same wave added Anthropic model support inside Word for orgs running Claude. Platform teams managing Copilot at 1,000-plus seats finally control which agents ship.

 

Cursor spells out the recipe for reliable agent runs

The August 13 update confirms agents work best when the repo is pre-cloned, dependencies are installed, and setup scripts run before the session starts. Cursor also launched an India Start plan at 649 rupees (about $7.75) per month with always-on cloud agents. Try Cursor →

 
Quick Hits
• Make. Visual automation with 2,000-plus integrations for wiring AI triggers into the tools you already run. Try Make →
• ElevenLabs. Text to lifelike speech with an API clean enough to drop into a content or voice-agent pipeline in an afternoon. Try ElevenLabs →
• Firecrawl. Turns any URL into LLM-ready markdown, so your RAG pipeline or agent gets clean input instead of raw HTML. Try Firecrawl →
From the Pondero Stack
Illustration for the 5-layer enterprise agent reference architecture guide

The 5-layer enterprise agent reference architecture

The architecture your platform team will actually own in production, layer by layer. Start here if you are past the pilot.

 
Illustration for the enterprise agent platform RFP scorecard guide

The agent-platform RFP scorecard

Score vendors on harness architecture, eval maturity, and governance before the POC ends badly.

 
Illustration for the six enterprise agent deployment patterns guide

Six enterprise deployment patterns from the public record

The patterns every named 2026 enterprise deployment shares, pulled from what companies have said on the record.

 
Illustration for the data boundaries RAG fine-tuning DLP guide

Data boundaries: RAG vs fine-tuning vs context windows

Which control goes where, mapped through a DLP lens so sensitive data stays inside the line.

 

How was today's brief?

★★★★★ Nailed it  |  ★★★ Solid  |  ★ Missed

Jonathan Hildebrandt Jonathan Hildebrandt
Co-founder and primary operator of Pondero. Writes the Pondero Brief.

Affiliate disclosure  ·  Unsubscribe  ·  Manage preferences

Pondero earns commissions on some links. This does not affect our editorial picks.

Don't miss what's next. Subscribe to Pondero AI:
Older → Pondero Brief: Meta just put a frontier agentic model on a gaming GPU
pondero.ai
Bluesky
LinkedIn
Twitter
LinkedIn
Powered by Buttondown, the easiest way to start and grow your newsletter.