Start Here: A Map of Capital & Compute

Capital & Compute covers one question from many angles: what does artificial intelligence actually cost, and who pays it. That means token prices, coding-agent bills, benchmark credibility, memory and GPU markets, data-center capital, and the economics underneath all of it. There are 194 published analyses and 33 free tools.

This page is the map. If you would rather just browse, the full archive is ordered newest first.

Three ways in

You have a decision to make. Go straight to a tool. The calculators and trackers below answer specific questions with numbers rather than prose.
You want to understand a subject. Pick one of the six topics and read its starting analysis, then follow the links from there.
You are a machine. The manifest at /llms.txt describes the site and every dataset. Structured data ships as JSON at the endpoints listed at the foot of this page.

The six topics

Every post belongs to one of these, and each topic hub carries its own full archive.

AI costs (135 posts). Token prices are the sticker. Context windows, retries, cache misses, subscription floors and the work an agent abandons halfway decide the bill. A model that looks half the price per million tokens routinely costs more per finished task, and the gap only appears once the same job is priced end to end across providers. Start with Why Is My AI API Bill So High? 8 Causes and Fixes.
Coding agents (83 posts). The model matters. The harness around it usually matters more: how an agent reads a repository, what it retries, how much context it wastes, and where the loop stops. Two teams running the same model see cost and success rates diverge by multiples once the scaffolding and workflow differ. Start with The 2026 AI Coding Agent Landscape: Leaders, Costs, Harness.
Models & benchmarks (90 posts). Leaderboards compress messy reality into a single number, then labs quote whichever number flatters the release. Suites saturate, contamination inflates scores, and the same model under different scaffolding can move ten points. What matters is which benchmarks still separate frontier models, and what each one refuses to measure. Start with Best Open-Weight AI Models in 2026.
Local AI & hardware (33 posts). Memory capacity sets what a machine can load, bandwidth sets how fast it answers, and utilization decides whether owning hardware beats renting it. With DRAM in a shortage cycle the arithmetic keeps moving: a build that paid for itself last quarter may not this one, and the break-even turns on how many hours the box actually runs. Start with How Much RAM to Run a Local LLM? 2026 Sizing Guide.
Compute infrastructure (25 posts). Behind every model sits a physical stack measured in gigawatts, substations, GPUs and construction schedules, and a financial one measured in debt, leases and depreciation assumptions. Capacity announced today energizes years out, so the distance between what is promised and what is running is where the interesting numbers live. Start with What Is an AI Data Center? Cost, Power, and Scale.
AI markets (51 posts). Adoption, productivity, publishing and national strategy show where AI creates value and where it only moves it from one balance sheet to another. Measured gains rarely match the claims, search referral traffic is repricing entire publishing models, and countries subsidize compute for reasons that are strategic before they are economic. Start with AI Search Broke the Web's Business Model. What Replaces It..

The tools

All free, no signup, no paywall. Calculators run in the browser; trackers are rebuilt from source-verified data rather than fetched live, so a number on the page is a number someone checked. See the tools hub for the full set.

Can I Run This LLM?. Pick your GPU, Mac, or RAM and see which local models you can run, how fast, and whether owning, renting a cloud GPU, or the API is cheaper.
Cost-Per-Task Calculator. Model what an AI coding agent really costs per task, not per month, and find the cheapest model for the work.
Cost Per Task Ranking. Every AI model ranked by what one real coding task costs to finish, computed from verified per-token prices, cheapest first.
Model Comparison. Put two or three models head to head on the same task and see which is genuinely cheaper to finish the work.
Subscription vs API. Should you pay for Claude Max, Cursor, or Codex, or use the API? Enter your usage and see the break-even.
Value Leaderboard. The top LLMs ranked two ways: by benchmark score and by value, the coding and intelligence points you buy per dollar.
Non-China AI Models. Frontier AI models ranked by value with the Chinese labs excluded, for teams under government, data-residency, or procurement rules that rule them out.
AI Benchmarks. A directory of 116 AI benchmarks across 12 categories: what each measures, who built it, the year, the current top score, and whether it still separates frontier models.
Benchmark Trust Ranking. Every major AI benchmark ranked by how much reason there is to distrust its headline number, scored on contamination, saturation, gameability and real-world gap.
Free AI Models. The AI models you can call at $0 on OpenRouter, Google AI Studio, Groq and more, filtered to the genuinely good ones by independent benchmark.
Model Release Tracker. The latest and upcoming AI models with release dates and per-token pricing, verified against primary sources.
Releases by Month. The month-by-month archive of every AI model release since June 2026, with the price each model launched at, the lab announcement behind every date, and a link to that month full analysis.
Inference Providers. Where to run a model: 35 inference providers across five categories, from frontier labs to open-weight hosts, speed specialists and GPU clouds, with representative pricing.
Coding Plan Pricing. Monthly subscription prices for every major AI coding tool, sorted cheapest first and dated to source.
AI Coding Agents. The 2026 landscape of coding harnesses: maker, interface, pricing model, and what each is best for.
Claude Skills & Agent Directory. A curated, source-linked directory of Claude Skills, community agent skills, MCP servers and prompt libraries, rated by what is actually worth installing.
Best Open Agent Harnesses. Pi, OpenCode, Aider, Goose, DeepSeek Harness, Hermes and more, compared by license, local-model support and what happens when context fills.
Memory Price Tracker. What a RAM kit costs right now per GB, a buy-or-wait verdict by buyer type, and the DRAM spot market against flat government indices.
Should You Buy RAM Now?. A live buy-or-wait check: price your exact kit or module against the tracked range, see whether the quote you were given is fair, and what would flip the answer.
DDR5 Price Tracker. DDR5 kit and per-GB prices by capacity, and why the 64GB tier is now better value per gigabyte than 16GB.
DDR4 Price Tracker. DDR4 costs about half as much per GB as DDR5 at retail while its chip spot price sets records. The cheap escape hatch, and why it is closing.
HBM Price Tracker. What high bandwidth memory actually costs per gigabit and per GB, and why HBM3E on contract undercuts a consumer DDR5 kit.
Server RAM Price Tracker. Live DDR5 ECC RDIMM street prices per GB and per module, the 3Q26 contract forecast, and the long-term agreements that shield large cloud buyers from it.
RAM Price Chart. Every published DRAM spot reading since November 2025 charted over time, with the move between each, the index level, and the unobserved winter that holds half the run.
GPU Price Tracker. What Nvidia and AMD cards cost against official MSRP, the 2026 kit-price hikes driving the gap, and a buy-or-wait verdict by situation.
Nvidia GPU Price Tracker. Every RTX 50 model against its official MSRP, from the $1,999 RTX 5090 to the $299 RTX 5060.
AMD GPU Price Tracker. AMD official MSRP increases across the RX 9000 lineup, versus street price by model.
Why VRAM Costs So Much. What a GDDR7 memory module costs, how much of a GPU price is memory, and the VRAM premium isolated in one card pair.
SSD Price Tracker. What a terabyte of flash costs by capacity, why the cheapest gigabyte sits in the middle of the range, and the one storage channel whose price has already turned.
Hosting Price Hike Tracker. Every announced VPS, cloud and dedicated server price increase of 2026, what channel each applies to, and the memory costs driving them.
PC Build Cost Calculator. What a gaming PC actually costs in 2026, combining live GPU and RAM tracker prices with a dated baseline for everything else.
AI Usage by Country. Where AI is actually used, by country: usage intensity, adoption rates and IMF readiness, each from a primary source and plotted against national income.
AI Pricing Keyword Trends. A keyword cloud of 45 AI pricing and model topics from real autocomplete suggestions, sized by frequency signal rather than ad-platform volume.

How to judge this site

Read the parts that let you check our work rather than take it on faith:

Editorial standards: how claims are sourced and fact-checked, and exactly how and where we use AI.
About: what this publication is, what it covers, and who writes it.
Disclaimer and disclosures: editorial content, not financial advice.
Advertise and get listed: paid listings and sponsorship, and why no ranking position is for sale.

Most of what we publish has a shelf life. Posts carry the date they were last updated, and the data pages carry their own refresh dates. If you find a figure that has moved, send a correction and it gets fixed and re-dated.

What the numbers here will and will not do

The trackers follow one rule that is worth knowing before you read them: prices from different markets are never averaged together. A memory chip has a spot price, a contract price, a retail kit price and a server street price, and those four can move in opposite directions in the same week. Blending them produces a single tidy figure that is wrong in every direction at once, so each channel is kept in its own column with its own source and its own date.

The same discipline applies to what is missing. Where a source has not published a value, the cell stays blank rather than being filled with an estimate, and where a figure rests on a single listing it is flagged as thin rather than presented as a market. Forecasts are labelled as forecasts and attributed to whoever made them. The cost of this is that the pages are less quotable than a single headline number; the benefit is that when a figure here disagrees with one you read elsewhere, you can see exactly which market each of you is describing.

Getting your product listed

The directories take submissions: AI tools, models, inference providers, benchmarks, and agent skills. Every number recorded has to be verifiable against a page the vendor publishes, and enhanced and featured listings are available as paid options. How to submit or correct an entry.

Machine-readable access

Crawlers and answer engines are welcome. Nothing here is gated, and the datasets are published as JSON so you can cite the data rather than scrape the page.

/llms.txt: site manifest, pillar descriptions, and a full post index.
/rss.xml: every new post.
/sitemap-index.xml: every URL.
/ai-benchmarks.json: dataset behind the matching page.
/ai-inference-providers.json: dataset behind the matching page.
/ai-model-leaderboard.json: dataset behind the matching page.
/ai-model-releases.json: dataset behind the matching page.
/claude-skills.json: dataset behind the matching page.
/free-ai-models.json: dataset behind the matching page.
/hosting-price-hikes.json: dataset behind the matching page.
/memory-prices.json: dataset behind the matching page.
/ssd-prices.json: dataset behind the matching page.

If you cite us, link the canonical URL (all of them use a trailing slash) and name the date you read it. Our numbers move.