Our AI subsidy is $794 Billion in 2026 alone, that's why Harvey & Legora aren't bankrupt

Our AI subsidy is $794 Billion in 2026 alone, that's why Harvey & Legora aren't bankrupt

Everyone gets cheap AI because AI true costs are carried by capex spending

Let’s genuinely say thank you to hyperscalers 🙏🏽, the Chinese government 🇨🇳, pension funds via VCs, plus debt hidden by Banks 🏦

Other criticism

I did not account for caching so this is all BS! 👇🏽 I did by blending plus caching isn't free. GPT-5.6 bills cache writes at 1.25× uncached input, and Anthropic's is 1.25× (5-min) or 2× (1-hour). A low reuse rate makes caching cost more, not less.

Long-horizon legal agents makes it worse, why? Agentic legal work is output-heavy so the cacheable share shrinks with long-horizon agents 👈🏽

You can route and use cheaper models? 👇🏽 A weaker model reaches the same answer by thinking longer, calling more tools, and retrying more often so erase the entire saving. Also see China 🇨🇳 comment above on strong open models.

Finally, where did you get the data?👇🏽

Harvey $300M ARR, 13 trillion tokens, 2,000 customers 👉🏽 https://lnkd.in/eWs8Zgc3

Legora $100M ARR 👉🏽 https://lnkd.in/eKQjqyet

Legora 1,000+ customers 👉🏽 https://lnkd.in/eU5tvrui

OpenAI GPT-5.6 Sol $5 in / $30 out 👉🏽 https://lnkd.in/eJXnEj3C

Anthropic Fable 5 $10 in / $50 out 👉🏽 https://lnkd.in/euTSg7_U

Legora moves to usage-based pricing 👉🏽 https://lnkd.in/eF-5FBMr

Harvey vs Legora pricing divergence 👉🏽 https://lnkd.in/edxzGMmg

My own 5B-token binge 👉🏽 https://lnkd.in/eNCJjWyq

Bonus critique: inference margins are healthy 👉🏽 https://lnkd.in/ehJGcpBa Well, why go through pain and humiliation of constantly needing to raise debt?

The Weeknd 🎧, how was yours? 👉🏽 https://lnkd.in/evP8igyG