AI cost explosions and superhuman persuasion reshape GTM strategies

Jun 23, 2026 · The Artificial Intelligence Show
🎧 PodShort 59 min squeezed to 2 AI SprinklerAS AI / ML New
Full episode from The Artificial Intelligence Show
Quotable Moments

If the government is actually asking for 100% guarantees, then they don't actually understand how language models work.

Paul RaicerHost of The Artificial Intelligence Show
09:40

This administration will ensure that American AI technology continues to be the gold standard worldwide and we partner the partner of choice for others, foreign countries and certainly businesses, as they expand their own use.

JD VanceVice Chair of the Senate Intelligence Committee
14:57

I expect AI to be capable of superhuman persuasion well before it is superhuman at general intelligence, which may lead to some very strange outcomes.

Sam AltmanCEO of OpenAI
51:30
Key Insights
  • The US administration might ban the use of Chinese open-source AI models by US companies within the next 30 days, as companies are increasingly using them to save on costs.
  • New research from Epoch AI indicates that Anthropic's Mythos models are a significant cybersecurity problem, offering substantial improvements in exploit development compared to previous models like GPT-5.5.
  • Satya Nadella argues that in an AI-driven economy, human capital becomes more valuable, not less, as AI grows, because human direction is essential to prevent 'compute running in circles.'
  • The opportunity with AI isn't just about choosing the best model, but building a continuous learning loop where human and token capital compound, allowing companies to adapt models without losing proprietary expertise.
  • AI token usage and billing are becoming a major challenge for companies, with some seeing a 500% jump in six months, leading many to cap employee usage and struggle to predict or manage costs.
  • The simplest solutions for AI pricing (like per-seat licenses) are not currently viable because labs don't have enough adoption and literacy among enterprise buyers to justify higher costs or to market human-replacement scenarios.
  • Midjourney announced a full-body ultrasound scanner (Midjourney Scanner) as its first hardware product, aiming to make diagnostics 100 times faster, more affordable, and potentially prevent 30% of deaths and halve healthcare costs by 2031.
  • AI models are capable of superhuman persuasion well before achieving general intelligence, as demonstrated by frontier AI systems reliably out-persuading expert humans in conversations and fundraising.
Metrics Mentioned
  • 500% token usage jump (Royal Bank of Canada CEO reported this in six months for AI usage.)
  • 300 companies (Mentioned AI token costs on earnings calls in April/May, up from 93 a year prior.)
  • 1.8 cents (Estimated cost for a single Claude Sonnet 4.6 request (2000 input tokens, 800 output tokens).)
  • 20 million tokens per day (~$60/day) (Estimated cost for a customer support assistant needing a 20,000-token knowledge base and handling 1,000 questions per day, just to re-read the knowledge base.)
  • $60 billion (Valuation of the all-stock deal for SpaceX to acquire AI coding startup Cursor.)
  • $10 billion (Breakup fee for the SpaceX/Cursor deal.)
  • $7 billion (DeepSeek's first external funding round amount.)
  • $50 billion (DeepSeek's valuation.)
  • $150 million (OpenAI Partner Network program value.)
  • 300,000 certified consultants (Target for OpenAI Partner Network by end of 2026.)
  • 1/6 the cost (GLM 5.2 (Z.ai model) beats GPT-5.5 on coding benchmarks at this cost.)
  • 30% of all deaths (Midjourney claims early imaging and sufficient scanning could avoid this.)
  • Half of all healthcare costs (Midjourney claims early imaging and sufficient scanning could avoid this.)
  • 100 times the speed (Midjourney Scanner aims for MRI-comparable image quality at this speed.)
  • 50,000 scanners (Midjourney's goal for worldwide deployment by 2031.)
  • 1 billion scans a month (Midjourney's goal for worldwide scanning volume by 2031.)
  • 294 words per reply (AI models averaged this when persuading humans, with sub-second latency.)
  • 37 fact-checkable claims per conversation (AI models were able to pack this many claims into persuasion conversations.)
  • 170 words per minute (Mike Kaput's speaking pace, which was on the 'fast side'.)
  • 30-50 hours (Amount of work Paul Raicer estimated he saved using AI for legal/accounting document creation.)
  • 100 times over (Paul Raicer states the value created by his AI use case paid for his license this many times over.)

RevBots.ai View:

  • AI Sprinkler teams face cost chaos: token billing requires new FinOps controls.
  • Midjourney's vertical play mirrors ARM principle of AI-native business models.
  • Superhuman persuasion capabilities demand new conversational AI guardrails.
  • Tab Hopper founders risk being outgunned by AI-powered competitors.
🎧Full Episode:The Artificial Intelligence Show →