Episode Summary
Executive Summary: The episode centers on two major AI shifts: on-prem AI for regulated industries and the broader reorganization of companies around AI-driven productivity. It also covers proof-of-human/identity infrastructure, a wave of layoffs and restructuring, startup formation trends, and the rise of consumer health tools. Jason argues AI should be decentralized, privacy-preserving, and state-regulated rather than federally controlled.
Main Topics: On-prem AI for regulated industries (Priority: 5/5): GoAbacus pitches hardware-plus-software AI infrastructure for banks, hospitals, credit unions, and insurers that want AI without sending data to public cloud providers. Proof of human and agent authorization (Priority: 5/5): Yanez discusses biometric verification and uniqueness systems to verify humans, prevent bots, and authorize AI agents to act on behalf of real users. AI-driven layoffs and company restructuring (Priority: 5/5): The hosts frame recent layoffs at major tech and fintech companies as part cost-cutting, part AI transformation, with AI increasing output and reducing headcount needs. Startup formation and the new labor model (Priority: 4/5): Jason argues AI is enabling a wave of lean startups, independent contractors, and small teams that can outperform larger legacy organizations. Consumer-led healthcare and wearable data (Priority: 3/5): The discussion highlights Whoop, Function, Superpower, and similar products as building a new healthcare stack based on personal data and AI-assisted care. AI regulation and federalism (Priority: 4/5): Jason strongly opposes federal executive control over AI models and argues for state-level experimentation and regulatory diversity.
Key Arguments: Regulated industries need AI that stays inside their infrastructure; privacy and compliance make public cloud AI unacceptable for many buyers. Fixed-cost on-prem AI is more attractive than usage-based pricing for enterprises that need predictable spend. Local models and smaller specialized models can be sufficient for deterministic enterprise tasks. Training on customer data through a shared network can lower costs if done carefully and with customer consent around weights, not raw data. Proof-of-human systems will become essential as bots and AI-generated slop proliferate across social, commerce, and identity workflows. Recent layoffs are not only about bloated staffing; they also reflect AI allowing fewer people to produce more output. AI-first companies should require every employee to use AI for repetitive tasks and workflow design. The future of work is smaller teams, more independent contractors, and startups formed by laid-off employees. Consumer healthcare will increasingly start with wearable, lab, and app data instead of traditional doctor-first workflows. AI and other contested technologies should be governed more locally; Jason argues federal executive power is too centralized and prone to abuse.
Data Points: Go One orders: Over 1,600 orders - David Moscatelli says the on-prem AI device received this many orders about 25-30 days after launch. Concurrent users per Go One: Up to 2,000 users - The Go One appliance is said to support 2,000 concurrent users, with chaining for more. Go One Max capacity: Up to 8,000 concurrent users - A larger version of the hardware is mentioned as coming soon. Go One starting price: $250,000 - Starting capex for the standard enterprise appliance. Go One Max starting price: $350,000 - Starting capex for the higher-capacity version. Model refresh cadence: Every 6 months - GoAbacus says its core model ships on a semiannual cycle. Hardware refresh cadence: Every year - The appliance is replaced annually as part of the service. Client training discount: 20% off monthly fee or 50% off hardware price - Customers who share model weights receive incentives. GoAbacus data set: 30 million queries - Legacy thumbs-up/thumbs-down question-answer data used for training financial services models. Yanez alpha projects: About 5 projects - Jose Caldera says the product is in alpha with five projects coming live soon. BitTensor subnet: Subnet 54 - Yanez says it operates as subnet 54 in the BitTensor ecosystem. BitTensor subnet count: 128 subnets - Jason references the network as having 128 competing subnets. Yanez market cap: About $9-10 million - Jason cites the subnet token/project valuation as very early-stage. Cloudflare layoffs: 20% of staff - Jason corrects the earlier estimate and says Cloudflare is cutting about 20%. Coinbase layoffs: 14% of staff / about 700 people - Cited as part of the broader AI-era restructuring trend. Block layoffs: 4,000 people in February - Used as a prior example of aggressive restructuring. Crypto.com layoffs: 12% of staff - Another example of tech/crypto workforce reductions. Block AI adoption: 100% of employees using AI tools - Block says all employees are using AI in their work as of early April. Productivity lift at Block: Perks and code changes per engineer up over 2.5x - Used to support the claim that AI is already improving output. Block EPS beat: 85 cents vs. 68 cents expected - Jason and Alex cite earnings that beat expectations. Block EPS year-over-year: 56 cents a year ago - Shows earnings improvement over the prior year. Anthropic round: $50B at $900B pre-money valuation - Discussion of a potential huge private financing round. Anthropic revenue run rate: $45B annualized - Reported by the Financial Times in the transcript. Founder University Japan cohort 1: 60 companies - Jason describes the first Japan cohort as having 60 participating companies. Talent policy at one company: Top one-third classified as AI/ultra-high performers - Jason explains his return-to-office exceptions for top performers. Whoop/healthcare cost examples: $20K-$50K/year concierge care; some up to $100K/year - Jason contrasts traditional concierge medicine with emerging consumer-led alternatives.
Pivotal Quotes: "If there's only one way to really know if your information is being shared or not, which is to not share it." — Jason Calacanis: Explaining why regulated industries prefer on-prem AI rather than public cloud AI. "If you don't have the time to get to it, you don't have a job." — Jason Calacanis: Describing an AI-first workplace where employees must use AI for repetitive work and productivity gains. "I wouldn't trust Biden or Kamala or Trump or J.D. Vance with AI regulation for everything." — Jason Calacanis: Arguing against centralized federal control over AI and for state-level experimentation.
Implications: The episode suggests AI adoption is moving from experimentation to infrastructure, with privacy, cost control, and trust driving on-prem deployments. It also signals a labor shift toward smaller teams, more automation, and decentralized regulation, while identity and proof-of-human tools become increasingly important.
About This Week in Startups
Jason Calacanis covers startups, tech, markets, media, and all the hottest topics in business and technology. He also interviews the world’s greatest founders, operators, investors, and innovators.