[00:00] Nina Park: Welcome to Model Behavior. This program examines how AI systems are built, deployed, and operated
[00:07] Nina Park: in real professional environments. Joining us today is a guest who brings a systems-level
[00:12] Nina Park: perspective on AI and automation, blending technical depth with engineering insight.
[00:18] Nina Park: It is great to have you here.
[00:20] Thatcher Collins: We start today with a significant shift in the financial landscape surrounding artificial
[00:25] Thatcher Collins: intelligence.
[00:26] Thatcher Collins: A new analysis from Axios shows that the eight largest tech companies with AI ambitions now represent nearly half of the S&P 500.
[00:35] Thatcher Collins: To fuel this growth, tech firms are projected to issue over $1 trillion in debt this year alone, with hyperscalers spending up to $700 billion from their balance sheets.
[00:48] Chad Thompson: I mean, it is a massive concentration of risk, Thatcher.
[00:52] Chad Thompson: The market has moved past questioning if AI works to questioning if the monetization can justify these capital inflows.
[00:59] Chad Thompson: When half of all venture capital dollars are flowing into a single sector, the entire economy is essentially betting on the successful deployment of these systems.
[01:10] Nina Park: Speaking of deployment, Alibaba recently released QN 3.5, a model that signals a new phase in open weight efficiency.
[01:20] Nina Park: That's true. The parameter count here is massive, but the active weights tell a different story.
[01:26] Thatcher Collins: Mm-hmm. Nina, QN 3.5 has 397 billion total parameters, but only 17 billion are active during any forward pass.
[01:37] Thatcher Collins: This allows it to compete with frontier-closed models like GPT-5.2 and Claude 4.5 while remaining significantly more efficient to run.
[01:46] Thatcher Collins: It's natively multimodal and specifically designed for egentic tasks, which aligns with the broader industry move toward autonomous workflows.
[01:54] Chad Thompson: That trend is being reinforced elsewhere too.
[01:58] Chad Thompson: OpenAI just hired the creator of the OpenClaw framework.
[02:02] Chad Thompson: This suggests OpenAI is pivoting toward a multi-agent future where specialized tools interact, rather than just relying on a single chat interface.
[02:11] Chad Thompson: We are also seeing Meta Integrate AI into its Ads Manager to automate data analysis, showing that agents are no longer theoretical.
[02:21] Chad Thompson: They are becoming standard features in business software.
[02:25] Nina Park: The scale of this adoption is particularly evident in India right now.
[02:30] Nina Park: At the India AI Impact Summit in New Delhi, OpenAI announced that India has over 100 million weekly active chat GPT users, making it their second largest market.
[02:42] Nina Park: Anthropic also opened a new office in Bengaluru this week, reporting that the revenue run rate in India has doubled in just four months.
[02:51] Thatcher Collins: The enterprise numbers are even more striking, Nina.
[02:55] Thatcher Collins: Cognizant is deploying Gemini and Claude to its 350,000 global employees to modernized legacy systems.
[03:04] Thatcher Collins: they are effectively becoming a massive testing ground for agentic reliability
[03:09] Thatcher Collins: before they package these same services for their own corporate clients.
[03:13] Chad Thompson: However, that rapid expansion is hitting friction at the government level.
[03:18] Chad Thompson: Reports indicate the defense secretary is considering designating anthropic as a supply chain risk.
[03:25] Chad Thompson: This stems from their refusal to remove safeguards for military use.
[03:31] Chad Thompson: While competitors like OpenAI and XAI have reportedly agreed to allow use for all lawful purposes,
[03:40] Chad Thompson: Anthropic is maintaining restrictions, which could lock them out of Pentagon contracts.
[03:45] Nina Park: That tension between safety and utility is also appearing in consumer hardware.
[03:51] Nina Park: Meta is planning to add facial recognition to its smart glasses by the end of the year
[03:56] Nina Park: under a feature called Name Tag.
[03:59] Nina Park: While Meta says it won't be a universal lookup tool, using social media profile data for real-time identification is already raising significant privacy concerns.
[04:11] Nina Park: Absolutely. It highlights the recurring theme of 2026. The technology is scaling faster than the governance frameworks can adapt.
[04:21] Nina Park: Whether it is $1 quadrillion in debt or 100 million users in a single country,
[04:28] Nina Park: the sheer volume of these deployments is forcing a high-stakes reckoning for the industry.
[04:34] Nina Park: Thank you for listening to Model Behavior.
[04:37] Thatcher Collins: Thank you for listening to Model Behavior, a Neural Newscast editorial segment.
[04:43] Thatcher Collins: For more technical analysis, visit mb.neuralnewscast.com.
[04:48] Thatcher Collins: Neural Newscast is AI-assisted, human-reviewed.
[04:53] Thatcher Collins: View our AI transparency policy at neuralnewscast.com.