Latest / Neural Newscast / Microsoft Models In-House while OpenAI Files for IPO [Model Behavior]
Transcript
- [00:00] Announcer: From Neural Newscast, this is Model Behavior, [00:02] Announcer: AI-focused news and analysis on the models shaping our world. [00:12] Announcer: Welcome to Model Behavior. This show examines how artificial intelligence systems are built, [00:17] Announcer: deployed, and operated in real professional environments. Today is June 9th, [00:23] Announcer: 2026. We are currently tracking a structural shift in the industry [00:27] Nina Park: as Microsoft asserts its independence from its primary partner and the two biggest names [00:33] Nina Park: in frontier AI prepare for public markets. Joining me to discuss these developments is Thatcher. [00:39] Thatcher Collins: Nina, the headlines from Microsoft's Build conference on June 2 are still reverberating across [00:44] Thatcher Collins: the sector. Microsoft announced seven new AI models built entirely [00:50] Thatcher Collins: in-house by their superintelligence team. This represents a significant pivot away [00:54] Thatcher Collins: from the previous arrangement where Microsoft's branding was effectively a wrapper [00:58] Thatcher Collins: for OpenAI's GPT models. They have invested roughly $13 billion [01:04] Thatcher Collins: in OpenAI over the years. But this move suggests they want to own the core capability directly. [01:08] Nina Park: That focus on ownership is a major strategic hedge. Microsoft released [01:13] Nina Park: a comprehensive 109-page technical report detailing their progress. The transparency in that report [01:20] Nina Park: has been praised by independent researchers, however, which is rare in an era where these documents [01:25] Nina Park: are often marketing materials. However, Microsoft is specifically comparing its results [01:30] Nina Park: to older versions [01:31] Nina Park: of Cloud. which complicates their narrative of being toe-to-toe with the current state-of-the-art. [01:36] Thatcher Collins: That is an important distinction, Nina. While the transparency is welcome, [01:40] Thatcher Collins: independent verification of these performance claims is still in the early stages. [01:45] Thatcher Collins: The real story is the from scratch nature of the training. Microsoft is working to remove [01:51] Thatcher Collins: the strategic exposure of relying on a partner whose interests might not always align with [01:55] Thatcher Collins: their own This shift towards self-reliance is happening just [01:58] Thatcher Collins: as their partners are looking toward the entry into public markets. [02:01] Nina Park: Exactly. Yesterday, June 8th, OpenAI submitted its draft S1 to the SEC for a confidential IPO. [02:09] Nina Park: This follows Anthropic's own filing back on June 1st citing Reuters reports Anthropic is targeting [02:15] Nina Park: a valuation of approximately $965 billion which briefly overtook OpenAI on the secondary markets. [02:22] Nina Park: OpenAI is targeting a debut as early as September 2026, [02:26] Nina Park: with a valuation range between $730 billion and $852 billion. [02:31] Thatcher Collins: It is fascinating to see that valuation gap, Nina. Anthropic is moving from a position [02:37] Thatcher Collins: of relative strength, telling investors it expects to report its first profitable quarter here in [02:42] Thatcher Collins: the second quarter of 2026. OpenAI, despite raising $122 billion this past March, [02:49] Thatcher Collins: has not reached that profitability milestone. When these S1 filings become public, [02:55] Thatcher Collins: the market will finally see the actual revenue trajectories behind these trillion-dollar ambitions. [03:01] Nina Park: The public markets are a different beast than the private venture world. [03:05] Nina Park: While they prepare for listings, these companies are also issuing warnings about the technology. [03:10] Nina Park: Anthropic recently called for a brake pedal on AI development, citing risks [03:15] Nina Park: of recursive self improvement. Reuters reported that as of May 2026 over 80% of the code merged into [03:23] Nina Park: Anthropic's own codebase was authored by Claude. [03:27] Nina Park: That is no longer a theoretical milestone; it is a shift in how these models [03:31] Thatcher Collins: are built." "I find that statistic particularly jarring, Nina. If AI is building the next generation [03:37] Thatcher Collins: of AI, the acceleration curve changes. However, there is a friction point here. [03:43] Thatcher Collins: Is this genuine safety concern or regulatory capture? When a near trillion dollar [03:49] Thatcher Collins: company calls for a pause, smaller competitors often hear an attempt to freeze the race now [03:54] Thatcher Collins: that the leaders are ahead. Anthropic and OpenAI are asking for trust as they become [03:58] Thatcher Collins: central authorities in AI governance, while their valuations depend on continued scaling. [04:04] Nina Park: That tension will be a primary focus for public market investors. For the users, [04:10] Nina Park: the models keep arriving. We have seen a rapid evolution through the Claude family, [04:15] Nina Park: moving from the 3.5 breakthrough last year to the current flagship, OPUS 4.8, [04:20] Nina Park: which was released in May. These models are increasingly focused on agentic workflows [04:25] Nina Park: and real-world reliability. That is what enterprise customers [04:28] Nina Park: are actually buying. They want governable intelligence they can defend to their boards. [04:34] Thatcher Collins: Right, reliability is the product now. Whether it's Microsoft building [04:39] Thatcher Collins: its own superintelligence stack or Anthropic scaling the Cloud4 family, [04:44] Thatcher Collins: the industry is shifting from asking can it do this to how do we govern it as it does this? [04:49] Thatcher Collins: The coming IPOs in the second half of 2026 [04:52] Thatcher Collins: will be the ultimate stress test for whether the public markets believe in that governance story. [04:57] SPEAKER_03: Thank you for listening to Model Behavior, [04:59] Nina Park: a Neural Newscast editorial segment. For more on these technical reports and IPO filings, [05:06] Nina Park: visit mbn.neuralnewscast.com. Neural Newscast is AI-assisted, [05:12] Nina Park: human-reviewed. View our AI transparency policy at neuralnewscast.com. [05:19] Announcer: This has been Model Behavior on Neural Newscast. Examining the systems behind the story.