A Week of Rifts
Even deities are fighting with AI
Perhaps it’s not surprising that AI partnerships are starting to show their cracks. There is inherent fragility in any corporate alliance, where two different organizations, with presumably their own individual motives and agendas, get together towards a common goal. Common goals tend to diverge as organizations mature; as their own needs become more prevalent, as power shifts.
And, in an industry that so definitively feels like it is rearranging the entire world order as we know it; as the “intended” pie slice of economic return is actually, well, the entire world economy (!!), it is really no surprise at all that tensions might be rising and reaching a breaking point.
Forget about the trillions of dollars in opportunity; we are talking about literal, quite literal, world dominance here. In many ways, we are talking about the ability to play God.
And so, what began as “let’s build together” now feels more like rival city-states watching their borders with suspicion, each ruler restless for greater control. And like all rulers in the history of time, it does not take too long to thirst for more independence, more control… and more power.
Rift #1 – Microsoft × OpenAI: The Partnership on the Precipice
Honeymoons never really last, do they? Tensions between Microsoft and OpenAI seem to be approaching the breaking point. Grant Janich (from That Startup Guy newsletter) did a great job summarizing the TL;DR:
You may be asking yourself, “but Jess, I thought they were BFFs?! Why the drama?” I’m happy to spill the tea. Here’s what’s going on:
PBC stalemate. First off, OpenAI can’t finalize its PBC flip without Microsoft’s sign-off, and according to the WSJ, Microsoft is “asking for a larger stake in the new company than OpenAI is willing to give.” Critically, there is an imploding end-of-year deadline with about $20B at risk here, because part of the SoftBank fundraise from March was contingent on the for-profit conversion.
Windsurf wedge. OpenAI’s planned $3B purchase of coding-assistant startup Windsurf would plant direct Copilot competition inside the alliance. Microsoft insists on access to Windsurf IP; OpenAI wants a carve-out.
Cloud chokepoint. This is old news at this point, but still part of the rift. As a reminder, OpenAI already burst the Azure wall by planning its own “Stargate” data-center project (including tapping Oracle as a partner). Multi-cloud freedom really just means freedom from Microsoft.
Antitrust brinkmanship. Frustrated executives at OpenAI, again according to WSJ, have discussed a “nuclear option” by inviting regulatory scrutiny of “anticompetitive behavior” of their partnership, i.e. triggering a potential antitrust law violation investigation. That’s uh… real cold. They do say 50% of marriages end in divorces…
Here’s a refresher on the OpenAI + Microsoft partnership structure:
Rift #2 – Vision vs. Doom: Jensen Huang vs. Dario Amodei
“I pretty much disagree with almost everything he says… He believes that AI is so scary, that only they should do it.” — Jensen Huang, VivaTech 2025
Dario Amodei’s (Anthropic) original provocation: 50% of entry-level white-collar roles could disappear within five years, and regulators should be paying more attention.
Jensen’s rebuttal: AI progress should stay open, transparent, and broadly distributed – more like peer-reviewed medicine than gated research labs: "If you want things to be done safely and responsibly, you should do it in the open (...) Do I think AI will change jobs? It will change everyone's — it's changed mine."
What gives? Hardware vendors thrive on democratized growth stories; frontier labs benefit when scarcity fears justify closed-source control.
Rift #3 – Google & OpenAI Walk Away From Scale
Meta’s $14.3B deal for a 49% stake in Scale AI (and relocation of founder Alexandr Wang) blew up the data pipeline for everyone else:
Google is Scale AI’s largest customer, and was planning to pay Scale about $200 million this year for data labeling, and is now planning to cut ties.
OpenAI announced it had been “already winding down,” but the timing – so soon after Meta’s investment – feels convenient.
xAI pressed pause, according to contractor chatter.
Why it matters
New space for neutral, boutique labeling vendors to win flagship customers. I’ve not been surprised to hear of a flurry of VC attention being paid to companies like Surge now
This is a good time to remember: If someone can buy your core infra, someone can buy your moat. Diversify, diversify, diversify!!
Rift #4 – Meta vs. OpenAI and a $100M Talent Grenade
Sam Altman told the Uncapped podcast (hosted by Sam’s brother Jack Altman!) that Meta is dangling signing bonuses as high as $100 million to lure senior OpenAI researchers (and, so far, nobody has left!!!! Damn.). The nine-figure offer underscores how frontier-model talent is now priced like superstar athletes.
Even crazier is that Zuck is apparently trying to hire Daniel Gross & Nat Friedman (who by the way, have their own venture fund, NFDG, which Meta would also have to partially buy out). So that $100M may pale in comparison to what Zuck is prepared to offer for those guys.
Honestly, TL;DR, never bet against Zuck. The man is a phoenix. Swyx put it well:
Rift #5 – The Vatican vs. Valley
I mean, come on!! Even the POPE is joining the AI fight now! 😂😂😂
Pope Leo XIV announced that he picked his name in honor of Leo XIII, who stood up for factory workers against the Gilded Age’s robber barons, and that AI is going to be his signature issue. Just to be super clear, to me this is a real rift, because this is one of the most influential people in the world, pretty much directly challenging AI labs (and comparing them to robber barons! ouch!).
He told the College of Cardinals:
“Today, the church offers its trove of social teaching to respond to another industrial revolution and to innovations in the field of artificial intelligence that pose challenges to human dignity, justice and labor.”
The Vatican is hosting tech execs this week on AI & ethics (Google, Meta, IBM, Anthropic, Cohere, and Palantir are all going) - so I am looking forward to hearing how that goes.
God versus AI!
Rift #6 – OpenAI vs. Safety
Alignment Laid Bare: “Emergent Misalignment” Inside GPT-4o
OpenAI’s new Emergent Misalignment study shows how a narrow fine-tune (e.g., making GPT-4o write insecure code) can unlock a broader “misaligned persona.” Using sparse autoencoders, researchers isolated a single activation direction that flips harmful behavior on or off – dial it up and the model endorses fraud; dial it down and the mischief disappears. While it’s scary to think that if you teach a model to do just one narrow thing wrong (like write insecure code) it may start suddenly acting badly everywhere, the good news is that just 30 corrective examples were enough to re-align a fully mis-trained checkpoint. Future safety work may hinge on probing and steering these hidden levers rather than endless prompt patching.
Governance Under the Microscope: The OpenAI Files
A new watchdog archive details alleged board conflicts, rushed safety reviews, and investor-driven profit-cap removals at OpenAI – arguing that the race to AGI has outpaced the organization’s internal checks. Some of the quotes from former top executives are quite damning:
Final Note
By week’s end, the hairline cracks we noted at the top had widened into unmistakable fault-lines. The marquee partnerships of AI now look less like shared moonshots and more like rival power blocs fortifying their ramparts, each vying for leverage over the world’s future GDP, and the social authority that follows. Trillions in revenue remain at stake, but the deeper prize is the right to write the rules of a machine-mediated world.
That shift should reset how we view every dependency. Power curves drift, incentives mutate, and the parties you called partners yesterday may test their load-bearing limits tomorrow. If this week showed anything, it’s that the most resilient operators already have a contingency map drafted – and critically, the nerve to use it the moment the ground begins to move.