June 25 2026 - AI Shakeups: OpenAI Chips, Talent Exodus & RobotsAI Shakeups: OpenAI Chips, Talent Exodus & Robots artwork

June 25 2026 - AI Shakeups: OpenAI Chips, Talent Exodus & RobotsAI Shakeups: OpenAI Chips, Talent Exodus & Robots

The AI Signal & The AI Noise

June 25, 2026

OpenAI and Broadcom reveal 'Jalapeño' custom LLM chip; top Google researchers defect to Anthropic; Gradium claims faster, lower-latency real-time translation with voice cloning; Agility Robotics files for a $2.5B SPAC to scale Digit. Big bets, big risks.
Speakers: Morgan
**SPEAKER_1** (0:00)
Welcome back to the show. It is Thursday, and we have some absolutely wild AI news today.

**Morgan** (0:07)
Hey, everyone. Yes, Thursday is always packed, but today we are looking at some massive hardware shifts and big talent moves.

**SPEAKER_1** (0:15)
Dude, seriously, we are talking custom silicon, robotics going public, translation wars, and researchers jumping ship. It is crazy.

**Morgan** (0:26)
It really is. The landscape is shifting so fast, it's hard to keep up. But we've picked the four most interesting stories for you today.

**SPEAKER_1** (0:35)
Oh yeah, we did. No boring press releases here, just the stuff that actually matters. So let's skip the fluff and get right into it.

**Morgan** (0:44)
Sounds good. Let's start with the biggest hardware story of the week. What is OpenAI up to now?

**SPEAKER_1** (0:51)
Okay, so OpenAI and Broadcom just unveiled Jalapeño. It is a custom chip built specifically for LLM inference, and it is set to run at scale by late 2026

**Morgan** (1:07)
Wait, Jalapeño? Like the pepper?
That is a spicy name for silicon. But seriously, why is OpenAI building their own chips? Aren't they Nvidia's biggest fan?

**SPEAKER_1** (1:20)
Well, yeah, but Nvidia chips are like ridiculously expensive and hard to get. By designing their own with Broadcom, OpenAI wants to control their own destiny, especially for inference.

**Morgan** (1:33)
Right, but custom silicon is notoriously hard to get right. Google has TPUs, Amazon has Tranium, and now OpenAI is entering the ring.
Late 2026 is still a long way off.

**SPEAKER_1** (1:46)
True, but Broadcom is a powerhouse. They helped Google build those TPUs, so OpenAI isn't starting from scratch. They're basically renting Broadcom's brain power to bypass NVIDIA.

**Morgan** (2:00)
That is a fair point, but think about the cost. Designing ASICs costs hundreds of millions of dollars. Will this actually lower costs for ChatGPT users or is this just a margin play?

**SPEAKER_1** (2:12)
It is definitely about margins, but also speed. If they can run GPT-5 or whatever comes next on custom hardware, they can make it way faster. Like, imagine instant voice responses.

**Morgan** (2:26)
I mean, we'll see.
If Jalapeño can actually handle the massive load of ChatGPT without melting, it might be a game changer. But late 2026 is basically a lifetime in AI years.

**SPEAKER_1** (2:40)
True, but the hype is real. It shows OpenAI wants to be a full stack tech giant, not just a software company.
They want the hardware too.

**Morgan** (2:51)
Which is incredibly ambitious, maybe even a bit risky.
But hey, if you want to win the AI race, you have to control the hardware. It makes sense.

**SPEAKER_1** (3:01)
Exactly.
Speaking of things moving fast, let's talk about where all the top talent is going. Google is having a rough week, dude.

**Morgan** (3:12)
Yes, the great talent migration. I saw on TechCrunch that Google is having a serious brain drain.

**SPEAKER_1** (3:19)
Totally.
Top AI researchers, Jonas Adler and Alexander Pritzl just left Google for Anthropic. It's like a mass exodus, dude. Wait again?

**Morgan** (3:31)
Didn't they just lose some massive names recently?
Like who is that other guy?
John Jumper?

**SPEAKER_1** (3:37)
Yes, John Jumper, the Alpha Fold genius and Gnome Shazira both left recently too. Now Adler and Pritzl are heading to Anthropic.

**Morgan** (3:47)
This is getting critical for Google.
Jonas Adler was leading some of their core generative AI research. Losing people of that caliber to a direct rival like Anthropic is a massive blow.

**SPEAKER_1** (3:59)
Totally. Anthropic must be throwing some insane money at these guys. Or maybe they just want to work on Claude instead of Gemini.
What do you think is driving this?

**Morgan** (4:10)
I think it's culture. Google is a massive, slow moving giant now.
Researchers want to ship things fast. At Anthropic, they probably feel like they can actually build without all the corporate red tape.

**SPEAKER_1** (4:23)
That makes total sense. But like, Google is Google. They have the compute, the data, the money.
How do you walk away from that?

**Morgan** (4:33)
Compute doesn't matter if your projects get bogged down in committee meetings. Plus, Anthropic has Amazon's backing, so they aren't exactly hurting for compute.

**SPEAKER_1** (4:42)
True, but Google has been trying to restructure to keep these people. Remember when they brought Noam Shazir back for billions?
Clearly, that strategy isn't working for everyone.

**Morgan** (4:53)
Exactly. You can't just buy loyalty with billions if the research environment is frustrating.
If Google doesn't fix its internal culture, they will keep losing the talent war.

**SPEAKER_1** (5:05)
Man, the talent war is just brutal. Google has to be sweating right now.
But, hey, speaking of rivals beating each other, check out this translation news.

5 more minutes of transcript below

Feed this to your agent

Try it now — copy, paste, done:

curl -H "x-api-key: pt_demo" \
  https://spoken.md/transcripts/1000651996090

Works with Claude, ChatGPT, Cursor, and any agent that makes HTTP calls.

From $0.10 per transcript. No subscription. Credits never expire.

Using your own key:

curl -H "x-api-key: YOUR_KEY" \
  https://spoken.md/transcripts/1000774135398