Own Your AI Weights or Rent Them?

SUMMARY: Brandon and Aaron discuss the pros and cons of owning or renting your model weights. What does that mean for the Enterprise, and what should you be considering?

SHOW: 1055

SHOW TRANSCRIPT: The Enterprise AI Show #1055 Transcript

SHOW VIDEO: https://youtu.be/uc0GZBLgUeo

SHOW SPONSORS:

• Nasuni - Activate your data for AI and request a demo

 Topic: Own Your Weights or Rent Them?

• Why now? Alex Karp had a spicy CNBC segment arguing enterprises should "own their weights" rather than rent models from the big labs — sparking a widely-shared response from Jamin Ball on Clouded Judgement. Substack
• Past: Same shape as the "own vs. rent" debate the industry has had before — on-prem vs. SaaS, buy vs. build for ERP/CRM — just replayed one layer down, at the model layer instead of the app layer.
• Present: A weight file is really just a frozen snapshot that degrades in relative terms as frontier models keep improving — what actually matters is owning the RL/training loop that keeps producing better weights, not the weights themselves. A model RL'd against a company's actual workflows can beat a frontier generalist model on that one task, and do it far more cheaply — but that leaves enterprises managing a sprawl of task-specific models that all need governing, versioning, and securing.
• Future: Ball frames it as a stated-preference vs. revealed-preference problem — everyone says they want model sovereignty, but the spend data shows enterprises keep writing bigger checks to the frontier labs every quarter because most don't have the talent or infra to run the loop. Where's the market for a company that closes that gap — makes "owning the loop" accessible without the complexity tax? Tie back to your Show #4 (off-the-shelf AI, harnesses) — this is basically that debate's sequel, one layer deeper. (Aaron’s hot take, and another episode: maybe it’s not about the weights at all…)

FEEDBACK?

• Email: show @ the enterprise ai show dot come
• Bluesky: @TheEntAIShow.bsky.social
• Twitter/X: @TheEntAIShow
• Instagram: @TheEntAIShow

Transcript

# The Enterprise AI Show \#1055 - Own Your Weights or Rent Them?

[00:00:00] 


## Welcome Back Setup

**Aaron:** And we're back. And we are continuing our Models and Markets series, and joined once again by our guest host the person who is secretly trying to take over- ... the podcast, , Brandon Richard. How you doing, Brandon? 

**Brandon:** Fantastic. Good to be on again. 

**Aaron:** So what we're gonna do, , again, for those that aren't familiar, we pick a topic, we talk about why now, we talk about the past, we compare it to something in the past, and then we talk about present and future state as well.


## Karp Rant Sparks Debate

**Aaron:** And this is something we, we talked about it a little bit with the Alex Karp and, , going back a, a bit now, , his infamous and , CNBC segment- ... and the, owning your alpha and owning your weights and owning all of your AI and what makes you unique. And It actually, it's funny.

The more time has gone by, the more I feel like it has kinda changed the market. At first I was like, "Oh my gosh, this is a bit like a Jerry Maguire moment," but he seems to have gotten legs with all of it. Jim Inbal picked it up you know- ... on, [00:01:00] on, on "Clouded Judgment" and on his substack.

And for those that are out there, again, we recommend you go take a look at his. And i- it's interesting to me, and there's, we've got some other ones we're gonna record after this about Nvidia and others are kinda picking up this story, and everyone's-- It became really cool overnight to kinda bag on Anthropic and OpenAI- Yep

and closed models. And so let's talk about that, because I feel like it had its moment and it be- was very timely, but it almost became timeless all in one shot. What's your thoughts on that? 

**Brandon:** I guess we have to kinda tease apart. That rant. There's a lot in there. Yeah, there's a lot- that he said. 


## What Owning Weights Means

**Brandon:** So I think let's just focus in on what I think is one of the things he was trying to say in there, right? And he kinda says it, I think it, for him he kinda says it in the wrong way. He says , something along those like, "I wanna own my weights," right? Or something like that is what he says.

Yeah. And I think he, he's quick to kinda point out, especially to a financial audience "Oh, most people aren't technical. They don't get it." And it's like, even when-- I'd almost throw that back to him, is "Do you get what you're talking [00:02:00] about here?" 'Cause the point is, it's okay, let's, 101 here, right?

It's , you do all the trainings, the output is the weights, right? And that is, becomes, if you will, the intelligence of how all these tokens are created. So for the most part, once it's like out there, it's it's old. It's immediately old. So there's like somebody building a new model with new weights that has more information.

So I think what he's really trying, and I don't think it's-- So like owning the old weights, like you don't wanna own the old weights of like whatever OpenAI 3.0. Like it's dead. No one's using that anymore, right? All those old models are retired for the most part. So what I think he's really just trying to say is training, right?

Like we are allowing these Anthropic and OpenAI, they are training data constantly and looking for more data, and the more da-data that they have access to, the better models that they can make. No argument there. So I think his core kinda thing is is, and again, like we have to look at like everyone's motivations, right?

So for them, for o- Anthropic and [00:03:00] OpenAI, it makes total sense. It's like we wanna build the greatest models we can, and we wanna have the most information, and we wanna train it because that's how we're valued. That's how we make money. 


## Palantir Versus Frontier Labs

**Brandon:** If we look at Palantir, to my knowledge They are not creating a front- frontier model of any kind, tell me if I'm wrong, okay? And they have come out and explicitly stated that they want, that middle layer of their company, whose-- I just forgot the name of it, but whatever, like the intelligence layer, let's call it, that they provide, they wanna sit on top of models, and they want to be, if you will, right?

They wanna have the expertise of "Hey, listen, we know about all your secret data, and we'll pro- we'll use the model, but we'll protect the model from seeing all the secret data." The secret data could be defense data, it could be, manufacturing, whatever, right? Now, of course, they're seeing that data, right?

So it's okay, so like someone sees this data at some point. But he's making the case of no, these companies are gonna steal everything." And it's I don't know. To me, it doesn't, just doesn't like make sense of what he's trying to say there. It's yes, these companies are out to make the best model they can, and any data that they [00:04:00] can get, they'll use.

Just like Palantir is trying to interface and use the best LLM to provide, the, the data and the information for their clients, right? And they're gonna do it. So it makes total sense to me. I can see why they're competing, and I can see why he believes that like I want to have access to public LLM models that are open source, that are the best, and that's what we should do going forward.

Now, if you just kinda leave it there, fair-- Like that simple, in my mind, a simple argument is fair. Like you could say, "I believe in open models. I want it to be open models. I'm going to invest in building open models," which is part of the-- I think that what gets lost in that whole interview, that's why he was originally on the show, right?

Is to say is that he, his company, and NVIDIA have partnered to basically build an open model, right? Makes total sense. Now, I think where he like loses me is just when it's like all these allegations, about- Yeah ... these model makers and all , their devious. He even gets into like international politics [00:05:00] and all this other stuff.

I was like- He loves a good conspiracy theory ... he really just like to me, 

He, yeah, he just kinda loses me, and I just, and I-- And so to kinda just kinda take it back, it's like fair argument to make, "I wanna use open models. They're better for my company. I think they're better going forward." Fair argument.

Just like it's a fair argument to say, "Hey, listen, we can innovate faster in proprietary models like OpenAI and Anthropic are doing, and we see that those models are producing, the, the best models." At least right now they're the best models. Maybe not forever, but for the moment, they seem to lead the leaderboards.

So I look at it as two people competing, slightly different strategies. This is normal business. It will play out over many years to see who actually wins out and was right. 

**Aaron:** Yeah. 


## Owning the Account Edge

**Aaron:** And here's the thing. So , two things to that Number one, every company is always looking for an edge. And , like, whenever you hear this story or you hear others talk in the industry, it's always what's their edge, right?

And here's why , I'll say that. The models now, even though , yes, we're up to what? 3 trillion parameters or thereabouts as, as the top end right [00:06:00] now. The data, the publicly available data. Now, sometimes models get trained on sketchy data or data they shouldn't be, and the, I'm not here to judge that or talk about that.

But the public amount of data is pretty much gone through. So now I remember I wanna say it's Fable f- I think Fable 5 now with Anthropic, like you can't opt out of them capturing your inputs. And I think that also started to trigger some of that. Why? Because they're looking for an edge.

They need more data. Yeah. And guess what? If you can't opt out, it's more data to make their models better. So they're looking for an edge. And it all goes back to, I always try and take something on this and com- compare it to something different. And for those of us that have been in a, an industry background and done field sales, there's a concept of owning the account.

And so when I worked at a reseller, it was a really big deal of like who owned the account. Was it the reseller or was it the vendor? , If you're selling them VMware, did the VMware, , have the better relationship or did [00:07:00] you as the partner? Because if they ever got unhappy, who are they gonna kick out?

You know- ... kind of thing. And so like I also saw his little rant as a little bit of "I wanna own the accounts." Yeah. "I wanna be the one, you know, the rest, everything else is commodity. I wanna be the one you talk to and I wanna be the one you call and I be, I wanna be your trusted advisor," and all those things we used to always say when we, we sold to the field.

And so for me, this is nothing more than a, like everyone's trying to commoditize everyone else play. 

**Brandon:** Yeah. It's the old manager of managers, right? Like I wanna be the manager on top of everything else. And I think, he makes it, you kinda, the way you say it is it comes across as very like succinct and almost measured.

But like what he's, I think when you watch his rant, what he's saying is m- is more than that. He's saying that like they're stealing from you. They are literally stealing from you. You don't even know what's happening and they're taking everything from you. They're evil. I am your Batman. [00:08:00] I am your protector.

I am here. I am the only one that stands between you and y- you... The phrase he likes to use is alpha and, but that's just another word of saying like your competitive differentiator. Only I am the one that's willing to protect your competitive differentiation because you have to use my thing. And then he would go on and say why that is and then he would go and start to bring in geopolitics into the argument.

And it's like, okay, like I think he just goes so far with it, right? It gets a little, like to me it's a little, you know- I would agree ... outside of, it's just, and again, maybe that's why he's a billionaire. He's just , people are attracted to these just like grandiose kind of strategies and thoughts.

Whereas I look at it like, okay you're more like a regular software CEO who's seeking to protect their , base and where they belong inside the customers, which is fine. Again, that part is fine to me. I just think the other parts... , and I just think it and, , and I think one thing we don't talk about is and you know a lot more about this than I, but it's , we're just seeing this weird cycle I like to talk about.


## Distillation Open Source Cycle

**Brandon:** A, a frontier model gets created, right? And then in six months, the... It's usually an open source mo- [00:09:00] and a, a model out of China gets close to it. And then, but what people complain about is certainly Anthropic and OpenAI complain is that the way they're doing that, to your point, is they're distilling Anthropic and OpenAI's model in China, where they don't really, they're not respecting the terms of use, right?

'Cause they have different values. And then, but they, that's how they do it, right? They basically, they... It's not like they have more data or they have new algorithms. They basically are like, "We're gonna distill everything that you're doing down." And then they take that, and they release it open source, so it then comes back to the United States.

And then Amazon and Azure, as we talked about other shows, would be like, "Look, we got the op- Look, we got an open source model." And it's really, it's like it was that, that model's grandfather was probably the model that's one server over, right? Is Anthropic's probably sitting in the same room. , And again I know Anthropic and OpenAI would be like, "This is wrong.

We have to s- stop this." And it's just probably, you're probably right, but this is just not how the world works. They're not gonna stop doing it. And of course, Anthropic and OpenAI that's how they're making better models. They're distilling their own models. So it's just , everyone's distilling [00:10:00] everybody's models, and everyone's complaining that everyone else is cheating, but it's just like at some point it's not cheating.

If everyone's doing it, it's like, I don't know, we talked about college football a little bit with Brian. It's like at some point if everyone's doing it, it's not really cheating, it's just how the sport works, and that's kinda how I feel like open- building models is. It's like it's just a free-for-all.

And maybe it shouldn't be that way, but that's how it is. 


## Weights Degrade Fast

**Aaron:** And again, we'll go back to this, like the weight file, right? The weight file is it's nothing more than a froz- frozen snapshot. Yeah. And these days, it's a frozen snapshot that degrades pretty quickly. We're talking about days, weeks, maybe months, right?

Because the m- the, the industry just keeps moving forward. And what I find super interesting too is a lot of folks kinda talk about and Palantir, even Palantir does and Jamon talks about it on in his "Clouded Judgment," the, the newsletter on this, is okay, you wanna own your weights.

Why would you wanna own your weights? And it's this whole idea of you wanna own your weights, and then you wanna constantly improve them over time. Maybe it's reinforcement learning, all these other things. But take the model, make it better, and effectively have your [00:11:00] own model, right? I don't know.

I'm increasingly wondering like, okay, is that the best way forward? And that's something we probably don't have talk-- time to talk about today. We'll probably talk about that on another episode here. But the whole idea here is I'm actually gonna say, okay is it owning your weights? Is it owning your model?

Is it owning your alpha? You need to own something. And maybe that's the problem right now, and that goes into the future of like what exactly Do people need to own? 

**Brandon:** Yeah. I don't know that it's 

**Aaron:** that clear yet. 

**Brandon:** We won't get into it in detail today. What's your thoughts on that? 


## Model or Harness Prompts

**Brandon:** But I think the thing that, you know the, the question worth asking here is is it the model or is it the harness, right?

Because, and I highly recommend anyone go through some training. Pick your favorite, OpenAI, Anthropic, Aaron, you've got some companies to recommend, and be like, okay, like what do you used to do when you like build an agent, right? And because one of the things I think you'll learn is one of the things that you have to do as an agent, basically you're responsible for the system prompt, right?

You have to give it the prompt, right? And so what are these code harnesses? What is Claude Code and [00:12:00] Codex doing? It's like behind the scenes, they've written all these prompts, right? They-- Like you don't see them, but there's like all this prompt engineering that they've thought about, and that's what gives the model the tone, and that g- what gives the model why it can understand what you're doing.

And they've also taken the time to start to build out skills and build out tools, right? And it's like that stuff I am finding myself is like I'm starting to get more attached to that than I am actually the model. 'Cause I mentioned like Claude Code, be like, I'm happy. A lot of times I'm happy to live in the, the old times and use Sonnet because it's cheaper and I can just do more, right?

But I'm like, I have all my harness that I'm used to in Claude versus if I have to leave and go to something else, then I have to relearn the harness, and the harness is slightly different, doesn't do things quite the same thing. So back to like someone like Palantir and and the whole argument is maybe that's really what they should be able doing is listen, we have built, and I'm sure they've already done this, right?

They've, they have their own integration with these models. They probably built their own prompts, and people that like Palantir probably use the UI, and they [00:13:00] like that UI. So that, that really may be the place going forward. So it's really not owning the weights 'cause you just wanna plug in any model that's good.

It's owning all of that prompt and all of those system prompts to actually use it. So maybe that's where it'll ultimately end up, where Alex will be happy because he owns all the prompts, and they're Palantir prompts, and they're really good, and they're done it the way he wants, and that's a way to differentiate yourself going forward.

**Aaron:** Yeah. 


## Lock In Is the Future

**Aaron:** Ultimately, the, and again, the future, I'm gonna say, goes back to something that has been in the past in our industry for as long as I can remember, and that is all everyone is doing is posturing to lock you in. 

**Brandon:** That's right. That's right. No one wants to be locked in. Everyone wants to lock you in. No, no one wants to be locked in, 

**Aaron:** and every company on the planet wants to lock you 

**Brandon:** in.

Abs- Amen, brother. 

**Aaron:** So I think we'll stop there. We just solved both the past and the future all in one statement. All right. 


## Wrap Up and Feedback

**Aaron:** Brandon, I appreciate your time. Thank you very much. And everyone out there, thank you very much for listening. We certainly appreciate your time. And if you have any [00:14:00] feedback, we would love to hear it.

And if you have any ideas for guests and topics for future episodes, please send them our way as well. And with that, I'm gonna wrap it up, and we will talk to everyone next week.