Staniszewski: Only a few global companies will pre-train foundational AI models
Mati Staniszewski · The $11B Bet That Voice Will Replace Everything | Mati Staniszewski x Nikhil Kamath | WTF Online · Nikhil Kamath · Mar 11, 2026 · at 40:48
ElevenLabs CEO Mati Staniszewski explains why foundational AI infrastructure will remain concentrated among a few global players despite data localization.
0:00 / 0:23exact quote · 23.7s
▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark
“Like let's take building foundational AI models. I think the kind of, the base will be pre-trained by few companies in the world, just given the resources required. They'll be open source of course offerings in that bucket. But then Probably, as I think about the government space or companies in all of the sectors, they will use those to fine tune it based on their data.”
quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →
More from Mati Staniszewski
Staniszewski: No AI bubble brewing in Silicon Valley because revenue is real
“I don't think that the bubble is brewing. I think there's, like, what I think is a very different part, as I compare it to other cycles, I think there's a clear value, clear revenue numbers for a lot of the companies.”
Staniszewski: Headphones will likely be the primary voice AI form factor
“I do think it most likely will be the most natural form factor given it's already one of the key devices.”
Staniszewski: Skeptical of valuations for GPU and inference providers built on NVIDIA
“It's all the, like, GPU providers that are built on top of NVIDIA that, like, help you get Effectively inference capacity. Like I'm a little bit skeptic on some evaluations there”
Staniszewski: Deep domain integrations create defensibility for AI startups
“If you know domain extremely well, then The AI model, this, you know, gives you a good starting point, but then integrations with the system, the legacy systems of how you deliver that, telephony how you bring the knowledge, that will take so much time. Then l…”