← All episodes

FULL INTERVIEW: Why I Think Nvidia Is Perfectly Positioned In The AI Race

| 14 products mentioned
TBPN TBPN host
Tae Kim guest
Watch on YouTube nvidia ai infrastructure gpu demand inference computing ai agents semiconductor supply chain competitive positioning

Tae Kim argues that Nvidia is perfectly positioned to capitalize on the AI infrastructure boom despite recent stock volatility and market concerns about peak AI capex. The episode breaks down why massive inference demand from AI agents and coding assistants will drive sustained GPU demand, why Nvidia's acquisition of Grock and its supply chain advantages create defensibility, and why concerns about GPU depreciation and compute gluts are overblown given current demand dynamics. For builders evaluating AI infrastructure investments or the long-term viability of AI scaling, this is a framework for thinking through supply, demand, and competitive positioning.

Key takeaways
  • Inference demand is exploding due to AI agents and coding assistants—engineers at Meta, Google, and Nvidia are all experiencing compute shortages, and individual developers are hitting rate limits across multiple AI platforms, suggesting the constraint is real and structural.
  • Nvidia's Grock acquisition solves a specific infrastructure need: pairing Grock's low-latency inference (25% of demand) with Vera Rubin GPUs (75% of demand) creates an economically efficient solution for the current wave of agentic workloads that competitors can't easily replicate.
  • TSMC wafer capacity is the real constraint, not demand—Nvidia's relationship and prepayment leverage means they'll get priority allocation, but industry-wide fab capacity shortages will persist for years, making GPU supply the limiting factor for the broader AI market.
  • GPU depreciation is not currently a risk because secondhand GPU rental prices maintain 90-95% of original pricing over 5-6 year lifecycles, and demand outpaces supply across all hardware ages; a depreciation crisis only occurs if there's a genuine compute glut, which Kim doesn't forecast.
  • AI agent adoption will drive token demand across every vertical, not just coding—from customer service to drug discovery to chip design simulation; the analogy is spreadsheets and calculators, which didn't eliminate knowledge work but enabled workers to operate at higher abstraction levels while the underlying tool demand grew massively.
  • CPU demand is massively underestimated because agentic AI requires orchestration, tool calls, database queries, and web searches—the ARM CEO cited 4x more CPU cores needed versus last year, meaning the infrastructure play extends beyond GPUs.

Recommendations (4)

Cursor
Cursor uses

"I use Cursor daily for coding"

Tae Kim · ▶ 3:15

ChatGPT
ChatGPT uses

"I would try to use a chatbot, they would get it wrong. I did it like a few weeks ago and all the chat bots got perfect."

Tae Kim · ▶ 25:36

Gemini
Gemini uses

"I put in Gemini. I put in ChatGPT and just to make sure they're right and they're right."

Tae Kim · ▶ 26:10

Tequi Context recommends

"go subscribe to Tequi Context on Substack"

TBPN · ▶ 28:54

Mentioned (10)

B200 GPU
B200 GPU "people are like building bots to pick up any kind of B200 GPU they can that they're waiting like ..." ▶ 4:04
H100
H100 "where do we stand now with the fear that GPUs will depreciate precipitously and H100s will be wor..." ▶ 21:04
H200
H200 "Jensen literally said at GTC, they got licensed approvals on both the US and China side. So, we'r..." ▶ 10:09
Groq
Groq "Jensen acquired Groq, acquired the assets of Groq and the people of Groq" ▶ 4:44
Vera Rubin
Vera Rubin "the combination of integrating Groq's technology together with Vera Rubin lets Nvidia serve this ..." ▶ 4:50
DeepSeek "everyone was freaking out about DeepSeek, the super efficient models. We're going to destroy AI c..." ▶ 1:02
Qwen "the top people at Qwen left and we don't know where they went. Qwen is an amazing model. It's kin..." ▶ 8:53
CoreWeave
CoreWeave "CoreWeave has talked about these things are lasting five to six years and they're getting like al..." ▶ 21:11
Colossus 2 "the biggest win has been Colossus 2. Colossus 2 which was built very fast." ▶ 18:56
Starlink
Starlink "what Elon did with Starlink. I mean, that's a telecom infrastructure play and this will be an AI ..." ▶ 19:35