NVIDIA's Missed Opportunity – Jensen Huang
TL;DR
Jensen Huang says Anthropic is the exception, not the trend — he argues TPU and custom-ASIC growth is “100% Anthropic,” framing its Broadcom and Google TPU deals as a one-off driven by financing structure rather than a broad rejection of Nvidia.
Nvidia’s real miss wasn’t chips — it was capital — Huang says Nvidia wasn’t in a position to make the multi-billion-dollar investments Anthropic needed early on, while Google and AWS were willing to fund the lab and secure its compute usage in return.
He rejects the idea that custom accelerators only need to be ‘good enough’ — when Dwarkesh suggests hyperscalers can tolerate something “not more than 70% worse” to avoid Nvidia’s 70% margins, Huang counters that ASIC vendors also run rich margins, citing a hypothetical 65% margin and asking, “What are you really saving?”
Huang’s core claim is that building something better than Nvidia is much harder than people assume — he points to Nvidia’s scale, annual release cadence, and “big leaps every single year,” plus the number of ASIC projects that have been canceled, to argue the economics and engineering are brutal.
He explicitly calls this a lesson Nvidia won’t repeat — Huang says he didn’t fully internalize that VCs wouldn’t casually fund an OpenAI or Anthropic with $5–10 billion, and now says he’s “delighted to invest in OpenAI” and Anthropic to help them scale alongside Nvidia compute.
The Breakdown
Dwarkesh Pushes on the TPU Question
The clip opens with Dwarkesh pressing a pretty direct challenge: if Nvidia’s price-performance story is so strong, why are major AI labs like Anthropic and Google leaning into TPUs and Broadcom-built custom silicon? He points to Anthropic’s newly announced multi-gigawatt deal with Broadcom and Google, and notes that the market no longer looks “all Nvidia” the way it once did.
Jensen Says the Premise Is Wrong
Huang doesn’t just disagree — he interrupts to say the premise itself is too important to leave uncorrected because it matters “to AI,” “the future of science,” and “the future of the industry.” His actual answer is blunt: Anthropic is a “unique instance and not a trend,” and without Anthropic, he says, there would barely be TPU growth at all.
“There’s Only One Anthropic”
Huang keeps hammering that point: there isn’t some broad abundance of ASIC opportunities, there’s “only one Anthropic.” When Dwarkesh brings up OpenAI’s AMD relationship and work on its own “Titan” accelerator, Huang concedes people will try alternatives, but says OpenAI remains “vastly Nvidia” and they’ll still do a lot of work together.
Nvidia’s Confidence: Go Ahead, Try the Other Chips
One of the stickier moments is Huang saying he’s not offended when customers experiment, because “if they don’t try these other things, how would they know how good ours is?” It’s classic Jensen swagger, but he pairs it with a serious point: Nvidia has to continuously “earn” its position, and many touted ASIC efforts have simply been canceled.
The Margin Debate Gets Specific
Dwarkesh offers the skeptical view in one sharp line: maybe competitors don’t need to beat Nvidia — they just need to be less than “70% worse” if they’re avoiding Nvidia’s 70% margins. Huang pushes back hard, saying ASIC suppliers like Broadcom also make extremely high margins; if Nvidia is at 70% and an ASIC vendor is at 65%, he asks, “What are you really saving?”
The Real Miss: Nvidia Didn’t Finance the Labs Early
Then the conversation turns from chips to capital, and this is the actual reveal of the clip. Huang says Nvidia’s mistake was not deeply internalizing how hard it would be to build a foundation model lab, and how much those labs needed multi-billion-dollar backing from their compute suppliers because “a VC would never put in 5, 10 billion” on spec.
Google and AWS Could Do What Nvidia Couldn’t
In Huang’s telling, Google and AWS won Anthropic not just with hardware, but with balance sheets. They could make huge early investments and tie that to compute usage, while Nvidia simply “wasn’t in a position” to do that at the time — though if the company had been as large then as it is now, he says he would have been “more than happy” to.
“I’m Not Going to Make That Same Mistake Again”
The clip ends with Huang saying the lesson has landed. He says he’s now delighted to invest in OpenAI and Anthropic and help them scale, making clear that Nvidia no longer sees itself as just a chip supplier — it wants to be financially embedded in the labs shaping demand.