**SPEAKER_1** (0:02)
Bloomberg Audio Studios, podcasts, radio, news. Bloomberg Tech is live from the heart of Silicon Valley with Ed Ludlow in San Francisco.
**Ed Ludlow** (0:22)
This is Bloomberg Tech. Coming up, OpenAI unveils its first custom AI chip called Jalapeno, developed in partnership with Broadcom. Plus, SK Hynix is planning to raise $29 billion in a landmark US listing, racing to increase capacity to meet memory chip demand. And Cerebras reports quarterly earnings for the first time since going public. CEO Andrew Feldman joins us later this hour.
Let's get to our top story, custom silicon. OpenAI is tackling one of AI's biggest challenges, the supply and cost of computing. The company unveiled Jalapeno Intelligence Processor. That's the wafer, a custom AI chip developed with Broadcom, which the companies say carries a 50% lower cost versus a typical AI GPU. The focus, faster, cheaper AI inference. Shares of Broadcom opened pretty higher, around 2%.
We're off-session highs, but still up a percentage point or so in a market where tech at the index level, kind of flat. This is OutPerformance, Bloomberg Tech editor, Seth Figerman is here with the details. This is the big picture, right? OpenAI is compute constrained, heavily reliant on NVIDIA. They want to diversify. What do we need to know?
**Seth Figerman** (1:35)
Yeah, that's right. I think OpenAI wants to own more of the infrastructure stack, to have more flexibility here, and then I think importantly use the expertise that it's developed into what makes AI models run better and more efficiently and apply that to the hardware in the same way that I think we're seeing Google do with its TPUs and the hope would be that over time, they can build a chip that's more cost efficient and also better performing and eventually cuts down its own costs.
**Ed Ludlow** (2:00)
Bloomberg News spoke to both Hock Tan, the CEO of Broadcom, and OpenAI's hardware chief. I guess, you know, from Broadcom's perspective, there's quite a lot of fighting talk, right? They see a lot of demand for ASICs or custom silicon. But what was interesting to me is Hock Tan saying to us, they see a world where every Frontier lab goes to have a custom chip. Why?
**Seth Figerman** (2:23)
Yeah, I mean, it gets back to what we said about OpenAI a minute ago, I think giving them the versatility to kind of control their infrastructure stack and build custom hardware that meets their needs. And I think Hock Tan's comment, he's saying there's not that many of these Frontier model developers. So I think if you read between the lines, kind of talking about, is Anthropic going to move in this direction, and that's certainly a possibility. And to your other point, I think Hock Tan sounds quite bullish. Generally, on the prospects for this chip and ships like it going forward, he sees no limit to the demand for this kind of infrastructure, and is predicting that the demand and rollout for this chip next year will exceed his prior estimates for 1.3 gigawatts of supply.
**Ed Ludlow** (3:02)
On the OpenAI side, it's the latest example of them being prepared to spend. Did we get any sense from our conversation with the company about how this custom chip will rank in their overall plan for compute, whether it's leasing data center, buying NVIDIA chips, etc.?
**Seth Figerman** (3:20)
Yeah, they remain pretty tight-lipped on what the financing for this will look like. We have previously reported that OpenAI planned to spend tens of billions of dollars on these chips with Broadcom, part of its hundreds of billions of dollars in infrastructure commitments. Broadcom has previously said that they are standing up a chip financing vehicle that could help OpenAI's efforts, but it's unclear if that's the primary financing mechanism for what OpenAI will do here.
**Ed Ludlow** (3:43)
Bloomberg's AI editor, Seth Figerman, thank you very much indeed. SK Hynix plans to raise $29.4 billion in its US market debut, a deal large enough to rank among the top five share sales of all time. The listing comes at a time when markets are hungry for memory chips, and semiconductor companies are rushing for capital to expand capacity. Meet that demand. Bloomberg's Peter Elstrom, who leads our coverage of Asia Tech, joins us. That's part of it, right?
Expand memory fab capacity, raise a lot of money. There's also this idea of like valuation and prestige of an ADR situation where you get exposure to American investors.
**Peter Elstrom** (4:24)
Yeah, that's right. SK Hynix is a company we don't talk about that often on the show. Obviously, a leading memory chip company, but they often play second fiddle to Samsung, even in South Korea. But what we've seen from them over the past three years is really they've come from quite far behind to essentially the same market cap size as Samsung at this point. And that's largely because they've been so successful in these memory chips for AI applications, the HBM chips that we've talked about before. They essentially, HBM chips are DRAM chips stacked on top of each other, and their technology for doing that stacking has been very good. They got NVIDIA's blessing. They've been able to jump ahead of Samsung in that market quite effectively. Samsung has struggled a little bit. So now what we're seeing with this $29 billion fundraising is really them announcing to the US, hey, that they're ready to raise this kind of capital. They want more exposure in the US. And yeah, they'd like more of that PE multiple that you see with the US tech companies too. They've traded at a discount to that. If they get some of that ADR exposure, that may give them a little more opportunity to increase that valuation.
37 more minutes of transcript below
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/1000774087519