The AI Race, Starship and Silicon Valley's Defense Bet

**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. China's latest AI breakthrough shaking up the industry. What Moonshot's Kimi K3 means for the race with OpenAI, Anthropic and the chip makers powering it. Plus, SpaceX gears up for a second attempt at a Starship test flight later this week. The stock's underperforming most of the other big Nasdaq IPOs. Andreessen Horowitz signed defense investor and military vet, Connor Love, to the firm's American Dynamism team. Love joins the show.
Good morning from San Francisco. The surprise release of Kimi K3 from China's Moonshot, it's still everything that people are talking about, reading about. We are rebounding in technology stocks.
Modestly, we're off-session highs where we opened earlier. Remember, chip stocks entered a bear market Friday, in part because of all of the questions posed by what Kimi K3, a 2.8 trillion parameter model represents. Some of those were like, well, if we have more efficient AI, do we need to spend as much on compute? NVIDIA is an example of a name that's rebounding. A lot of focus has also shifted to the intense memory requirements for what is happening in open weighted models. Moonshot's gone from AI upstart to global contender almost overnight. Kimi K3 drawing praise for rivaling the best models from OpenAI and Anthropic, but demand has been so strong that the company says it's temporarily limiting new subscriptions as it adds to work, works to add more capacity. Bloomberg's executive editor for Asia Tech, Peter Elstrom is with us from DC.
There are so many reports on the Bloomberg this morning. So many places we could start, but I want to go to one of the pieces the team did about, okay, it's Monday morning. We probably went into the weekend thinking, oh no, we don't need to spend as much on compute hardware in that context. Now we have a piece on the Bloomberg saying this puts a lot of emphasis on memory. What's the team in Asia writing about?

**Peter Elstrom** (2:25)
Well, yeah, let's broaden out the lens a little bit here.
What's going on is China is holding this World AI Conference in Shanghai, and we're seeing their companies come forward with new offerings in all sorts of different areas. Moonshot is certainly the one that got the most attention over the weekend. We saw this company that was probably not that well-known outside of China, if at all, and all of a sudden people are paying attention to it. As you mentioned, its performance is very close to the frontier models, 2.8 trillion parameters. Essentially, the brain synapses of this model are very significant. But also, you saw Alibaba come out with a new version of their Gwenn model that is very close behind. It's 2.4 trillion parameters. It's a very powerful model too. I think the big takeaway here that China AI is much closer behind the US in terms of AI capabilities than we really realized. If you thought China was 6 to 12 months behind before, now it looks like maybe they're only a couple of months behind. They're catching up quite quickly. And it's a broader selection of competitors than we really anticipated. People have been focusing mostly on DeepSeq in particular, maybe a little bit on Alibaba and then a couple of the smaller models. But now we're seeing companies like Moonshot come forward and they're very competitive. And a lot of people didn't even realize that they were going to be in this race.

**Ed Ludlow** (3:47)
There's some really important Bloomberg reporting on what Moonshot is looking at financially. We're going to get to that with Bloomberg's beta dip shots in just a moment. But away from the numbers, what do we know about the people behind Moonshot? What do we actually know about this startup and its team that have been just thrust into the limelight in such a quick and short space of time?

**Peter Elstrom** (4:10)
Well, the founder of Moonshot is a 33-year-old, Yang Jilin. He graduated from Carnegie Mellon. He got his degree in the United States. He decided to go back to China because of the possibilities that they had there. And really, before DeepSeek, he was kind of seen as the rising star within the AI universe. He's a big music fan. He named the company in Chinese after Dark Side of the Moon, the Pink Floyd album, which apparently is his favorite. They have conference rooms. But beyond being this geek and this music fan at the same time, he's been at kind of the cutting edge of AI developments there. And he made a very conscious choice. Rather than trying to make a lean, efficient model, he decided to go very big with this version of Kimi.