CCTV Dialogue | Xi Wang Aims to Be the AI Industry's 'P&L Optimizer'
Drive down the per-Token price
"The next battle in AI isn't about who's smarter — it's about who can deliver intelligence more stably and cheaply."
On October 17, CCTV-2's Dialogue visited Hangzhou's "New Eight Steeds" — Zhan Wang, Co-CEO of Xi Wang (Sunrise), said on the program that what customers care about most right now is whether AI services are affordable, reliable, and fast.
"Xi Wang aims to be the profit-and-loss optimizer for the AI industry, committed to reducing the per-Token cost."
Why "All-in Inference" Was Written Into Company Strategy
"Among domestic GPU companies, we're the first to eat this crab."
"In 2024, we found that the vast majority of customer use cases for our S2 chip fell on the inference side," Wang said. It's foreseeable that the importance of AI inference will continue to grow, and the overall scale of inference computing will soon surpass training.
"We further see that intelligence is becoming a standardized commodity that can be procured on demand for the first time, giving ordinary users access to top-tier AI capabilities. The core factor affecting the democratization of intelligence will be who can produce intelligence more stably and cheaply."

Why Become the AI Industry's Profit-and-Loss Optimizer
"Token costs will appear on everyone's bill, just like water and electricity."
"From now through the next decade, we'll officially enter the era of human-machine symbiosis, with an average of two to three Agents serving each person," Wang predicted. In the future, the number of Agents could exceed that of humans, and Token demand growth is just beginning — humans will no longer be the main consumers of Tokens.
People will need to start accounting for Tokens and actively seek better cost performance. But the computing gap is widening. Token demand is growing exponentially while compute supply grows linearly. Wang judges that the compute gap will be difficult to alleviate within the next five years.
"So Xi Wang aims to be a domestic GPU company focused on inference computing, solving the last-mile problem for AI democratization."

What Qualifies as a Sufficient Compute Foundation
"For users, it comes down to three things: affordable, reliable, fast."
This is the starting point for Xi Wang's technology innovation.
"The S3 launched in January differs from mainstream GPUs — it only does inference, not pursuing the training-inference integration," Wang explained. First, it trims training modules, redirecting saved transistor and power budgets toward improving inference efficiency. Second, it swaps memory. It was the first to adopt an LPDDR6 solution, abandoning the traditional HBM approach. Third, it rewrites the interconnect architecture while natively supporting FP4 low precision, further boosting inference efficiency. "According to our measurements, compared to the previous-generation S2, the Sunrise S3 delivers roughly 5x inference performance improvement and about 90% reduction in per-Token inference cost."
"Why do supernodes matter today? Large models have already reached trillion-parameter scale, and they'll only get bigger." Xi Wang's 256-card supernode, built on high-speed interconnect architecture, enables 256 cards to achieve first-level interconnection, working together like one giant GPU. "This is an inference beast," Wang said.

"Only by driving down the per-Token cost will large numbers of intelligent applications truly take off across all industries, achieving intelligence democratization." This is the change Xi Wang hopes to bring about.


Heart Capital was founded in 2022 as a venture capital fund focused on investing in China's early-stage technology startups.
The Heart Capital team is primarily composed of founding partners and core investors from Lightspeed China, along with senior investors from industry backgrounds. The team's past investments include MetaX (688802.SH), Xpeng Motors (NYSE: XPEV, 09868.HK), Full Truck Alliance (NYSE: YMM), SUNMI (06810.HK), RoboSense (02498.HK), Ambiq Micro (NYSE: AMBQ), Hanshow Technology Co., Ltd. (301275.SZ), FinVolution (NYSE: FINV), HERE (NASDAQ: HERE), as well as LandSpace, MicroNano Space, Baichuan, Yunmanman Cold Chain Logistics, World Logistics, FanDeng Reading, and Lanhu.
Rooted in China with a global outlook, Heart Capital is dedicated to early-stage support and partnership with entrepreneurial teams that have the potential to become world-class companies in China's technology sector. Heart Capital champions the value of "heart," believing that technology can serve as a bridge connecting minds. Heart Capital looks forward to accompanying more young Chinese entrepreneurs onto the world stage.
