345: Chapter 345 CUDA Ecosystem Crisis

November in Shenzhen, the sky was as blue as if washed clean.

In front of the Shenzhen World Exhibition & Convention Center, the crowd surged.

The stream of people coming out of the subway station merged into a long dragon, winding its way toward the entrance of the exhibition center.

They carried backpacks, dragged suitcases, and spoke different languages—English, Japanese, Korean, German, French.

Security personnel held megaphones and repeatedly shouted, "Please show your credentials and enter in an orderly manner," but their voices were drowned out by the noise of the crowd.

In the largest exhibition hall of the convention center, all five thousand seats were filled.

The stage was designed with a futuristic style; on the massive circular LED screen, the Xingchen emblem slowly rotated.

Blue and white lights were projected from every angle, illuminating the entire space like a scene from a science fiction movie.

There were still over two thousand people standing at the entrance; they crowded into the aisles, standing on tiptoes, stretching their necks, only able to watch the live broadcast via the big screen.

They were all developers.

Coming from fifty different countries, speaking fifty different languages, but at this moment, their gazes were all focused on one place—the woman in the white suit at the center of the stage.

Li Wanqing stood on the stage, a giant countdown timer behind her.

The 38-year-old president of Xinghuo Chip was wearing a simply tailored white suit today, her short hair groomed meticulously.

There was no podium in front of her, no script, only a microphone.

"Over the past twenty years, CUDA has cultivated millions of developers. But who does CUDA belong to? It belongs to Nvidia."

Her voice carried through the speakers across the entire hall.

"Developers have written countless lines of code using CUDA, but this code can only run on Nvidia hardware. If one day, Nvidia raises prices or stops supply, what will you do?"

The audience fell silent.

Some people began to whisper to each other.

Li Wanqing continued, "Today, Xingchen is launching the 'Pangu Developer Program'—"

She turned around and pointed to the big screen behind her.

Pangu Developer Program / 1 Billion rmb Subsidy / Free Development Boards: 100,000 sets / One-on-one Technical Support: 7x24 hours / Open Source Toolchain: 100% Free

The audience began to stir.

Some people stood up.

"Live demonstration—a CUDA program, ported to the Pangu ecosystem."

On the big screen, code began to scroll. Li Wanqing signaled, and an engineer walked onto the stage and sat at the console.

Fifteen minutes. On the screen, the program ran successfully.

Performance comparison data popped up: Pangu vs CUDA: 30% performance increase.

Someone stood up to applaud, then more people followed. Finally, the whole audience stood up. The applause was like thunder, lasting for a long time.

November 14th, Palo Alto.

University Avenue is one of the most famous streets in Silicon Valley, lined with cafes, bookstores, and small boutiques.

Stanford students like to kill time here, entrepreneurs talk about financing, and programmers chat about technology.

At three in the afternoon, sunlight filtered through the sycamore trees on the roadside, casting dappled shadows on the ground.

In the "Coffee Society" by the roadside, two young developers sat in a corner booth.

The one wearing glasses was named Michael, twenty-eight years old, who had worked at Google for three years. The other was named Chris, thirty years old, a technical backbone at an AI startup. The two were alumni of Stanford and had kept in touch since graduation.

Michael lowered his voice, as if afraid of being overheard: "Are you still using CUDA?"

Chris smiled bitterly, stirring the latte in his cup: "No choice for company projects; clients specify using Nvidia cards. But privately, I've already started learning Pangu."

Michael's eyes lit up: "How is it?"

Chris put down the spoon and leaned forward: "It's very easy to get started. Xingchen's documentation is written much more clearly than Nvidia's. The key is—the subsidy is too aggressive. Write a technical blog post, get five thousand dollars. Port an open-source project, get twenty thousand dollars. I earned thirty thousand last month."

Michael was silent for three seconds: "Can you introduce me?"

Chris laughed, and the two clinked their coffee cups. Outside the window, a young man wearing an Nvidia T-shirt hurried past, a look of anxiety on his face.

At the same time, the Stanford University Computer Science Department.

The Computer Science building is one of the oldest buildings at Stanford, with red brick walls covered in ivy. In a lecture hall on the second floor, over two hundred seats were packed full.

Professor David Cheney stood on the podium, sixty years old, with white hair and gold-rimmed glasses. He had taught computer architecture at Stanford for thirty years and was one of the early evangelists of CUDA.

But today, he put down his lecture notes and looked at the students below: "I have some news to tell you all. For next semester's 'Parallel Computing' course, I am considering teaching the Pangu ecosystem."

A buzz arose from the audience.

A student raised their hand: "Professor, didn't Nvidia just announce that CUDA is free?"

Cheney smiled, a complex meaning in that smile: "Free? CUDA has always been free to download. The problem is, free doesn't mean open. Xingchen's Pangu is truly open."

He paused: "Besides, you need to find jobs after you graduate. Now, AI companies in Shenzhen, Shanghai Municipality, and Beijing are all using Xingchen's chips. If you learn CUDA, where will you find a job?"

The students began to take notes seriously.

Santa Clara, Nvidia headquarters is a huge gray building, located in an industrial park in the suburbs of Santa Clara. Its design is like an alien spaceship, low, flat, and full of futuristic feel.

In the headquarters' largest presentation hall, it seemed somewhat empty at this moment.

Jensen Huang stood on the stage, behind him a line of large text: "CUDA for Everyone."

The 55-year-old Nvidia founder was wearing his iconic black leather jacket, his standard attire for countless press conferences. But today, his expression was not as confident as usual.

Below the stage, there were less than two hundred people. Most were media, and some loyal fans. There were very few real developers. The presentation hall was too big; two hundred people sat in the front rows, and the rest were empty seats, looking particularly desolate.

Jensen Huang's voice echoed in the hall, with a hint of imperceptible emptiness: "From today on, CUDA is completely free and permanently open source. Any developer can freely use, modify, and distribute it."

No one applauded.

A reporter raised their hand, a senior tech reporter from Wired magazine: "Mr. Huang, Xingchen's Pangu program has been released for ten days. More than fifty thousand developers worldwide have registered, and twenty thousand open-source projects have been ported to the Pangu ecosystem. Do you think it's still in time to make it free now?"

Jensen Huang was silent for three seconds. For the first time, an expression appeared on his face that had never been there before—it was a complex emotion mixed with exhaustion, helplessness, and unwillingness.

"Nvidia will win back the developers."

He said, with no confidence audible in his tone.

At the same time, Li Wanqing sat in her office, watching the live broadcast of the Nvidia press conference on her phone. When Jensen Huang finished that sentence, she smiled.

She picked up her phone, opened Twitter, and typed a few words: "You are too late."

The attached image was a photo—at the entrance of the Shenzhen World Exhibition & Convention Center, developers queuing to register for the Pangu program stretched from the entrance all the way to the subway station. The line wound for hundreds of meters, a sea of heads, quite a spectacle.

Posted.

Three minutes later, retweets broke ten thousand. Ten minutes later, retweets broke one hundred thousand. Half an hour later, retweets reached five hundred thousand.

In the comment section, it was all messages from developers: "Just got back from Shenzhen, Pangu is really good."

"Already learning it, waiting to job hop."

"Leader Huang, the times have changed."

Li Wanqing put down her phone and walked to the window. Outside, the Shenzhen skyline glowed with golden light in the sunset. In the distance, several planes were landing; who knew how many programmers were on board, coming to Shenzhen to attend the Pangu Developer Conference.

In Palo Alto, at that coffee shop, Michael and Chris were sitting in their usual spots again.

Michael looked at the Twitter post on his phone and laughed: "Li Wanqing tweeted, 'Too late.'"

Chris leaned over to take a look and also laughed: "Indeed, too late. CUDA is free? What were they doing earlier?"

Michael picked up his coffee cup and looked out the window: "Do you think Nvidia can still make a comeback?"

Chris thought for a moment: "A comeback? Unless they can build chips cheaper than Xingchen. But is that possible?"

The two looked at each other and smiled. Outside, night fell. The lights of Silicon Valley lit up one by one, but among those lights, how many developers were opening the Pangu documentation?