Thread by @wmhuo168

Nvidia is cooked. Huawei already won. Barron’s won’t tell you, but here’s what they’re afraid to print.

barrons.com/articles/nvidi…

The Barron’s 2025 Nvidia year-end puff piece isn’t analysis. It’s a nostalgic hallucination that clings to past glory while ignoring the collapsing fundamentals beneath Jensen’s empire. Let’s tear it apart. (1/11)

“AI demand remains Nvidia-centric,” they say. Only if you ignore half the world. In China, CUDA is being quietly deprecated across sectors. Telecom, energy, defense, and even academia are ripping out Nvidia’s stack and replacing it with domestic AI clusters. (2/11)

What Nvidia offers in 2025 is PowerPoint. What Huawei delivers is a national-scale quantum-grade optical AI backbone with CloudMatrix clusters that already train models like DeepSeek and PanGu. Nvidia dreams in slides. Huawei operates in silicon and photons. (3/11)

Blackwell is hyped as a game-changer. Yet Blackwell doesn’t ship in volume until late 2025, and not to China. Meanwhile, Ascend 910C and homegrown compute fabrics are already in the field running real-time inference and massive LLM workloads. (4/11)

Barron’s handwaves past the obvious: Nvidia has been de facto banned from the world’s most aggressive AI economy. That’s not a speed bump. That’s a total market disintegration. And it’s irreversible. (5/11)

The “China needs Nvidia” myth is laughable. China has gone full-stack: chips, frameworks, interconnects, operating systems, even compiler chains. There’s no hole for Nvidia to fill anymore. CUDA is a stranded asset. (6/11)

Globally, Nvidia’s moat is leaking. BRICS nations and Belt and Road partners are buying Huawei’s AI stacks bundled with digital sovereignty and zero dependency on American silicon or Washington’s leash. (7/11)

The core lie of the Barron’s piece is temporal: it talks as if 2023’s Nvidia still exists in 2025. But that version of Nvidia was built on China revenues, cheap fabs, and unchallenged software dominance. All three are gone. (8/11)

Every major Chinese hyperscaler is now training on domestic chips. Benchmarks are emerging that show Huawei’s clusters beating Nvidia’s on energy, bandwidth, and real-world throughput. Not marketing FLOPS. Field performance. (9/11)

Nvidia is entering its IBM phase: still treated as important, still inflated on legacy perception, but no longer dictating the future. The age of CUDA imperialism is ending—not with a bang, but with a quiet global decoupling. (10/11)

Verdict: Barron’s didn’t write analysis. It wrote a eulogy dressed up in hopium. Huawei builds. Nvidia pitches. The market has moved on. (11/11)