We didn't need another earnings call to know where this is heading. But when Jensen Huang stepped onto the stage at a DBS Bank forum in Singapore earlier this year and named three Chinese AI models โ DeepSeek, Kimi, and GLM โ as examples of "excellent performance," something shifted in the global AI narrative. Not because a CEO praised competitors. That happens daily. But because the world's most powerful computing infrastructure gatekeeper chose, in front of an audience of institutional investors and Asian capital allocators, to validate the very models that Western export controls were designed to contain.
The subtext matters more than the soundbite. Huang didn't just say these models were good. He said open source is "the best way to improve efficiency." He said "the market will decide which models are best for different applications."
Those two sentences, when placed side by side, constitute one of the most strategically loaded statements in recent AI industry history. And unpacking why requires understanding not what Huang said, but what structural position allows him to say it.
The Open-Source Paradox Nobody Wants to Name
Open-source AI models are not a charity project. They are a demand-generation mechanism for compute. Every DeepSeek weight file downloaded, every Kimi API call made, every GLM fine-tune deployed โ each one creates downstream inference demand. And inference runs on GPUs. Specifically, it runs on NVIDIA GPUs in the vast majority of deployments, because CUDA remains the path of least resistance for any team that wants to ship a model to production without rebuilding their entire software stack.
This is not speculation. It is mechanical. When Meta released Llama weights, NVIDIA's data center revenue didn't suffer. It accelerated. When Mistral open-sourced its models, European cloud providers bought more A100s and H100s. The pattern is consistent: open weights act as a distribution channel for compute consumption.
Huang understands this better than anyone. His endorsement of Chinese open models isn't a geopolitical olive branch. It's a business development strategy dressed in the language of industry pluralism.
What the China Angle Really Signals
Here is where the story gets more interesting. The three models Huang named โ DeepSeek, Kimi, GLM โ share a defining characteristic beyond their origin: they were all developed under compute constraints. DeepSeek's efficiency innovations, including its Mixture-of-Experts architecture and FP8 training pipeline, were not academic exercises. They were survival adaptations. When you cannot buy unlimited H100s, you learn to do more with less.
Huang praising these models is, in effect, acknowledging that constraint breeds innovation. That is a double-edged admission for NVIDIA.
On one hand, it validates the narrative that efficiency-focused AI development will drive broader adoption, which means more inference demand across more diverse deployment scenarios. That is unambiguously good for GPU sales.
On the other hand, it implicitly concedes that the compute-scaling arms race โ the very dynamic that has driven NVIDIA's valuation to stratospheric heights โ may be entering a phase where diminishing returns force a pivot toward architectural efficiency rather than brute-force scaling. If DeepSeek can approach GPT-4-class performance with a fraction of the training compute, what does that mean for the assumption that ever-larger clusters are the only path forward?
Huang's answer appears to be: it doesn't matter, because the market will sort it out. And critically, he will sell shovels to whoever wins.
The Software Moat as Strategic Core
Buried in the forum transcript is a line that deserves more attention than it received: Huang emphasized that software will play a key role in efficiently utilizing infrastructure.
This is not a throwaway comment. It is a direct response to the most persistent bear case against NVIDIA โ that hardware eventually commoditizes, that competitors will catch up on raw FLOPS, that the moat is illusory.
Huang's counterargument, delivered implicitly through his broader framing, is that the moat is not silicon. It is CUDA. It is TensorRT-LLM. It is the decade-plus of developer tooling, library integrations, and optimization workflows that make NVIDIA hardware the default choice even when alternative chips offer comparable theoretical performance.
I have seen this firsthand. In 2024, I helped lead a pilot project integrating decentralized compute networks with AI agents for content verification in the Philippines. We tested multiple hardware backends. The team's consistent finding was not that NVIDIA GPUs were faster in absolute terms โ though they often were โ but that the software ecosystem made them dramatically easier to deploy, debug, and scale. Switching to non-NVIDIA hardware meant rebuilding pipelines, retraining team members, and accepting a longer time-to-production. For most organizations, that friction is itself the deciding factor.
This is what Huang means by software. He is not talking about model software. He is talking about the invisible layer of tooling that makes NVIDIA hardware sticky. And by praising open-source models that run efficiently on that tooling, he reinforces the ecosystem lock-in without ever having to argue for it directly.
The Quantum Timeline and the Hedging Strategy
Huang's claim that quantum computing will become mainstream in three to five years is the most aggressive timeline offered by any major technology CEO. It is also, I suspect, deliberately aggressive.
NVIDIA has been investing in quantum-classical hybrid computing for years. Its cuQuantum platform and CUDA-Q framework are designed to position NVIDIA as the orchestration layer for a future in which quantum processors handle specific computational tasks while classical GPUs manage everything else.
If quantum computing becomes viable in three to five years, NVIDIA wants to be the company that bridges the two paradigms. If it takes fifteen years, NVIDIA loses nothing by having made the bet. The statement is asymmetric in its risk profile. It costs nothing to predict an aggressive timeline and potentially positions the company as a first mover if that timeline proves roughly correct.
What concerns me is how the market will read this. Quantum computing concept stocks are notoriously volatile and prone to narrative-driven speculation. A single sentence from Huang could trigger a wave of capital flows into companies with questionable fundamentals and distant commercial prospects. I have watched this pattern before โ in 2021, during the NFT boom, I saw dormitory-mates pour savings into projects that had no product, no team, and no verifiable code. The same psychological dynamics apply here. A credible figure says something optimistic. The market extrapolates. Retail investors get hurt.
The difference is that quantum computing, unlike NFTs, has genuine long-term potential. The problem is timeline ambiguity. Three to five years is a span that can accommodate enormous variance in outcomes. Investors who position based on that statement alone are making a bet on narrative, not technology.
The Contrarian Test: What If Open Source Doesn't Win?
There is a blind spot in the open-source euphoria that Huang's remarks have amplified. The assumption is that open models will continue to narrow the gap with closed frontier models, and that this narrowing is permanent.
But consider the counterfactual. What if the next generation of AI capabilities โ advanced reasoning, long-horizon planning, multi-step tool use โ requires training regimes that are not easily replicable in open environments? What if the data pipelines and reinforcement learning infrastructure needed for frontier performance are inherently capital-intensive and proprietary?
In that scenario, the open-source ecosystem stagnates at a capability plateau while closed labs continue to advance. The gap widens rather than narrows. And the open-source advocates who celebrated their efficiency gains find themselves optimized for a game that no longer matters.
I am not predicting this outcome. I am noting that it is possible, and that Huang's framing โ "the market will decide" โ conveniently allows him to avoid taking a position on which scenario is more likely. His neutrality is not philosophical. It is strategic. He benefits from both outcomes, as long as both outcomes require GPUs.
The Institutional Divide and the Education Imperative
Last year, I founded a platform aimed at translating complex blockchain and AI regulatory frameworks into accessible guides for small businesses in Manila. We partnered with local banks to create a curriculum for SME owners, focusing on compliance and basic wallet security. The experience taught me something that applies directly to the AI landscape: the gap between institutional understanding and retail comprehension is widening, not closing.
When Jensen Huang speaks at a DBS forum, his words reach institutional investors who have the resources to act on nuanced signals. They understand that "market will decide" means "I'm neutral on models but bullish on compute." They adjust their portfolios accordingly.
Retail participants do not have that context. They hear "Jensen Huang praises Chinese AI models" and either over-interpret it as a geopolitical shift or dismiss it as public relations. Neither response captures the strategic reality.
The education gap is not a minor issue. It is the mechanism by which information asymmetry becomes wealth asymmetry. And it is why I believe that accessible, technically grounded analysis โ not hype, not doom, but clear explanation of incentive structures โ is the most valuable contribution anyone in this space can make.
The Takeaway: What to Watch Next
Huang's remarks are not a prediction. They are a positioning statement. They tell us where NVIDIA sees the next phase of growth โ in inference demand driven by diverse model deployment, in software ecosystem lock-in, and in optionality across compute paradigms.
The signals to watch are concrete. NVIDIA's next data center revenue report, particularly China segment performance, will indicate whether open-source enthusiasm translates into actual hardware demand. US export control policy discussions will reveal whether Huang's China-friendly rhetoric has political consequences. And the trajectory of Chinese model-plus-domestic-chip integration will determine whether NVIDIA's long-term position in the world's second-largest AI market is as secure as its CEO's diplomatic language suggests.
We didn't get a technical roadmap from Huang. We got something more valuable: a clear view of how the most important infrastructure company in AI thinks about the next three years. The neutrality he projects is not a lack of conviction. It is the luxury of a position so central that every outcome benefits it.
That is not a criticism. It is a description of power. And understanding power โ how it is structured, how it sustains itself, how it shapes the choices available to everyone else โ is the first step toward making informed decisions in a landscape designed to favor the incumbent.
The question is not whether Huang is right about open source, quantum computing, or the future of AI models. The question is whether we, as a community, are equipped to read the signals correctly when someone with his vantage point speaks. That capacity โ not any particular prediction โ is the real determinant of who benefits from the next phase of this industry.
Build that capacity. Then the noise becomes legible.