Buying a growth stock for the dividend is like dating a rockstar because they have a great 401(k). It misses the entire point of the relationship.
The financial press is currently tripping over itself to crown Cambricon as China’s "little Nvidia" because it managed to squeeze out a profit in 2025 and—God help us—issue its first dividend. They see a maturing tech giant. I see a white flag. In the high-stakes arms race of AI silicon, cash is ammunition. When a company stops firing that ammunition into R&D and starts handing it back to shareholders, it isn't winning. It’s conceding that it has run out of ideas on how to beat the competition.
The Profit Trap
Common wisdom suggests that profitability is the ultimate validator for a semiconductor startup. It isn't. Not in this cycle.
Nvidia’s dominance isn’t built on healthy quarterly margins; it’s built on the fact that Jensen Huang spends money like a man who knows the window of opportunity is closing. For a specialized AI chip designer like Cambricon to pivot toward "capital discipline" and "shareholder returns" in the middle of the greatest hardware gold rush in human history is a tactical disaster.
If Cambricon truly had a roadmap to unseat the H100 or the B200, every single cent of that 2025 profit would be flowing into tape-outs for 2nm nodes or securing High Bandwidth Memory (HBM) supply chains. By paying a dividend, the board is effectively admitting they can’t find a better use for that cash than letting it sit in a retail investor’s brokerage account.
I’ve watched companies play this "maturity" card before. It usually happens right before they become a legacy footnote.
The Fallacy of the Little Nvidia Label
Stop calling them "little Nvidia." It’s an insult to the complexity of the stack.
Nvidia isn't a chip company. It’s a software company that happens to sell silicon. The moat is CUDA. Cambricon’s struggle has never been about whether their Bangxiu architecture can crunch TFLOPS. It’s about the ecosystem.
The "lazy consensus" says that US export sanctions create a vacuum that Cambricon will naturally fill. This ignores the friction of migration. If you are a Chinese cloud provider, switching from an industry-standard stack to Cambricon’s proprietary environment isn't a "seamless" transition—it’s a localized nightmare.
- Software Debt: Every hour a developer spends optimizing for a niche architecture is an hour they aren't building the next LLM.
- Interconnect Bottlenecks: Raw compute is cheap; moving data between ten thousand chips is expensive. Cambricon hasn't proven it can match NVLink’s throughput at scale.
By choosing to pay a dividend, Cambricon is signaling it lacks the stomach to subsidize the massive software ecosystem development required to actually compete. They are settling for being a second-tier provider for state-backed projects rather than a global hegemon.
The Geopolitical Safety Net is a Noose
Investors love the "national champion" narrative. They think the Chinese government will never let Cambricon fail. They’re right. But they’re also wrong about what that means for your portfolio.
Being a national champion means you are a utility. Utilities have capped upside. They exist to serve the state's strategic needs—specifically, ensuring domestic supply of mid-grade AI accelerators. That is a fine business if you want 4% yields and zero volatility. It is a terrible business if you are looking for the exponential growth characteristic of the AI sector.
When you see a 2025 dividend, you aren't seeing the fruits of a market victory. You are seeing the constraints of a company that is being forced to act like a "responsible" corporate citizen while its global rivals are burning billions to invent the future.
The Arithmetic of Failure
Let’s look at the math. If we assume a hypothetical R&D budget for a next-generation AI chip:
$$C_{total} = C_{design} + C_{tapeout} + C_{software}$$
In the 2026 landscape, $C_{software}$ is often $3x$ to $5x$ the cost of the physical hardware design. If Cambricon is diverting $D$ (dividends) away from $C_{software}$, the value of their hardware $V_h$ drops because the barrier to entry for users becomes too high.
$$V_h \propto \frac{1}{Difficulty_{integration}}$$
By choosing $D$ over $C_{software}$, they are intentionally lowering the utility of their own product to satisfy short-term price stability. It’s a classic agency problem. The executives want to keep the stock price buoyed to hit their own bonus triggers, even if it starves the company’s long-term technical viability.
What You Should Actually Be Asking
Instead of asking "How much is the dividend?" you should be asking: "Why didn't they buy more lithography equipment?"
The real winners in the next three years won't be the companies that show "profitable growth." They will be the ones that show "strategic deficits"—massive, calculated spending on the specific bottlenecks of AI:
- Advanced Packaging: CoWoS capacity is the real kingmaker.
- Memory Integration: Getting HBM3e on-die is more important than a quarterly report.
- Compiler Maturity: Making the hardware invisible to the researcher.
Cambricon is doing none of these at the scale required to lead. They are managed by accountants now, not visionaries.
If you want a dividend, buy a tobacco stock or a toll road. If you are in AI, you should demand that the company burns every dollar it makes until it either owns the market or goes to zero trying. Anything else is just a slow-motion liquidation.
Sell the dividend. Buy the burn.
