How Nvidia's Leadership Vacuum Opens the Door for Samsung
Jensen Huang's departure from NVIDIA creates an unexpected competitive opening for Samsung Exynos and shifts the balance away from CUDA lock-in. Here's who wins, who loses, and what happens next.
The Unlikely Beneficiary of a Silicon Shake-Up
When Jensen Huang stepped down as NVIDIA CEO, the market did something surprising: it rewarded Samsung Electronics instead of punishing it.
On November 13, Samsung shares climbed 2.4% to 213,000 won and SK Hynix gained 2.2% to 1,282,000 won. The KOSPI index touched 5,500, while the tech-heavy KOSDAQ surged nearly 4%. Meanwhile, across the Pacific, NVIDIA shed 4.3%, Micron dropped 1.6%, and Super Micro Computer plummeted 11.3%. AMD rose 5.9%. Broadcom gained 5.5%.
The divergence tells a story. The world now prices NVIDIA differently without its face, and Korea — long the backbone supplier of high-bandwidth memory for the AI revolution — stands to gain from whatever chaos follows.
But this is not simply a rotation trade. It is a structural shift in how AI compute will be manufactured, designed, and distributed.
Why Samsung Wins Without Lifting a Finger
Samsung’s advantage under this new scenario rests on one word: diversification.
The company already supplies HBM3E memory for AMD’s MI350X and MI355X accelerators, and it is locked in as the priority provider of HBM4 — the sixth generation of high-bandwidth memory — for AMD’s upcoming Instinct MI455X. That next-generation chip carries 432GB of HBM4 per device, a full 50% more than NVIDIA’s Vera Rubin architecture at 288GB.
In July, Samsung also expanded its memory-supply partnership with Broadcom for next-generation AI accelerator designs. Google, which builds its own custom silicon alongside chipmakers like AMD and Broadcom, also benefited in the session.
The logic is straightforward. If cloud providers diversify their GPU purchases away from NVIDIA — whether toward AMD accelerators or Broadcom-designed ASICs — every alternate path feeds Samsung’s memory orders. The supply chain does not collapse. It splinters. And Samsung sits at every junction.
An investor calling Samsung the “NVIDIA plus-one” winner is not being generous. They are observing reality.
The TSMC Problem
SK Hynix faces a different calculus, and its situation reveals why this moment matters for everyone except NVIDIA loyalists.
According to Bloomberg, roughly 27% of SK Hynix’s revenue comes directly from NVIDIA. The company lives closer to one customer than Samsung does. That relationship built the modern HBM market. But it also concentrates risk.
Here is where Taiwan Semiconductor Manufacturing Company enters the picture.
China’s Trend Taiwan reported on November 10 that NVIDIA may reduce the advanced packaging volume it entrusts to TSMC. For years, Huang personally managed the relationship — he told a podcast audience in April that three decades of business with TSMC had never required a single signed contract. That trust was real. But trust does not transfer automatically.
If TSMC’s most constrained capacity — advanced packaging for AI accelerators — opens up for other buyers, AMD, Broadcom, MediaTek, and other ASIC designers will grab it. Those companies use SK Hynix memory. So even a company heavily concentrated on NVIDIA could find itself benefiting indirectly.
But only if the HBM pipeline actually reroutes. SemiAnalysis flagged the exact question hanging over the new NVIDIA leadership: can a successor honor the unwritten agreements that Huang cemented through personal relationships? There is no guarantee.
The Neocloud Squeeze
While Samsung trades higher, the AI cloud companies that relied on NVIDIA capital are getting colder treatment.
CoreWeave, which grew through NVIDIA-backed GPU capacity, fell nearly 7% in extended trading. Nevius dropped more than 16%. Wedbush noted that if NVIDIA’s new management adopts a more conservative stance on investment and credit guarantees, these operators will find their expansion plans financing themselves into lower gear.
That dynamic is precisely why it matters that Samsung and SK Hynix have alternative customers waiting. If NVIDIA cuts back on funding the very infrastructures that buy their memory, the hole could widen — unless non-NVIDIA silicon picks up enough volume to fill it.
That pivot is underway. Cerebras, another American AI inference chip maker once labeled a potential NVIDIA rival, jumped over 17% after raising its annual guidance on the back of accelerating cloud revenue.
What Happens Next
Several things are converging at once, and none of them favor stagnation.
First, NVIDIA’s market position weakens not because their product disappeared overnight but because a single charismatic figure held the ecosystem together through informal channels. Replacing him requires rebuilding trust with TSMC, with HBM suppliers, and with the cloud giants who design around CUDA compatibility. That takes time.
Second, every competitor that secures HBM supply early gains leverage. Samsung’s multi-customer strategy means they are not betting on one player winning — they are betting that multiple players will need memory simultaneously. That is a safer hedge than SK Hynix’s current concentration, though both are exposed to the speed of diversification.
Third, Samsung Exynos and its GPU ambitions sit in the background of all of this. Samsung has spent years trying to prove its in-house graphics architecture could compete with ARM Mali and Qualcomm Adreno, let alone NVIDIA’s offerings. A distracted NVIDIA and an accelerated push toward heterogeneous compute architectures create an environment where even incremental progress looks meaningful. Whether Samsung can translate memory dominance into actual GPU design influence remains the open question.
The Bottom Line
Jensen Huang’s exit created an immediate pricing dislocation. The deeper story is slower-moving: the AI semiconductor ecosystem is fracturing along lines that benefit multi-vendor suppliers over single-vendor dependency. Samsung is already positioned there. SK Hynix is close behind, but farther from the edge.
NVIDIA will survive. The company’s products are too embedded. But the era of uncontested GPU supremacy — held together by personal relationships and informal agreements — appears to be closing. Whoever controls the next layer of hardware diversity gets richer. Memory makers are watching closely.