Nvidia와 Broadcom은 모두 견조한 실적을 기록했으며, 분석가 논의는 어느 AI 칩 강자가 더 큰 상승 가능성을 제공하는지에 집중되고 있습니다.

The artificial intelligence hardware race once looked straightforward: Nvidia sold general-purpose graphics processing units (GPUs), while Broadcom sold custom silicon, known as XPUs. That distinction is dissolving. After both companies reported earnings, it's clear that they now compete as architects of the entire data center, not merely chip specialists. The real question for investors is no longer which company makes the better accelerator, but who can capture more of the AI infrastructure stack as hyperscalers spend record sums on both custom ASICs and general-purpose chips.

Nvidia is no longer a pure-play GPU designer. During the second quarter, its data center segment generated $89 billion in revenue, up 117% year over year. Alongside GPU sales, Grace CPUs—launched in 2021—have generated more than $5 billion in sales over the last year. The new Vera Rubin CPUs are expected to generate roughly $20 billion in sales by next quarter, with management forecasting revenue to more than double in fiscal 2028 as the platform ramps. Nvidia's software layer continues expanding through CUDA, Nemotron, Cosmos, and Isaac. The company's $12.9 billion acquisition of Hugging Face is the clearest signal of this shift, underscoring how Nvidia is becoming the place where generative models are shared, not just the silicon they run on.

Broadcom occupies the opposite end of the chip spectrum. Custom accelerators drive growth, with XPU shipments comprising 73% of the company's $16.7 billion in AI semiconductor sales last quarter. Networking solutions are poised to grow just as fast in the upcoming quarter, while infrastructure software generated $8.8 billion in revenue—up 29% year over year. This growth demonstrates that the VMware acquisition has proven transformative for Broadcom's semiconductor business. Custom chips designed for Alphabet's Google Cloud, OpenAI, Meta Platforms, and Anthropic sit atop Ethernet switches, optics, and software that Broadcom provides. CEO Hock Tan's pitch is straightforward: custom chips designed around a specific customer's model requirements are cheaper and more power-efficient than one-size-fits-all GPUs.

Inside the data center, Nvidia and Broadcom operate on different levels. Nvidia positions itself as a reusable AI factory that any cloud provider or enterprise can plug into seamlessly. Broadcom aims to become a purpose-built engine for a handful of giant customers that will reorder by the gigawatt. The critical insight is that hyperscalers are buying both solutions, making the choice between the two stocks considerably more complex.

During its second-quarter earnings call, Nvidia revealed an expanded relationship with Amazon Web Services that includes another 2 million GPUs through fiscal 2029. The company is also helping fund a 12-gigawatt data center campus that OpenAI is building in Ohio alongside SoftBank Energy. With every major cloud provider already placing orders for Vera Rubin and the company's non-hyperscaler customer base growing 138% year over year to $40 billion, it's nearly impossible to envision the infrastructure supercycle without Nvidia touching multiple sockets.

Broadcom's story is nearly as large but far more concentrated. Anthropic is scaling from 1 gigawatt of Ironwood this year to 5 gigawatts of TPU v8i in 2027, giving Broadcom line of sight to 10 gigawatts by 2028. OpenAI is narrowly behind, with 1.3 gigawatts of Jalapeno inference chips deployed in 2027 and more than 5 gigawatts in 2028. Google Cloud remains a multiyear design partner for its Tensor Processing Units (TPUs), and Meta has three generations of custom MTIA chips booked through 2027.

The near-term math between the two companies is striking. Nvidia guided current-quarter revenue to $108 billion, excluding China data center compute. Fiscal 2028 revenue growth is pegged at roughly 70%, with CFO Colette Kress calling it a "supply-constrained outlook." Unconstrained demand would call for growth above 70%, but memory, wafers, power, and shells are bottlenecks through the end of next fiscal year.

Broadcom guided current-quarter revenue to approximately $34.8 billion, with AI semiconductors growing 236% year over year to $21.7 billion. For full-year fiscal 2026, AI revenue was raised to $58 billion, reflecting 186% growth from the previous year. Over the next two years, management forecasts AI revenue of $115 billion in fiscal 2027, surging to $230 billion by fiscal 2028—meaning Broadcom expects to double its AI business for two consecutive years.

While growth rates favor Broadcom's AI business, breadth and cash generation favor Nvidia. Valuation is what ultimately breaks the tie. Nvidia trades at approximately 28 times on a trailing price-to-earnings basis and roughly 23 times forward earnings. Broadcom trades closer to 46 times trailing P/E and about 31 times forward P/E.

Investors are thus paying less for a company still guiding to 70% growth, with a broader customer set and a software distribution platform that Broadcom lacks. While custom silicon will continue gaining market share from the frontier labs that can afford it, Nvidia sells the scarcer asset: a comprehensive platform on which many models can run, backed by a balance sheet capable of financing and securing power and infrastructure years in advance. Although its valuation multiples may not reflect it, Nvidia is already distributing this platform across hyperscalers, neoclouds, and sovereign enterprises simultaneously. This is why Nvidia has more upside than Broadcom and is the better buy of the two.