Shares of chipmakers tied to artificial intelligence are increasingly being traded as bets on the broader data-center buildout rather than a narrow choice between graphics processors and custom silicon. After earnings updates from Nvidia and Broadcom, investors focused on each company’s ability to capture more of the AI infrastructure stack as hyperscalers continue to expand capacity.
While Nvidia is projecting revenue growth of roughly 70% in fiscal 2028 amid a “supply-constrained outlook,” Broadcom forecast an even steeper ramp for its AI semiconductor business, with management expecting AI revenue to more than double again through fiscal 2028. The strategic difference for investors centers on platforms and software reach versus customer concentration and custom-accelerator share.
Key takeaways
-
Nvidia is guiding for about 70% fiscal 2028 revenue growth, supported by ramping Blackwell deployments and expected scaling of Vera Rubin CPUs, according to management commentary.
-
Broadcom expects its AI chip revenue to keep roughly doubling across the next two years, driven by continued wins with hyperscalers and frontier labs, the company said.
-
Investors are weighing growth versus breadth: Nvidia’s platform approach and software layer are positioned for a wider set of customers, while Broadcom’s demand outlook is more concentrated.
-
The near-term bottleneck theme is less about demand and more about supply—power, networking, memory, wafers, and system-level capacity.
What drove the market’s focus after Nvidia and Broadcom earnings
The “GPU vs. XPU” framing is losing clarity as both companies expand beyond their original chip categories. Nvidia is increasingly presenting itself as an end-to-end AI infrastructure provider, combining accelerated computing with a growing software ecosystem. Broadcom, meanwhile, is leaning into custom silicon paired with networking and infrastructure software delivered through its VMware business.
In Nvidia’s case, investors highlighted the company’s data-center growth and the broader monetization of its platform. Management reported data-center segment revenue growth of 117% year over year in the second quarter, reaching $89 billion. Nvidia also pointed to Grace CPUs, which it said have generated more than $5 billion in sales over the last year, as well as expectations that Vera Rubin CPUs will generate roughly $20 billion in sales by next quarter.
Nvidia’s outlook also reinforced the platform narrative. Management forecast revenue to more than double in fiscal 2028 as the new platform ramps, while emphasizing supply constraints through the period. In parallel, Nvidia’s software efforts—through CUDA and additional products—remain central to its strategy to embed models across the AI stack. The company also cited its $12.9 billion acquisition of Hugging Face as evidence of a shift toward where generative models are shared, not only where they run.
Broadcom’s earnings update underscored a different balance of drivers. The company said custom accelerators remain the primary growth engine, with XPUs accounting for 73% of AI semiconductor sales of $16.7 billion in the most recent quarter. It also pointed to networking solutions and infrastructure software as additional contributors, including infrastructure software revenue of $8.8 billion, up 29% year over year.
How each company positions for hyperscaler spending
A key implication from both earnings disclosures is that hyperscalers are not choosing one ecosystem exclusively—they are buying both types of solutions. Nvidia’s software and platform approach is meant to be reusable across cloud providers, while Broadcom’s business model is built around purpose-designed systems tied closely to the requirements of major customers.
Nvidia described its expanded relationship with Amazon Web Services, including an additional commitment of 2 million GPUs through fiscal 2029. The company also referenced supporting a 12-gigawatt data center campus that OpenAI is building in Ohio alongside SoftBank Energy. In addition, Nvidia highlighted growth in non-hyperscaler customer demand, which management said rose 138% year over year to $40 billion, strengthening the view that the company is not solely dependent on a small set of mega customers.
Broadcom’s story is similarly tied to large AI deployments, but with more concentration. The company cited expected demand paths tied to specific customers and generations of accelerators, including deployments associated with Anthropic and OpenAI, as well as ongoing design partnership work with Google Cloud for TPUs. Meta’s custom MTIA chips were also described as booked through 2027.
Growth forecasts versus valuation: what investors may consider next
Nvidia’s near-term forecast emphasized that even with strong demand signals, execution is limited by supply-chain constraints. Management guided current-quarter revenue to $108 billion and said its fiscal 2028 growth outlook is around 70%, calling it a supply-constrained scenario. Nvidia also pointed to constraints from components and systems, including memory, wafers, power, and shells, through the end of the next fiscal year.
Broadcom’s guidance was more aggressive for AI semiconductors. The company guided the current quarter to about $34.8 billion in revenue and said AI semiconductors are expected to grow 236% year over year to $21.7 billion. For full-year fiscal 2026, Broadcom raised AI revenue to $58 billion, reflecting 186% growth from the prior year. Looking forward, management forecast AI revenue of $115 billion in fiscal 2027, then $230 billion in fiscal 2028—implying that Broadcom expects its AI business to double again across consecutive years.
Despite Broadcom’s faster projected AI growth, the article’s key investment framing is that valuation and customer breadth could still matter more for portfolio construction. Nvidia is described as trading at lower valuation multiples relative to Broadcom in trailing and forward price-to-earnings terms referenced in the report, while also offering a broader platform and software distribution reach. The underlying tension for investors remains the same: whether the market should price the platform benefits and wider demand base of Nvidia more favorably than Broadcom’s more concentrated custom-accelerator momentum.
Bigger picture: the AI infrastructure bottleneck is shifting from demand to capacity
Both companies’ guidance points to a common constraint theme across the AI buildout. As hyperscalers scale clusters, the limiting factors increasingly include power availability, networking capacity, and key components such as memory and manufacturing throughput. That matters because it can slow the rate at which revenue converts into installed base—even when customer demand is strong.
Looking ahead, investors may watch management commentary for updates on supply availability, system-level ramps, and whether software monetization continues to expand alongside hardware deployments. Next catalysts likely include future earnings reports from both companies and additional datapoints on hyperscaler capex and data-center build timing, alongside any guidance refinements tied to capacity constraints.
