Summary of Key Points
Broadcom's financial report for the second quarter of fiscal year 2026 generally met market expectations, but the future guidance for its most high-profile AI business fell short of institutional forecasts (AI revenue expected to be $16 billion in the next quarter, versus the market's estimate of $17 billion; over $10 billion in the following fiscal year, versus institutions' projection of over $13 billion). However, Broadcom's networking business, which challenges NVIDIA's dominant technology, represents a hidden core strength. Additionally, the debt burden from the VMware acquisition has been completely absorbed, leaving the company in a healthy financial position. Short-term market confidence may be affected by these guidance figures, but in the long run, the iteration of AI chips (such as the TPUv8) and the value of its networking business are expected to drive growth.
I. AI Business: Accelerating Growth, but Lower Guidance
Broadcom's AI revenue for this quarter was $10.8 billion (a 24% increase from the previous quarter), in line with expectations, mainly driven by shipments of Google's TPU chips. However, the guidance for the next quarter ($16 billion, a 52% increase) did not meet the market's expectation of $17 billion. The annual target of $5.6 billion and over $10 billion for the following fiscal year are also considered "conservative" by analysts (who predict over $13 billion).
Why is the guidance lower?
- The collaboration model with new customer Anthropic has changed: from purchasing complete data centers to buying chips only, resulting in chip revenue accounting for only 20-30% of the total data center cost (although the gross margin is higher, short-term revenue is lower).
- Major client Google is diversifying its suppliers (possibly turning to other manufacturers for some chips), which adds uncertainty to Broadcom's AI business growth.
- The company's CEO, Hock Tan, misread the year during the earnings call, which gave the market the impression of poor management performance.
There are also positives:
The next-generation TPUv8 chips (available in both training and inference versions) will support FP4 technology, matching NVIDIA's latest Blackwell series in computing power. If Google starts supplying its own TPU chips, Broadcom's share of the AI market could increase.
II. Networking Business: The Hidden Strength Behind AI
Many focus on Broadcom's AI chips, but its traditional networking business is the real "trump card," designed to undermine NVIDIA's dominance in high-end networking solutions.
NVIDIA monopolizes high-end AI cluster networks with its InfiniBand technology (low latency and stable transmission, but only available with NVIDIA equipment). Broadcom, on the other hand, is leading the "Super Ethernet Consortium" (UEC), using open-source, standardized Ethernet to challenge this dominance:
- Tomahawk 6 switch chip: The world's first single-chip to exceed 100 Tbps bandwidth, reducing data center network complexity from three layers to two, significantly lowering latency and failure rates.
- Jericho series routers: Achieve lossless transmission at the same level as InfiniBand within Ethernet infrastructure, while being more flexible and cost-effective.
As AI moves into the inference phase, the importance of "connection" exceeds mere computing power (similar to logistics: you need not only trucks but also good roads). The value of Broadcom's networking business has not yet been fully recognized by the market and could become a key driver for increased valuation in the future.
III. Financial Condition: Healthy Enough for Further Mergers & Acquisitions
Broadcom's revenue this quarter was $22.2 billion (a 48% year-over-year increase), with a gross margin of 76% (excluding acquisition-related expenses). Operating expenses were well-controlled (at 18.3%). The most critical metric is the debt ratio: total liabilities divided by adjusted EBITDA, which has dropped to 1.8, returning to pre-VMware acquisition levels—indicating that the debt burden from the VMware deal has been fully absorbed. Broadcom may be looking for new acquisitions in the future.
In terms of profitability, core operating profit was $12.8 billion (a 22% increase), driven primarily by the AI business, reflecting a very stable financial position.
IV. VMware Integration: Complete, Now Focusing on Subscription Growth
Broadcom's software business growth previously came from the VMware acquisition and the switch to a subscription-based pricing model. The integration is now complete, and the debt has been settled. The software business's guidance for the next quarter is promising ($8.9 billion), with future growth depending on the expansion of VMware's subscription user base—similar to how video streaming services have shifted from one-time purchases to recurring subscriptions, providing more stable long-term revenue.
V. Future Opportunities and Risks
Opportunities:
- Mass production of TPUv8 chips will enhance Broadcom's AI chip competitiveness.
- Anthropic is expected to start contributing revenue in the second half of the year, with six major customers requiring 10 GW of computing power (corresponding to over $10 billion in AI revenue).
- The value of Broadcom's networking business in the AI inference phase is being recognized by the market, potentially leading to higher valuation.
Risks:
- Lower-than-expected AI guidance may dampen market confidence in the short term.
- Major clients like Google diversifying their suppliers could reduce their reliance on Broadcom.
- Cyclical fluctuations in the semiconductor industry could result in sluggish growth in non-AI businesses.
Overall, although Broadcom's AI business guidance fell short of expectations, its solid foundation (networking strength and financial health) makes it a company worth watching in the long term. Short-term stock price volatility is possible, but if the value of its networking business is recognized or AI chip innovation exceeds expectations, there is room for growth.