Analysts continue to recommend holding Broadcom (AVGO.US) following its recent earnings report, as its artificial intelligence-driven competitive advantages remain intact. However, in a cautious market, there is little room for error in its valuation. AI revenue is accelerating structurally, driven by partnerships with hyperscale companies, but customer concentration, particularly with Google, poses a significant risk. Furthermore, while Broadcom's infrastructure software business boasts extremely high profit margins, customer attrition and migration trends threaten the sustainability of this high-margin segment. Technical analysis indicates downside risk, with $277 identified as a key support level. Patience and caution are advised before considering adding to or initiating new positions.
The competitive moat is real. Broadcom reported total fiscal first-quarter revenue of $19.3 billion, with the Semiconductor Solutions segment accounting for 65%. This is significant because in the previous fiscal year, the split was 58%/42%, and the primary driver of this change is AI. AI revenue has seen sustained acceleration, growing for eleven consecutive quarters. Revenue was $3.8 billion in fiscal 2023, $12.2 billion in fiscal 2024, and $19.9 billion in fiscal 2025, reaching $8.4 billion in the first quarter of fiscal 2026 alone. This shift is structural, reflecting an evolution in the philosophy of hyperscale data center computing infrastructure. This involves adopting XPUs specifically designed for these companies' workloads, enabling more predictable memory access patterns, fixed network topologies, and ultimately delivering models that not only execute millions of times but also reduce total cost of ownership by 40% to 65% at gigawatt scale. Broadcom co-designs these XPUs with its customers and partners with Taiwan Semiconductor (TSM.US) to translate the architecture into manufacturable chips, benefiting from not bearing manufacturing risk. Hyperscale data center operators are projected to invest $600 to $700 billion in capital expenditures, with 75% allocated to AI infrastructure. Additionally, Broadcom anticipates the AI chip market will exceed $100 billion by 2027. This substantiates that any negative developments would likely stem from hyperscalers slowing their own growth, not from a deterioration in Broadcom's competitive edge.
Another crucial aspect of this segment is a deeper analysis of its AI-related composition. Starting with Google's (GOOGL.US) TPU project, which is the core engine, each new generation of TPU deployed by Google, especially the significant evolution with Ironwood, translates to more orders, confirming an expanding dependency. Furthermore, Anthropic recently confirmed plans to procure up to one million Ironwood chips for its Claude model. This indicates Google has begun selling chip capacity directly to third parties. Meta Platforms (META.US) has also confirmed that hundreds of thousands of its jointly developed MTIA chips with Broadcom are in production use for Facebook and Instagram. The company plans to launch four new versions within the next two years, signifying another industry giant engaged in deep architectural collaboration with Broadcom that would be difficult to replace. Of course, the less attractive side must be mentioned, as not all hyperscalers rely on Broadcom. For instance, Amazon's (AMZN.US) AWS collaborates with Marvell Technology (MRVL.US) and AIChip on its Trainium chip, and Microsoft (MSFT.US) works with Maia. Consequently, Broadcom's most active XPU customers are ultimately Google, Meta, Anthropic, and OpenAI, indicating a concentration of AI revenue, with Google at the forefront. This represents a relevant risk: if a segment contributing 40% or 50% of revenue decided to switch suppliers or move to in-house designs, it would impact the financial model, making the current valuation appear disproportionate.
Broadcom's post-VMware transformation is progressing within management's expectations. The software business has achieved a gross margin of 93% and an operating margin exceeding 77%, generating $27 billion in revenue in the last full fiscal year 2025. In the latest report, this segment contributed $6.8 billion in revenue, a 1.4% year-over-year increase, contrasting sharply with the 26% growth seen in fiscal 2025. With operating margins holding steady, Remaining Performance Obligations (RPO) over $45 billion, and bookings exceeding $9.2 billion for the quarter, what explains this growth? The situation arises because previously, customers could purchase a component and use it indefinitely. However, with VMware's new business model, they must now pay an annual subscription for a complete software package that sometimes includes features they don't need. This ultimately affects billing, as charges are no longer per server but based on core counts. If a company's infrastructure has servers with 128 to 192 cores, the cost to Broadcom's clients is calculated on these cores, resulting in a multiplicative increase in company revenue without adding extra functionality. Therefore, this "forced" upgrade cycle is largely complete, and the growth seen in fiscal 2025 was significantly influenced by accounting recognition. Thus, this fiscal year will likely reveal the current growth rate as the business's true run-rate. Furthermore, at this pace, the erosion of the revenue base must be considered. For instance, Nutanix reported 2,700 VMware customer migrations in fiscal 2025, Red Hat accumulated approximately $500 million in OpenShift virtualization contracts over two years, and Gartner predicts that by 2028, 70% of enterprise customers will have migrated 50% of their workloads. So, although remaining customers are paying more (as indicated by a 19% increase in Annual Recurring Revenue), the profitable base is gradually shrinking. In response, management argues that VCF will become an indispensable part of AI, suggesting that as enterprises scale AI workloads, they will require more VMware resources, offsetting attrition. This view is understandable, but beyond VCF's technical integration with Nvidia (NVDA.US) and AMD (AMD.US), there is insufficient evidence, and no hyperscaler or AI frontier company has publicly commented on using VCF as primary infrastructure for training clusters. This suggests the real market lies in enterprise inference within private clouds and, to a lesser extent, frontier clusters.
In conclusion, although the infrastructure software business contributes only 35% of revenue, its margin profile makes it a critical lever for sustaining company profitability. Therefore, if customer attrition accelerates further, the impact on Broadcom would far exceed what the sales percentage suggests. The question is whether the market has already priced this scenario into the stock, or if this is merely a temporary adjustment within a longer growth trend.
Technical analysis reflects the risk. For Broadcom, the price structure suggests the market has priced in some risk, given the sequence of lower highs and lower lows since mid-December. On the weekly chart, the flattening 200-day moving average may already signal the direction of the primary underlying trend. The issue is more pronounced on the short-term daily chart, where the 30-day moving average is acting as dynamic resistance, its downward slope continuously pressuring the price. Additionally, volume has declined significantly within the $320 to $350 range, marking the largest drop in nearly 15 months. This could indicate a large number of sellers are trapped, potentially suppressing any rebound to gain liquidity. Therefore, at current price levels, and unless the trend changes, a move towards $277 seems the most probable outcome. This is a normal occurrence within a broadly bullish market. The risk is that if the price breaks below this level with a loss of volume support, the next significant trading range could be around $250. For now, caution remains the best strategy.
Broadcom's short-term rating is "Hold." Its competitive moat is intact, and AI revenue continues its healthy growth. One caveat is the concentration of business with Google, which could pose a risk if the industry slows, though there is currently no evidence of this or of Google reducing its reliance on Broadcom. Thus, the issue relates more to the industry's cyclical condition than this specific relationship. Another caveat is the exceptionally high margins of the infrastructure software business. This means that even at 35% of the revenue mix, the impact of accelerating customer attrition, as described, would be disproportionately large. These factors appear to be reflected in the current price action, contributing to the bearish technical structure, which is exacerbated by the current market environment and could lead investors to sell on rallies to hedge against potential broader market deterioration, thereby increasing liquidity risk.
