AI moat remains solid Broadcom(AVGO.US) on an unstoppable rise?

Zhitong Finance APP learned that, amid the recent release of financial reports, analysts still recommend holding Broadcom (AVGO.US) because its AI-driven moat remains intact; however, in a cautious market environment, there is almost no room for valuation to be wrong. AI revenue is structurally accelerating in growth, and partnerships with mega-cap companies are driving expansion. But customer concentration (especially Google) poses a major risk. Moreover, Broadcom’s infrastructure software business has very high profit margins, but customer churn and migration trends threaten the continued development of this high-profit area. The technical picture shows downside risk: $277 is a key support level. Before adding to positions or initiating a new position, patience and caution are warranted.

The moat is real

In the latest quarter, Broadcom’s total revenue was $19.3 billion, with the semiconductor solutions business accounting for 65%. This is important because in the prior fiscal year, this ratio was 58%/42%, and the main reason for this change was AI. AI revenue has continued to accelerate, growing for eleven consecutive quarters. Fiscal year 2023 revenue was $3.8 billion, fiscal year 2024 was $12.2 billion, fiscal year 2025 was $19.9 billion, and in just the first quarter of fiscal year 2026 it already reached $8.4 billion. This is structurally significant, because it requires considering the shift in mega-scale data center computing infrastructure paradigms. In other words, using XPU designed specifically for these companies’ workloads can deliver more predictable memory access patterns, fixed network topologies, and ultimately models that not only can execute millions of times, but also reduce total cost of ownership by 40% to 65% at the gigawatt scale. Broadcom co-designs these XPUs with its customers and, together with TSM.US, converts the architecture into manufacturable chips. The benefit of doing so is that it does not bear manufacturing risk.

Mega-scale data center operators are expected to invest $600 billion to $700 billion in capital expenditures, of which 75% will be used for AI infrastructure. In addition, Broadcom has already anticipated that the AI chip market size will exceed $100 billion in 2027. This is enough to show that if negative events occur, it would be due to mega-scale data center operators slowing their growth rather than Broadcom’s competitive advantages.

Regarding this segment, another important part is to dig into the composition related to AI revenue. First, look at Google’s (GOOGL.US) TPU program—this is the core engine, especially Ironwood’s significant evolution. Each generation of TPU deployed by Google implies more orders, confirming that this dependency relationship is expanding. Additionally, Anthropic recently publicly confirmed that it plans to procure up to one million Ironwood chips for Claude. This means Google has also begun selling chip capacity directly to third parties.

Meta Platforms (META.US) also confirmed that the hundreds of thousands of MTIA chips it jointly developed with Broadcom have been put into production use at Facebook and Instagram. In the next two years, the company plans to roll out four new versions, which means another industry giant is working closely with Broadcom on architecture design. Such cooperation is difficult to easily replace.

Of course, the less attractive side also must be mentioned, because not all mega-scale data centers rely on this company. For example, Amazon’s (AMZN.US) cloud service AWS works with Marvell Technology (MRVL.US) and AlChip to design Trainium chips, and Microsoft (MSFT.US) works with Maia. Therefore, for Broadcom, the most active XPU customers ultimately only include Google, Meta, Anthropic, and OpenAI. This indicates a trend toward concentration in AI revenue, with Google at the top.

This is important—this is a related risk: if 40% or 50% of revenue from this segment shifts to other suppliers or to internal design, it will affect the financial model, making the current valuation completely disproportionate.

Software business has high margins, but the earnings base is being squeezed

Broadcom’s post-VMware transformation is being carried out smoothly within management’s expectations. The software business’s gross margin has reached 93%, operating margin is above 77%, and it generated $27 billion in revenue in the last fiscal year 2025.

In the latest report, this business segment created $6.8 billion in revenue, up 1.4% year over year, which contrasts sharply with the 26% growth rate for all of fiscal year 2025. If the operating margin remains unchanged, RPO exceeds $45 billion, and this quarter’s order value is above $9.2 billion, then why is growth behind this outcome happening?

This situation arises because previously, customers could purchase a component and use it indefinitely. But with VMware’s new business model, they now must pay annual subscription fees for full software packages. Those packages sometimes include features that customers do not need, which ultimately affects billing, because billing is no longer calculated per server.

In other words, if a company’s infrastructure is similar to the current situation—one server can have 128 to 192 cores—then the cost borne by Broadcom clients is based on those cores. As a result, the company’s revenue grows multiple times without adding additional features.

So, this “forced” cycle is essentially already over. And much of the growth we saw in this business segment in fiscal year 2025 is largely the result of accounting treatment. Therefore, in this fiscal year, we will likely see that the current growth rate is the real level of this business.

Additionally, at this pace, revenue base erosion must be considered. For example, Nutanix reported in fiscal year 2025 the migration of 2,700 VMware customers. Red Hat accumulated roughly $500 million in OpenShift virtualization contracts over two years—plus, not to mention Gartner’s forecast that by 2028, 70% of enterprise customers will migrate 50% of their workloads.

Therefore, although the retained customers pay higher fees (as indicated by a 19% increase in annual recurring revenue), the profitability base is gradually shrinking. Faced with this situation, management argues that VCF will become an indispensable component of AI—that is, as enterprises expand their AI workload, they will need more VMware resources, thereby offsetting customer churn.

This viewpoint is understandable, but besides VCF’s technical integration with Nvidia (NVDA.US) and AMD (AMD.US), there is not enough evidence. At present, no mega-scale data center or AI frontier company has publicly commented on using VCF as the main infrastructure for its training clusters. This suggests that the real market should be enterprise inference in private clouds, and to a lesser extent, in frontier clusters.

In conclusion, although infrastructure software accounts for only 35% of the business mix, the segment’s profit margin makes it one of the key levers to sustain the company’s earnings power. Therefore, if customer churn rates further accelerate—even beyond the level described earlier—then Broadcom will lose its main growth momentum. Has the market already priced in this scenario into the stock price? Or, on the contrary, is this merely a temporary adjustment in the growth trend?

Technicals reflect the risks

For Broadcom, the technical structure of its price indicates that, given that the sequence of highs and lows has been trending downward since mid-December last year, the market has already digested some kind of risk. From the weekly chart, the 200-day moving average is starting to flatten, which may already be signaling the direction of the main underlying trend.

The main issue lies in the short-term daily chart, because the 30-day moving average (MA30) is forming dynamic resistance. Its slope keeps pushing the price downward. In addition, in the $320 to $350 range, trading volume has fallen sharply, marking the largest percentage decline in about the past 15 months. This could trap many sellers; to obtain liquidity, they will suppress any rebound.

Therefore, at current price levels, unless the trend changes, $277 seems to be the most likely direction. This is normal in an overall bullish market. The issue is that if the price breaks below that level while losing volume support, the most relevant trading range may be around $250. For now, caution is the best strategy.

Broadcom’s short-term rating is “Hold.” Its moat remains intact, and AI revenue continues to grow healthily. One of the “buts” is customer concentration tied to Google, because in an industry slowdown this could bring risks. But currently, there is no evidence that this is the case, nor evidence that Google will reduce its reliance on Broadcom. So the problem that needs to be addressed is more about the industry’s cyclical conditions rather than this specific relationship.

Another “but” is that infrastructure software has very high profit margins. This means that even though it makes up as much as 35% of the revenue mix, if the customer churn rate described in the article continues to accelerate, its impact will far exceed what the revenue-share proportion suggests. These factors are exactly what the current price action reflects, and what is causing this bearish-structure market. The current market environment is intensifying this bearish structure, and may cause investors to sell shares after a stock price rise to cope with potentially worse market conditions, thereby increasing liquidity risk.