Record AI Revenue, Cratering Stock: Broadcom Just Exposed the Incurable Disease of AI Investing

Looking at the next one to two months, the near-term trajectory for Broadcom shares is going to be set almost entirely by what the major hyperscalers say about their AI infrastructure spending during the Q2 earnings cycle running from late July through early August. If Google, Meta, Microsoft, and Amazon collectively raise their AI CapEx guidance — which I think is the base case — that becomes a direct catalyst for AVGO to recover the after-hours drop and push beyond it. The most critical data points to watch specifically are Google's commentary around TPU v6 chip demand levels, Meta's update on the next-generation MTIA co-development timeline, and any Amazon language around Trainium buildout plans. A constructive read from even two out of four of these companies would shift sector sentiment meaningfully in Broadcom's favor within a matter of weeks. Conversely, if any of the hyperscalers signals an AI investment pacing adjustment — even a modest one — Broadcom faces incremental downside pressure regardless of its own execution quality. My rough probability estimate for a 5–10% share price recovery within the next six weeks sits at approximately 60%, largely because the after-hours selloff had the hallmarks of emotional overselling rather than a genuine fundamental re-rating. A $140 million software miss does not logically justify the erasure of tens of billions in market cap when the core AI business is running at historic velocity.

The institutional positioning dynamics in the wake of this selloff represent the other critical short-term variable that most retail investors won't be tracking closely enough. Goldman Sachs and Morgan Stanley have both maintained buy ratings on Broadcom in recent published research, characterizing the custom ASIC franchise as an "irreplaceable strategic asset" — that kind of institutional conviction provides a meaningful floor for the stock as long-only funds interpret the dip as a reentry opportunity rather than an exit signal. On the other side of the trade, ARK Invest's Cathie Wood has been consistently vocal about AI infrastructure overinvestment risk throughout 2025 and 2026, and some hedge fund operators will likely use this volatility window to take profits on positions accumulated during the year-to-date 40% run-up. This divergence between long-only institutional buyers and momentum-driven sellers captures in miniature the broader uncertainty defining AI semiconductor valuations right now. From a technical analysis perspective, AVGO is likely to test support near the 200-day moving average — if that level holds with above-average volume confirmation, the setup for a V-shaped recovery improves substantially. If the 200-day breaks convincingly on volume, another 10–15% of downside opens up mechanically, bringing the stock to levels that would look genuinely attractive on a pure fundamental basis. The next three to four weeks will be decisive for establishing which direction becomes the dominant pattern.

Stepping back to the medium-term horizon spanning six months to two years, the most consequential variable becomes whether Broadcom's AI revenue concentration crosses the threshold that triggers a genuine market reclassification. If FY2027 AI revenue hits the $100 billion guidance target or exceeds it, AI's share of total revenue would cross 60% and potentially approach 70%. That crossing matters enormously because there's a well-documented tipping point in institutional equity market behavior: when a company transitions perceptually from "diversified infrastructure conglomerate with a large AI segment" to "AI pure-play with legacy software attached," the valuation multiple investors are willing to assign expands dramatically and quickly. Nvidia's remarkable re-rate in the second half of 2024 — when its AI revenue concentration cleared 65% and institutional allocation mandates shifted from generic "technology sector" buckets to dedicated "AI infrastructure" allocations — is the clearest available precedent. I believe the probability of Broadcom undergoing a comparable reclassification by the first half of 2027 is approximately 55%. If that reclassification actually happens, the P/E expansion from the current ~30x level toward 35–40x alone would justify $200–300 billion in additional market cap creation without requiring a single incremental dollar of fundamental earnings improvement. That asymmetric upside is the core reason why I don't think the post-earnings selloff represents the right moment to abandon the position.

On the competitive front, the medium-term story is more nuanced than the post-earnings selloff narrative suggests. Marvell Technology's extraordinary 32% single-day surge — triggered by nothing more than a Jensen Huang endorsement in a conference appearance — illustrates that the custom ASIC market is attracting serious competitive attention and capital allocation from the investment community. But the reality of custom chip co-development insulates Broadcom more than the Marvell headline implies: these are 2–3 year joint engineering programs with switching costs so high that existing customers simply don't walk away mid-cycle, regardless of what a competitor says at an industry conference. The more genuine medium-term threat is not Marvell but the gradual internal capability buildup within the hyperscalers themselves. Google's progressive internalization of its TPU architecture, Amazon's rapid maturation of Trainium and Inferentia, and Meta's evolving MTIA roadmap represent a structural directional trend that could, over a 3–5 year timeline, meaningfully compress the total addressable market Broadcom can capture through external design work. Within a strict two-year horizon, however, the probability of this threat materializing to a revenue-impacting degree is low — I'd put it at 25–30% or below — because full chip design internalization requires engineering organizations of a scale that even the hyperscalers are still assembling, and Broadcom's decades of accumulated ASIC design expertise creates a replication barrier that is genuinely difficult to overcome quickly.

Extending the view to the long-term horizon of two to five years, the fundamental question shifts entirely to the timing and shape of the AI investment monetization cycle. My baseline expectation is that 2028–2029 will bring what I'm calling the AI reality check: a period when hyperscalers face intensifying investor pressure to demonstrate concrete, quantifiable ROI on the hundreds of billions they've allocated to AI infrastructure buildout, and when the gap between AI infrastructure spending and actual AI-driven revenue generation becomes a central financial news narrative. In that environment, Broadcom's revenue growth rate naturally decelerates from 143% toward something in the 30–50% range — not because the business is failing, but because the law of large numbers and normalizing base comparisons are structural inevitabilities. McKinsey's projections put the global AI semiconductor market at roughly $80 billion in 2025, growing to over $300 billion by 2030. If Broadcom maintains a 15–20% share of that expanded market — consistent with its current trajectory — annual AI revenue in the $45–60 billion range becomes the long-term steady state, representing meaningful growth even from the current annualized run rate of approximately $43 billion. The bullish alternative scenario involves AI agent economies, autonomous vehicle systems, industrial robotics, and biomedical AI generating compute demand that extends well beyond the current LLM training wave, sustaining above-trend AI chip spending into the 2030s in ways the consensus isn't yet fully pricing.

Breaking the long-term into discrete scenarios clarifies the risk-reward picture considerably. The bull case, which I'd assign a 30% probability, involves FY2027 AI revenue clearing $100 billion and triggering the market reclassification described above, with P/E expanding to 35–40x and total market cap reaching $1.5–1.8 trillion. This scenario requires both hyperscaler AI CapEx remaining structurally elevated and Broadcom executing the VMware subscription transition cleanly over the next four to six quarters. The base case — my central expectation at 45% probability — sees AI revenue meeting guidance targets but software segment quarterly volatility continuing to disrupt sentiment, with AVGO trading in a wide band around current levels and market cap oscillating between $1.0 and $1.3 trillion. The bear case at 25% probability involves a meaningful hyperscaler AI investment pullback, accelerating competitive pressure from Marvell and internal hyperscaler chip programs, and P/E compression toward 20x or below, potentially putting market cap below $800 billion. Across all three scenarios, the binding variable is the same: Broadcom's long-term value is determined not by any single quarter's earnings print but by the duration and intensity of the AI infrastructure buildout supercycle — and the current evidence still points toward a cycle with years of runway remaining.

The ripple effects from this earnings event extend well beyond Broadcom's own price action and deserve tracking as a systemic indicator. The first-order effect is negative sentiment contagion across the entire semiconductor sector — Marvell, AMD, and ASML all declined alongside Broadcom in the after-market session, as investors reflexively repriced expectations for every AI chip-adjacent name. The second-order effect is that a single earnings print showing record-breaking results correlating with a stock decline materially strengthens the "AI overinvestment" narrative circulating across financial media and hedge fund strategy documents, creating a persistent headwind for sector-wide valuation multiples. The third-order effect — the one I consider most consequential for the broader AI investment thesis — is that Google, Meta, Microsoft, and Amazon now face heightened investor pressure to deliver clearer, more granular ROI metrics for their AI capital expenditure programs in their next earnings cycle. That pressure could cut either way: it might validate continued infrastructure spending, or it could trigger the first fractures in the hyperscaler AI spending coalition if ROI evidence proves thinner than expected.

For investors navigating this environment in practical terms, the core takeaway is this. If you already hold Broadcom, panic-selling into the after-hours decline is likely the worst available decision — the $10.8 billion in AI revenue and $73 billion backlog represent real, durable fundamental value that hasn't changed because of a software timing miss. A more disciplined approach is to hold through the hyperscaler earnings season and use the Q3 results in July–August to confirm whether the VMware software segment is stabilizing its revenue trajectory. For those considering a new position, the current price range near the 200-day moving average offers technical interest, but given that the broader AI sector expectations adjustment is still actively in process, a phased entry strategy over multiple tranches makes more sense than a single commitment. Most importantly: monitor Broadcom as a component of the AI semiconductor ecosystem rather than in isolation. With 80% of S&P 500 gains concentrated in 10 names, a single chip stock's decline can function as the first domino in a broader multiple-compression sequence. The pattern playing out here — record operational results, disappointed market expectations, brutal after-hours reaction — will repeat itself at each quarterly earnings cycle for the foreseeable future, and building that expectation explicitly into any AI semiconductor portfolio strategy is the most honest and practically useful response to the market environment we're actually operating in.