Initiating Coverage at Hold: A Record Print With AI Semis +46% and a FY26 Growth Rate Now Tied to Inference, But a Full Multiple and Concentrated XPU Customer Base Keep Us Patient

Rating Action: Initiating Coverage

This is our initiation report on Broadcom Inc. We begin coverage at Hold — a constructive but deliberately patient stance on a franchise that has, in the eighteen months since the VMware close, delivered a textbook example of how a disciplined acquirer extracts margin from an acquired enterprise-software asset while simultaneously inflecting an AI semiconductor business that is now pacing 60% YoY growth on a $4B+ quarterly base. Our Hold rating is not a critique of operational execution — the print is, by any objective measure, very good. It is a statement about the intersection of a fully-priced multiple, a narrowly-concentrated XPU customer base, and a non-AI semis floor that has taken longer to firm than we would have expected at this stage of the cycle.

Four reservations underpin the Hold rather than Outperform initiation:

1. Customer concentration in the AI piece is genuinely narrow. Hock continues to reference "three customers" deploying custom XPUs at scale (and "four prospects" in development). The print's AI upside this quarter was driven principally by AI networking (+170% YoY), not by a broadening of the XPU customer roster. The FY26 framing of "growth rate sustains" is conditional on those three customers continuing to scale on the trajectory implied by the existing pipeline. Any one of them pausing or pulling back — for hyperscaler-internal capex reasons that have nothing to do with Broadcom — reframes the FY26 algorithm materially.
2. Non-AI semis recovery is taking longer than expected. "Close to the bottom" with "bright spots" (broadband, enterprise networking, server storage sequentially up) is the right tone — but it is the same tone management has used for the better part of two quarters now. Wireless and industrial are seasonal/structural drags. The cyclical recovery in non-AI semis was supposed to be a tailwind alongside the AI ramp; it has been, at best, a non-headwind. We treat the non-AI piece as flat-to-modestly-up through FY26 in our base case rather than as a contributor to upside.
3. Gross-margin headwind from the XPU mix shift compounds. The Q3 guide of -130bps QoQ on consolidated GM is explicitly XPU-mix-driven; XPUs are slightly lower-margin than the rest of the semis portfolio (other than wireless). As the AI mix climbs through the year and into FY26, this dynamic continues. Software's 93% gross margin and 76% operating margin offset much of it — but the consolidated GM trajectory through FY26 is more likely to drift sideways than to expand. The earnings algorithm therefore relies more heavily on revenue compounding and on continued capital return than on margin expansion.
4. Valuation discounts continuity. AVGO has rerated meaningfully on the AI thesis through 2024-25 and has been priced as a marquee AI compounder rather than as a diversified semis-plus-software conglomerate. The current multiple already incorporates a multi-year continuation of AI semis growth and successful VMware monetization. The bull-case math works; it leaves only modest cushion for the inevitable normalization quarter when AI bookings cadence inflects or non-AI semis disappoints.

What gets us to Outperform: (a) the Q3/Q4 prints deliver the AI semis growth rate at or above the +60% YoY guide with continued networking strength, (b) management's "FY25 growth rate sustains into FY26" framing translates to a credible multi-billion-dollar XPU revenue inflection in fiscal H1 FY26 that can be triangulated from third-party AI data center capex disclosure, (c) one of the four XPU prospects transitions to a revenue-generating customer with timing that matters for FY26, or (d) a meaningful drawdown (15-20%) without thesis impairment that compresses the discounted-continuity premium. What gets us to Underperform: (a) any of the three XPU customers signals a slowdown in their AI capex commitments, (b) AI networking mix decelerates from the 40%-of-AI run-rate suggesting either competitive share loss or a normalization in scale-out spending, (c) the export-control regime tightens in a way that materially impairs the addressable AI semiconductor TAM in non-US geographies.

Results vs. Consensus

Q2 FY2025 was a clean print with a small but real upside surprise concentrated in AI networking and infrastructure software. The headline beats are modest in magnitude (revenue +0.1%, EPS +1.3%) but the composition is high-quality — this is not a beat driven by tax or a one-time benefit, it is a beat driven by AI networking running well above expectations and by VMware monetization continuing to convert.

Quality of Beat

• Revenue: Fully organic for the first time in the post-VMware era — Q2 FY24 was the first full quarter consolidating VMware, so the +20% YoY on the Q2 FY25 print is a clean organic comparison rather than a step-up benefiting from acquisition mechanics. The quality-of-revenue is high: AI semis +46% YoY (+~$1.4B in incremental revenue) plus VCF subscription conversions in software. Non-AI semis was a small drag but not a detraction from the organic compounder narrative.
• Gross margin: Consolidated GM of 79.4% on the print, ahead of the original guide on product mix. Semiconductor solutions GM of ~69% was up 140bps YoY on product mix; software GM of 93% (vs 88% a year ago) reflects the sustained shift to subscription. The Q3 GM guide of -130bps sequentially is the only forward-looking concern, and it is explicitly XPU-mix-driven rather than competitive-pricing-driven.
• Operating leverage: Q2 operating income of $9.8B was up 37% YoY; semiconductor operating margin expanded 200bps YoY to 57%; software operating margin expanded from 60% to 76%. The software operating-margin trajectory is one of the most underappreciated stories in the print — the disciplined VMware integration is converting acquired SG&A into structural margin at a faster rate than a baseline acquirer would deliver.
• Cash flow and capital return: $6.4B of free cash flow at 43% of revenue is below where a debt-free Broadcom would print — the VMware financing meaningfully impairs FCF conversion via interest expense and elevated cash taxes. As management has signaled, this is the principal reason the deleveraging path takes priority over share repurchases in the capital-allocation hierarchy. Even so, $4.2B of buybacks (25.3M shares retired) plus $2.8B of dividends in a single quarter is a substantial return-of-capital cadence.

Segment Performance

Semiconductors: AI Networking Carries the Quarter

Semiconductor solutions revenue of $8.4B (+17% YoY) accelerated from +11% in Q1, with the entire delta coming from the AI piece. Of the $4.4B of AI semis revenue, AI networking grew over 170% YoY and represents 40% of AI revenue — a striking inflection that surprised even management. Hock framed the surprise candidly:

"In fact, the increased density in scale up is 5 to 10x more than in scale out. And that's the part that kind of pleasantly surprised us and which is why this past quarter, Q2, the AI networking portion continues at about 40% from what we reported a quarter ago for Q1. And at that time, I said I expect it to drop. It hasn't." — Hock Tan, CEO

The mechanism is the migration of scale-up interconnect from proprietary protocols (NVLink) to Ethernet. With 72-128-XPU scale-up topologies becoming standard at the hyperscalers, and with each scale-up cluster requiring 5-10x more switching content than scale-out, Broadcom's Tomahawk + Jericho + NIC stack is collecting an outsized share of the bill of materials in each newly-deployed AI cluster. The newly-announced Tomahawk 6 (102.4 Tbps, the world's highest-radix Ethernet switch) extends that architectural advantage forward by enabling 100,000-XPU clusters in two tiers instead of three — a meaningful latency, bandwidth, and power-efficiency improvement that Hock argued will be central to next-generation frontier-model training.

Custom XPUs (the chip half of AI semis) grew "double digits" YoY — healthy but unspectacular relative to networking's +170% rate. Management explicitly addressed the apparent mix divergence: when Vivek Arya (BofA) suggested it implied XPU softness, Hock pushed back, calling the framing "trying to count how many angels on a head of a pin." Networking is hot but XPU is on its expected trajectory; neither is lumpy and neither is decelerating from where management saw it three months ago.

The FY26 framing is the new news. Management's previous disclosure had been a 2027 SAM target tied to three customers each deploying ~1M-XPU clusters. The Q2 disclosure adds a near-term overlay: "As we come into the second half of 2025 and with improved visibility and updates we are seeing in the way our hyperscale partners are deploying data centers, AI clusters, we are providing you some level of guidance visibility what we are seeing, how the trajectory of '26 might look like." The trajectory: FY25 growth rate sustains into FY26. The driver: hyperscalers "doubling down on inference" to monetize the training spend they have already done. As Hock put it: "Training makes [the model] smarter. You want to monetize inference. And that's what's driving it."

Stacy Rasgon translated the framing directly: at +60% AI growth in FY25, a sustained +60% in FY26 implies $30B+ of AI revenue for FY26. Hock declined to validate the precise math — "I'm giving you the trend" — but did not push back on the magnitude, which is in our reading a tacit confirmation that the algorithm is in roughly that range.

Non-AI semis of ~$4.0B was -5% YoY and remains the part of the print that bears watching. Broadband, enterprise networking, and server storage were sequentially up; industrial was down; wireless was down on seasonality. Q3 guidance keeps non-AI semis around $4B — broadband and enterprise networking continuing sequentially up but server storage, wireless, and industrial expected flat. Hock's "close to the bottom" framing is right but has been right for two quarters now; the cyclical inflection has not yet materialized.

Assessment: AI semis is doing exactly what the bull case requires, and the FY26 framing is incrementally constructive. Non-AI semis is a non-headwind rather than a tailwind. The semiconductor segment is firing, but the firing is overwhelmingly concentrated in AI — and the AI piece is concentrated in three customers.

Infrastructure Software: VMware Integration Continues to Deliver

Infrastructure software revenue of $6.6B (+25% YoY) beat management's $6.5B prior outlook, with the upside driven by continued conversion of legacy perpetual vSphere customers to the full VCF (VMware Cloud Foundation) subscription stack. Of the top-10K Broadcom customers, 87% have now adopted VCF — up from prior disclosure and on track to substantially complete the conversion cycle. ARR is growing double digits in core infrastructure software.

The operating-margin trajectory is the standout. Software OpEx of $1.1B against $6.6B of revenue produced a 76% operating margin — up dramatically from 60% a year ago. Hock and Kirsten Spears have consistently framed the VMware integration as a multi-year SG&A discipline campaign; the Q2 print is the cleanest evidence yet that the campaign is producing the structural margin lift that justified the LBO-style transaction.

The conversion runway has roughly 12-18 months left, per Hock's response to William Stein (Truist):

"Most of our VMware contracts are about typically 3 years... So based on that, the renewals, we are like 2/3 of the way, almost to the halfway — more than halfway through the renewals. So we probably have at least another year plus, maybe 1.5 years to go." — Hock Tan, CEO

Q3 software guidance of $6.7B (+16% YoY) implies a deceleration from +25% in Q2. This is a function of the YoY comp normalizing as the bulk of the conversion-driven uplift moves into the comparison base. The deceleration is expected and is not a quality-of-revenue concern; it is the natural shape of a multi-year subscription-conversion arc.

Assessment: Software is now mature in the integration sense. The remaining conversion runway will produce continued ARR growth but the rate of acceleration will moderate. The 76% operating margin, however, is now structural — that is the durable upside the VMware deal had to deliver, and it is delivering.

Key Topics & Management Commentary

Overall Management Tone: Hock Tan was characteristically blunt and visibility-focused. The notable tonal shifts: (1) more conviction than prior calls on the FY26 AI trajectory, anchored in inference acceleration as a new driver layered on top of the previously-flagged training demand; (2) more impatience with sell-side attempts to triangulate precise dollar guidance from his trajectory framing — the Stacy Rasgon "stop talking about SAM now" exchange was instructive; (3) explicit rejection of vertical-integration / new-protocol framings (NVLink Fusion, UALink) in favor of the open-standards Ethernet narrative. The overall tenor is of a CEO who feels he has more visibility than he had three months ago and is willing to say so, but who refuses to be backed into specific numerical commitments beyond the quarter's guide.

The FY26 AI Inference Pivot

The single most consequential disclosure on the call was the upsized FY26 AI framing. Management's previous public posture had been the 2027 SAM target tied to three customers each at ~1M-XPU scale. The Q2 disclosure introduces a new layer: hyperscaler partners are "doubling down on inference in order to monetize their platforms," and "we may actually see an acceleration of XPU demand into the back half of 2026 to meet urgent demand for inference on top of the demand we have indicated from training."

The implication, made explicit later in the call, is that the FY25 growth rate sustains into FY26. Hock confirmed this twice when pressed (by Harlan Sur and Vijay Rakesh):

"The rate we are seeing now so far in fiscal '25 and will presumably continue, we don't see any reason why it doesn't given lead time visibility in '25. What we are seeing today based on what we have visibility on '26 is to be able to ramp up this AI revenue in the same trajectory." — Hock Tan, CEO

The "lead time visibility" reference matters. AVGO's AI semis lead times remain at 35 weeks or better, which means the FY26 framing is not speculative — it is grounded in actual booked orders extending into the early portion of FY26. This is materially harder to dismiss than a forward-looking SAM extrapolation.

The inference angle is the genuinely new piece. C.J. Muse (Cantor) pressed on what workloads are driving the inference acceleration; Hock's answer was a thoughtful articulation of why custom silicon wins in inference once a hyperscaler has been on the learning curve for multiple cycles — the algorithm-to-silicon co-design generates compounding performance advantages that a merchant GPU cannot replicate. We agree with the directional argument; it is consistent with the broader thesis that inference, where TCO matters disproportionately, is the workload type where ASIC competition is most credible against merchant GPU.

Assessment: The FY26 framing is the new news, and it is constructive. We model AVGO AI semis at $30-32B in FY26 (vs. ~$19-20B implied for FY25), with networking growing slightly faster than XPUs in the mix.

The Tomahawk 6 Announcement and Scale-Up Architecture

The week-of-call announcement of Tomahawk 6 — a 102.4 Tbps switch enabling 100,000-XPU clusters in two tiers — is the architectural pivot that locks in the Ethernet-for-scale-up narrative. Hock's framing on the Ben Reitzes (Melius) and Tim Arcuri (UBS) questions was emphatic on this point: scale-up is rapidly converting to Ethernet for AVGO's hyperscaler customer base, and proprietary alternatives (whatever they are called — UALink, opened-up NVLink) are not relevant for that customer set.

"Scale up is very rapidly converting to Ethernet now, very much so. For our fairly narrow band of hyperscale customers, scale up is very much Ethernet... There is no reason to create a new standard for something that could be easily done in transferring bits and bytes of data." — Hock Tan, CEO

The "5-10x more density in scale-up vs. scale-out" datapoint is the dollar-content number that underlies the +170% AI networking growth. The current 40%-of-AI mix is, per Hock's response to Srini Pajjuri (Raymond James), expected to drift down toward "less than 30%" as XPU revenue scales faster than networking in FY26 — not because networking decelerates, but because the XPU base grows from a relatively smaller starting point. This is an important calibration: the +170% networking growth rate is not sustainable indefinitely, but the absolute dollar contribution of networking remains a structural multi-billion-dollar line item.

Tomahawk 6 is in proof-of-concept shipments today, with broader ramp expected through FY26. There is "tremendous demand" per Hock, but no large orders booked yet — we read that as a tailwind into FY26 rather than a Q3 lever.

Assessment: The Ethernet-for-scale-up architecture is increasingly hard to displace at AVGO's hyperscaler customer base. Tomahawk 6 is the next-generation product that secures the FY26+ networking content per cluster.

Capital Allocation: Deleveraging Takes Priority Over Buybacks

Aaron Rakers (Wells Fargo) asked the most strategically-important non-AI question on the call: how does AVGO think about capital return vs. M&A from here, given the integration progress on VMware? Hock's answer was a clear hierarchy:

1. Dividends: half of preceding-year free cash flow.
2. Debt paydown to a target of "no more than 2x debt/EBITDA" — the principal use of the non-dividend portion of FCF.
3. Opportunistic buybacks (the $4.2B of Q2 repurchases were "opportunistic" plus normal RSU-vest mechanics).
4. M&A "would be significant... would need debt in any case" — preserved as optionality but not a near-term path.

The deleveraging anchor is the most important point. Gross principal debt ended Q2 at $69.4B; subsequent to quarter end, $1.6B was repaid bringing it to $67.8B. The weighted-average coupon of 3.8% on $59.8B of fixed-rate debt is favorable, but the $8B of floating-rate debt at 5.3% is the piece that meaningfully impairs FCF conversion. As that floating-rate piece is amortized down, FCF conversion mechanically improves — and the $30B+ FY26 AI revenue base, applied to a deleveraged capital structure, drives the implicit FY27 EPS algorithm meaningfully higher than current Street estimates capture.

Assessment: The capital-return hierarchy is investor-friendly without being aggressive. The deleveraging path is the underappreciated FCF lever; we model it as a 200-300bps drag-removal over FY26-27.

Guidance & Outlook

The Q3 guide is constructive but not pyrotechnic. $15.8B beats the ~$15.7B Street consensus by ~$0.1B, AI semis at $5.1B (+60% YoY) is consistent with the trailing trajectory rather than an upside surprise relative to where the buyside had it modeled, and software at $6.7B (+16% YoY) is a clean continuation of the conversion arc with the comp now normalizing. The composition is what matters: AI semis is +60% YoY, non-AI is flat, software is +16%. The blended +21% consolidated guide is therefore overwhelmingly an AI story at the margin.

The 130bps sequential GM step-down is the principal modeling watchpoint. XPUs are slightly lower-margin than the rest of the semis portfolio; as the XPU mix rises, consolidated GM compresses. The 67% adjusted EBITDA margin in Q2 will likely settle to ~66% in Q3 and may drift slightly below that through FY26 as the AI mix continues to rise. This is a structural trade-off, not a competitive issue: AVGO is taking lower-GM revenue at very high incremental ROIC because of the AI volume, which is the right trade economically.

FY25 implied: H1 FY25 revenue of ~$29.0B (Q1 ~$14.0B + Q2 $15.0B) plus Q3 guide of $15.8B places 9-month FY25 revenue at ~$44.8B. If Q4 lands at ~$16.5-17.0B (consistent with continued AI sequential growth and modest non-AI improvement), full-year FY25 revenue tracks toward $61-62B — roughly +21% YoY against the $51.6B FY24 base. Adjusted EBITDA at ~66% would imply ~$40-41B of FY25 EBITDA. AI semis embedded in that: roughly $19-20B for the full year.

FY26 framing: If management's "FY25 growth rate sustains" framing translates to AI semis of $30-32B in FY26, and non-AI semis recovers to ~$17-18B (modestly above flat) and software grows ~10-12% to ~$28-30B, FY26 consolidated revenue lands at ~$75-80B — roughly +20-25% YoY. EPS scaling at the operating-leverage rate consistent with software margin expansion plus deleveraging benefit drives the FY26 EPS algorithm well above current Street consensus. We view this as the operative baseline for upgrade considerations.

Analyst Q&A Highlights

FY26 AI Trajectory and Inference

• Ross Seymore, Deutsche Bank: Asked whether the FY26 confidence comes from XPUs, networking, or both, and pressed on the inference angle. Hock confirmed both XPUs and networking are scaling, with inference "much more" visible than three months ago.
  Assessment: Direct confirmation that the FY26 framing is grounded in both pillars of AI semis, not just one.
• Harlan Sur, JPMorgan: Walked through the math: smoothed across the first 3 quarters of FY25, AI is +60% YoY; if that sustains into FY26 in line with the SAM growth CAGR, that is the "right way to think about it." Hock confirmed: "Yes."
  Assessment: The cleanest articulation of the FY26 algorithm. Management is explicitly tying FY26 AI revenue to the +60% growth rate, anchored in the SAM framework set out six months prior.
• Stacy Rasgon, Bernstein: Translated the framing into precise dollars: $19-20B in FY25 AI revenue at +60% growth implies $30B+ in FY26. Asked whether the SAM has gone up given inference is now layered onto training. Hock declined to engage on SAM ("Stop talking about SAM now") but did not push back on the FY26 dollar math.
  Assessment: The implicit confirmation. Hock will not validate the $30B number on the record but does not refute it; we read it as a tacit acknowledgment that the FY26 algorithm sits in that magnitude.

AI Networking Composition and Tomahawk 6

• Ben Reitzes, Melius Research: Asked what drove the upside in AI networking and whether Tomahawk 6 is part of the FY26 acceleration. Hock attributed the upside to scale-up density (5-10x more switching content vs. scale-out) and confirmed Tomahawk 6 is in proof-of-concept shipments today, with "tremendous demand" but no large bookings yet.
  Assessment: The 5-10x scale-up density datapoint is the structural reason AVGO networking is compounding faster than the AI cluster count alone would imply. Tomahawk 6 is a FY26 ramp, not a Q3/Q4 contributor.
• Blayne Curtis, Jefferies: Asked specifically about timing for shipping Ethernet scale-up to AVGO's main XPU customer. Hock's response: scale-up is "very much Ethernet" already at AVGO's hyperscaler customers.
  Assessment: This effectively closes the architectural debate at AVGO's customer base. The proprietary scale-up alternatives (NVLink Fusion, UALink) are competing for a different customer set.
• Karl Ackerman, BNP Paribas: Asked about co-packaged optics relevance for the higher-content scale-up opportunity. Hock indicated optical is 1-2 years away from displacing copper at >72-128-XPU scale-up sizes, and at that point CPO is "one path" but not the only one — pluggable optics are a viable interim solution.
  Assessment: Useful technical disclosure. The 512-radix Tomahawk 6 switch keeps copper interconnects viable at meaningfully larger scale-up sizes than the prior generation, which extends the time horizon before CPO becomes thesis-relevant.

Customer Concentration and Pipeline

• Vivek Arya, BofA: Asked what milestones to watch for the four prospects to transition into addressable-opportunity numbers. Hock declined: "We are not updating any numbers here... When we get a better visibility, clearer sense of where we are, and that probably won't happen until '26."
  Assessment: The prospect-to-customer transition is the single largest non-priced upside in the FY26-27 algorithm. Management's posture is to under-communicate until the conversion is mechanical — which is consistent with how they have managed the FY25 disclosure cadence. We do not model any prospect-to-customer transition in our base case.
• Srini Pajjuri, Raymond James: Asked whether any of the four prospects are contributing to the FY26 framing. Hock: "No comment. We don't talk about prospects. We only talk about customers."
  Assessment: Direct confirmation that the FY26 algorithm is built off the existing three customers. Any prospect-to-customer transition is incremental upside, not embedded.
• Vijay Rakesh, Mizuho: Pressed on whether new customers ramping in FY26 could push growth above the +60% framing. Hock declined to escalate the framing: "the rate of growth we are seeing in '25 will sustain into '26."
  Assessment: Management's posture is to set expectations at "rate sustains" and over-deliver if the prospect pipeline converts. We treat this as the appropriate analytical baseline.

Margin Trajectory and Revenue Mix

• Joshua Buchalter, TD Cowen: Pressed on the Q3 GM step-down being "below corporate average fall-through" given the implied $800M revenue increase but only $400-450M gross-profit increase. Kirsten attributed it to XPU margin dilution; Hock added that "there are more moving parts here than your simple analysis proves" and called the simple analysis "totally wrong."
  Assessment: The exchange suggests Hock is sensitive to attempts to pin down precise XPU GM levels. Our read is that XPU GM is meaningfully below the corporate average but not catastrophically so — the structural offset is the 76% software operating margin. We model XPU GM in the high-50s to low-60s and consolidated GM at 76-78% through FY26.
• Joe Moore, Morgan Stanley: Asked about export-control exposure given dynamic policy environment. Hock declined to forecast: "Nobody can give anybody comfort in this environment, Joe... I don't know whether it won't be."
  Assessment: The honest answer. AVGO has historically had less direct China data center exposure than NVIDIA, but the indirect exposure (AI semis going to hyperscaler customers that themselves have global exposure) is non-zero. Treat as an unbounded macro overhang.

Capital Allocation and M&A

• Aaron Rakers, Wells Fargo: Asked how AVGO weighs capital return against M&A given VMware integration progress. Hock laid out the four-tier hierarchy (dividends → debt paydown → opportunistic buybacks → M&A optionality), with the explicit framing that meaningful M&A "would need debt in any case" so deleveraging now preserves M&A capacity.
  Assessment: The most explicit M&A framing management has offered in recent quarters. The implicit message: another VMware-scale deal is plausible at some point, but the path runs through debt paydown first. We do not embed any M&A in our base case but recognize it as a real optionality value.
• William Stein, Truist: Asked how far along the VMware perpetual-to-subscription conversion is. Hock: "more than halfway through the renewals. So we probably have at least another year plus, maybe 1.5 years to go."
  Assessment: The conversion runway extends through roughly mid-FY27. Software revenue growth will continue to benefit from incremental conversions but the rate of acceleration moderates from here.

What They're NOT Saying

1. Specific XPU customer identities and revenue concentration. "Three customers" deploying at scale and "four prospects" in development is the disclosure. Industry-side reporting has identified the three (Google, Meta, ByteDance, with Apple as a fourth in development — though we treat industry attributions as inference, not management disclosure). Management will not name names and will not break out revenue concentration across the three. The implicit message is that one customer is materially larger than the other two; we model accordingly but the precise ratio is unknown.
2. The four prospects' identities and ramp timing. Hock's "we don't talk about prospects" line was emphatic. We assume two of the four are making credible progress toward customer status (consistent with the cadence of prior cycles), but the timing is unknown and management has explicitly said updates will not come before FY26. Any FY26 print that reveals a fourth XPU customer is meaningful incremental upside.
3. Precise FY26 dollar guide. "FY25 growth rate sustains" is the framing. Hock declined to validate the $30B+ math from Stacy Rasgon's question. The reluctance to commit to a specific number is rational management behavior — a precise dollar guide a year out invites quarter-by-quarter reconciliation pressure that constrains operating flexibility — but it does mean the FY26 algorithm has a wider error band than a precise guide would imply.
4. XPU gross-margin profile. Kirsten described XPUs as "slightly lower than the rest of the business other than wireless." That is genuinely vague. Hock's pushback on Buchalter's GM math suggests the precise XPU margin is sensitive intelligence the company prefers to leave unspecified. The implication: XPUs are dilutive to consolidated GM but not catastrophically so.
5. The non-AI semis cyclical inflection date. "Close to the bottom" has been the framing for two consecutive quarters now. Management is correctly cautious about calling a bottom prematurely, but the lack of a forward-looking inflection signal means the non-AI piece remains a flat-to-modestly-up contributor rather than a tailwind in our model. A genuine cyclical inflection would be a meaningful positive surprise we are not currently embedding.

Market Reaction

• After-hours move (June 5 evening): AVGO traded modestly weaker in the immediate after-hours session following the press release and call. The reaction was muted relative to the print's quality — the headline beat was small in magnitude (revenue +0.1%, EPS +1.3%), the Q3 guide was modestly above Street but not a blowout, and the gross-margin sequential step-down associated with rising XPU mix gave the bear case a near-term talking point.
• Next-day intraday context (June 6): The stock recovered into and through the cash session as sell-side reaction emphasized the FY26 framing and the AI networking inflection. The constructive elements of the print — +170% AI networking, FY26 AI growth rate sustained, VCF adoption at 87% of top-10K customers, 76% software operating margin — were re-weighted relative to the GM step-down.
• Pre-print context: AVGO had appreciated meaningfully over the trailing twelve months entering the print and was being priced as a marquee AI compounder rather than a diversified semis-plus-software conglomerate. Expectations were elevated, and a "clean beat without a blowout" outcome typically produces a muted-to-modestly-positive trading response on names with that setup.
• Volume: Elevated relative to trailing 30-day average, consistent with index-repositioning and rebalancing flows around a name that has become a top-ten S&P weight.

The market reaction is rational, in our view, and roughly proportionate to the print's information content. The headline beat was clean but small; the Q3 guide is constructive but not pyrotechnic; the FY26 framing is the most important new disclosure but is not a precise dollar commitment. We do not view the muted near-term reaction as an entry-level opportunity given the bias of the multiple to discount continuity.

Street Perspective

Debate: Is the FY26 "growth rate sustains" framing reliable enough to underwrite?

Bull view: Management has been deliberately conservative across the FY25 disclosure cadence and has consistently over-delivered relative to its own framings. Hock's explicit reference to "improved visibility" and "lead time visibility" (35-week lead times extending into FY26) anchors the framing in actual booked orders, not speculation. Inference adds a new demand layer on top of training, which expands the FY26 SAM relative to the prior framing. The $30-32B FY26 AI revenue path is achievable and may prove conservative if any of the four prospects converts to customer status during FY26.

Bear view: "FY25 growth rate sustains" is, on inspection, a fairly aggressive commitment. Sustaining +60% YoY growth on a $20B+ base for another full year requires either (a) substantial expansion at the existing three customers, (b) acceleration of one or more from the prospect pool, or (c) both. Hyperscaler capex commitments have historically been lumpy at this scale, and any single customer pause materially reframes the algorithm. The reliance on inference as a new driver is qualitatively right but quantitatively unprovable in advance.

Our take: Bull view is closer to right. The lead-time visibility argument is the meaningful constraint — AVGO is not speculating about FY26 demand, it is referring to bookings that are already on the schedule. We model AI semis at the bottom of the implied range ($30B for FY26) and treat anything above as upside. The asymmetric risk is a single-customer pause, which we do not currently embed.

Debate: Is the gross-margin compression a structural concern or a mix accounting issue?

Bull view: The XPU mix shift is the right kind of GM compression — it reflects AVGO winning more AI revenue at very high incremental ROIC, even though the marginal GM is below the corporate average. The 76% software operating margin offsets meaningfully, and at the consolidated level the operating-margin trajectory is still expanding. The GM compression is an accounting consequence of growth, not a competitive issue.

Bear view: A structural mix shift toward lower-margin product is, by definition, a structural margin headwind. If AI continues to grow as a share of total revenue through FY26-27, consolidated GM will continue to compress, and at some point the operating leverage from software margin expansion fully runs out (the 76% software margin is approaching the structural ceiling). EPS growth becomes increasingly dependent on revenue compounding alone, with no margin tailwind to amplify it.

Our take: Both arguments have merit. The bull view is right for FY25-26: software operating-margin expansion plus deleveraging benefit is more than enough to offset the GM mix headwind. The bear view becomes more relevant in FY27+: once software margin stabilizes near 78-80%, the GM compression starts to bite harder. We model consolidated operating margin expanding modestly through FY26 and stabilizing in FY27.

Debate: Does the equity discount continuity, or is there still room?

Bull view: AVGO is a unique asset — the only large-cap semis-plus-software platform with both a structural AI revenue tailwind and a structural margin expansion arc. Even at an elevated multiple, the FY26-27 EPS algorithm at +20-25% revenue growth, expanding operating margin, and aggressive deleveraging produces compelling forward returns. The premium multiple is justified by the optionality value of the prospect pipeline alone.

Bear view: The current multiple already prices a multi-year continuation of the AI compounder, successful VMware monetization, and a credible deleveraging path. There is little cushion for normal-course-of-business demand modulation in any of the three pillars. The first quarter where AI semis bookings cadence inflects, or where non-AI semis disappoints further, or where VMware conversions slow, will compress the multiple meaningfully.

Our take: The bear view is closer to right on the asymmetry of the setup. We do not view the current valuation as obviously wrong; we view it as offering modest forward returns relative to the operational risk being underwritten. The first hyperscaler signal of capex digestion at any of AVGO's three XPU customers will compress the multiple by 15-25%. We are comfortable being patient and waiting for either a cleaner setup or a non-thesis-impairing drawdown.

Model Implications

Valuation framing: At the current setup, AVGO trades at a forward P/E in the high-20s to low-30s on FY26 estimates — a multiple consistent with high-quality compounders carrying mid-teens-to-twenty-percent forward EPS growth, expanding operating margins, and meaningful capital return. The implied fair-value range lands roughly at-market — offering modest call-option upside if the FY26 algorithm prints at the upper end of our range and modest downside if non-AI semis disappoints further or if a single XPU customer pauses. This is the textbook setup for a Hold rating — a high-quality franchise at a price that already discounts continued execution.

The asymmetric scenarios that would change our rating: (i) a fourth XPU customer transitions from prospect to revenue with FY26 contribution → Outperform pull; (ii) any of the existing three XPU customers signals a capex pause → Underperform pull; (iii) non-AI semis prints a clear cyclical inflection (broadband, server storage, industrial all sequentially up for two consecutive quarters) → modest Outperform pull; (iv) export-control regime tightens further on US semis to non-US data centers → modest Underperform pressure but bounded.

Thesis Scorecard at Initiation

Overall: Five bull points are largely confirmed by this print, and four bear points are live or neutral. The central tension is that operational execution is excellent, the FY26 framing is incrementally constructive, and the VMware monetization arc continues to deliver — but the equity has already priced that trajectory, and the AI-customer concentration introduces a tail-risk that the current valuation does not adequately compensate. The Hold rating is a price judgment on a high-quality franchise, not an execution judgment. AVGO is in our universe of names we want to own at the right price; the right price is at-market or below, with a cleaner setup.

Action: Initiating at Hold (constructive bias). Upgrade triggers: (a) a fourth XPU customer transitions from prospect to revenue with FY26 contribution, (b) Q3/Q4 prints validate the +60% AI growth rate sustaining into early FY26 with no single-customer concentration anomalies, (c) non-AI semis prints a credible cyclical inflection across two consecutive quarters, (d) drawdown of 15-20% without thesis impairment that compresses the discounted-continuity premium. Downgrade triggers: (a) any of the three XPU customers signals a capex pause, (b) AI networking mix decelerates from the 40%-of-AI run-rate suggesting either competitive share loss or scale-out spending normalization, (c) export-control regime tightens in a way that materially impairs the non-US AI semiconductor TAM. We will revisit on the Q3 FY25 print.