Record Everything and $30B of AI Bookings — But the Stock Falls 13% Because $16B Wasn't $17B: Maintaining Outperform as the Priced-for-Perfection Reset Improves the Entry

Results vs. Consensus

Q2 FY2026 Scorecard

Year-Over-Year Comparison

Quarter-Over-Quarter Comparison

Quality of Beat

Revenue: The +48% YoY growth was driven entirely by AI semiconductors (+143% to $10.8B, now 49% of total revenue). The beat versus consensus was modest (~$50–90M), but the composition is high-quality: AI semis came in above the company's own outlook, networking represented ~40% of AI revenue, and non-AI semiconductors (+6%, with bookings above $6B) confirmed the long-awaited cyclical recovery is underway. Importantly, the beat was operational, not driven by software (which landed in line at $7.2B). The one nuance for modelers: gross margin optics will keep declining as AI semis mix up — but that is dilution-by-success, not a margin problem.

Margins: The 67.3% operating margin (+200bps YoY) is the cleanest number in the report and the direct rebuttal to the "AI is margin-dilutive" worry. Even as gross margin fell 32bps YoY (semiconductors becoming a larger share of mix), operating expenses stayed essentially flat in dollars while revenue grew 48% — producing a record operating margin and demonstrating extraordinary operating leverage. CFO Kirsten Spears was explicit that the gross-margin decline "does not represent a structural change in semiconductor margin," and urged investors to model the semiconductor (~70% GM) and software (93% GM) segments separately.

EPS & FCF: The $2.44 non-GAAP EPS (+54% YoY) is fully operational, with the ~16% tax rate the only notable below-the-line item (driven by global minimum tax and geographic income mix). Free cash flow of $10.26B (46% of revenue, a record) is the number that anchors the valuation: Broadcom is converting nearly half of a $22B revenue base to cash while it is still in the steep part of the AI ramp. The 86-day inventory (up from 68) is a deliberate forward-supply build for the H2 AI acceleration, not a demand-softness signal.

Segment Performance

Segment Revenue Mix — Q2 FY2026

AI Semiconductor — $10.8B, +143%, and Bookings 3x Shipments

AI semiconductor revenue of $10.8B (+143% YoY) was the headline, but the more important number was hidden in the prepared remarks: AI bookings exceeded $30B in the quarter against $10.8B shipped. A ~3x book-to-bill at this scale is extraordinary and is what extended management's visibility "all the way to 2028." Networking represented almost 40% of AI revenue; XPUs (custom accelerators) made up the balance. Management framed XPU and networking demand as "simply insatiable," with the six core customers placing large, early orders to lock in not just wafers and HBM but the power and infrastructure that gate deployment.

"During the quarter, bookings for AI semiconductors were over $30 billion against the $10.8 billion we shipped... Our visibility runs all the way to 2028 right now. Just three months ago, I can tell you visibility ran pretty much to 2027." — Hock Tan, President & CEO

Assessment: A 3x book-to-bill and a four-quarter extension of visibility in a single quarter is the strongest possible refutation of the "AI digestion / air-pocket" bear case for Broadcom's custom-silicon franchise. The orders are multi-year and supply-constrained by power and infrastructure, not demand — exactly the profile that supports the >$100B FY27 framework. The market's fixation on a $16B-vs-$17.2B Q3 number missed the signal entirely.

Non-AI Semiconductor — The Cyclical Recovery Finally Arrives

Non-AI semiconductors grew 6% YoY to $4.2B, with bookings exceeding $6B — "a clear indication we are on the path towards a full cyclical recovery." Broadband, server storage, and enterprise networking were all up, partially offset by a seasonal decline in wireless. Q3 non-AI semi revenue is guided to ~$4.5B (+12% YoY), an acceleration.

Assessment: The non-AI recovery is a quietly important second engine. After several quarters of bottoming, a return to double-digit growth (Q3 guide +12%) adds a few points of consolidated growth and diversifies the story beyond AI. Bookings running ~1.5x revenue here mirrors the AI dynamic — customers ordering ahead of a recovery they can see coming.

Infrastructure Software (VMware) — 17% ARR Growth, AI-Adjacent Tailwind

Infrastructure software revenue of $7.2B (+9% YoY, in line) carried a 93% gross margin and a ~79% operating margin (+310bps YoY). ARR grew 17% YoY, and the just-released VMware Cloud Foundation (VCF) 9.1 — which adds heterogeneous compute support across AMD, Intel, and NVIDIA platforms for running AI Kubernetes and virtualized workloads on a common private cloud — is driving "extremely strong" on-prem deployment. Q3 software revenue is guided to ~$8.9B (+31% YoY), a notable acceleration.

"With strong server demand globally, the deployment of VCF 9.1 for on-prem cloud computing is extremely strong, driving robust revenue growth... we expect that to continue for the next multiple quarters." — Hock Tan, President & CEO

Assessment: The Q3 software acceleration to +31% is underappreciated against the AI noise. VMware is benefiting from the same server-buildout wave as the silicon business — high CPU core counts shipping alongside GPUs pull VCF licenses. A 93%-gross-margin, 17%-ARR-growth software franchise is the ballast that makes Broadcom's consolidated margin and FCF profile so durable.

Key Topics & Management Commentary

Overall Management Tone: Matter-of-fact and unhurried in the face of a stock priced for a beat-and-raise on every line. Hock Tan presented records (revenue, margins, EBITDA, FCF) as routine execution and characterized AI demand as "insatiable," but he deliberately declined to play the quarterly expectations game — reiterating rather than raising the FY27 >$100B number and explicitly saying "we are not trying to guide you every quarter what 2027 would be like." The posture was confident and supply-focused (the constraint is power and infrastructure, not orders), with the one notable forward lean being the unveiling of the Apollo/Blackstone XPU financing platform. The call also marked a leadership transition: long-time CFO Kirsten Spears retires June 12 after 12 years, with Amy Teiner stepping in.

1. Why the Stock Fell: $16B vs. $17B and No Raise to the AI Targets

The proximate cause of the ~13% selloff was a Q3 AI semiconductor guide of $16B (below the ~$17.2B whisper) and the reiteration — not a raise — of the FY26 (~$56B) and FY27 (>$100B) AI targets. When asked to reconcile the half-over-half math, Hock walked through it directly: H1 AI shipped ~$19B, 2x the first half gets to ~$56B for the year, and 2027 remains "very easily" in excess of $100B "on the same trajectory."

"We continue to be in excess of $100 billion in 2027... But we are not trying to guide you every quarter what 2027 would be like." — Hock Tan, President & CEO

Assessment: This is an expectations problem, not a Broadcom problem. The $56B FY26 and >$100B FY27 frameworks were already in the numbers; the buy-side had positioned for an upward revision and a Q3 AI print above $17B. Management's refusal to chase quarterly target-creep is, if anything, a sign of discipline — but in a momentum stock, discipline gets sold. The fundamental trajectory is unchanged-to-stronger; only the positioning reset.

2. $30B of AI Bookings and Visibility Extending to 2028

The single most bullish disclosure was the >$30B of AI bookings in the quarter (versus $10.8B shipped) and the extension of visibility to 2028. Management explained the surge as customers ordering far ahead to secure not just silicon and HBM but the power and infrastructure that actually gate deployment.

"They are placing their orders early and they are placing their orders now... in fairly huge demand, which basically gives us a lot more visibility than we normally otherwise would have in semiconductors. Our visibility runs all the way to 2028 right now." — Hock Tan, President & CEO

Assessment: Book-to-bill of ~3x and a year of incremental visibility added in a single quarter is the kind of forward signal that, in a less euphoric tape, would have driven the stock up double digits. It de-risks the FY27 framework and points to FY28 being "a substantial growth" year. This is the datapoint long-term holders should anchor on.

3. The Six Core XPU Customers and Their Multi-Gigawatt Commitments

Broadcom now frames its AI franchise around six core customers, each with multi-year, multi-gigawatt programs: Google (April long-term agreement for multiple TPU generations + networking), Anthropic (1+ GW in 2026, +5 GW of next-gen TPU compute from 2027), OpenAI (silicon delivered, production late 2026; 1.3 GW in 2027 within a 10 GW-by-2029 deal), Meta (multiple generations of MTIA XPUs; 3 GW through 2028, initial 1 GW order received), plus two additional customers (shipments late 2026, accelerating 2027; $6B of POs to date).

"For Meta, in April we announced a partnership to deliver multiple generations of MTIA XPUs... we expect to deploy 3 gigawatts through the end of 2028. Initial order for 1 gigawatt... will start delivery in the second half of 2027." — Hock Tan, President & CEO

Assessment: The customer roster has evolved from "design wins" to contracted, gigawatt-scale, multi-generation programs — a qualitatively more durable revenue base. The breadth (four named hyperscalers/labs plus two more ramping) reduces single-customer risk even as Google remains the anchor. Each program is measured in gigawatts and years, which is why visibility now reaches 2028.

4. The AI XPU Platform: Engineering the Financing of Demand

The genuinely new strategic disclosure was the AI XPU platform — a vehicle built with Apollo, Blackstone, and other large-balance-sheet investors to deploy >20 gigawatts of compute through 2030, with a $35B first tranche already being launched by Apollo. The purpose is to fund Broadcom's chips for frontier-lab customers (Anthropic, OpenAI) who need access to capital as much as silicon.

"To deliver this vision, we are creating the AI XPU platform with Apollo and Blackstone and other leading investors to deploy more than 20 gigawatts of compute capacity through 2030. The first tranche of this platform, valued at $35 billion, is currently being launched by Apollo." — Hock Tan, President & CEO

Tan was careful, in a later exchange, to correct a characterization that Broadcom was "backstopping" the Anthropic deal — clarifying that Broadcom supplies its own TPU chips to provide compute capacity, with the financing platform enabling capital-constrained labs to access that technology.

Assessment: This is a double-edged disclosure. On one hand, it is a clever demand-pull mechanism that removes the financing bottleneck for cash-hungry frontier labs and locks in Broadcom silicon at the lowest power and cost. On the other, it introduces an element of vendor-adjacent financing — the kind of structure that bulls read as visionary supply-chain engineering and bears read as circular, demand-pull-forward risk if AI capex ever wobbles. We lean bullish (the capital comes from Apollo/Blackstone, not Broadcom's balance sheet), but it belongs on the watch list.

5. AI Networking: The Other Half of the Story

Networking represented ~40% of AI revenue and is a genuine technology-leadership franchise, not a commodity attach. Broadcom has been shipping the industry's only 100-terabit Ethernet switch (Tomahawk 6) for over a year, will tape out a next-generation 200-terabit switch this quarter, is the de facto standard in co-packaged optics (1.6T DSPs, CW and EML lasers), and leads in cross-data-center fabric (Jericho 3 and Jericho 4). For scale-up within racks, it enables direct-attach copper on industry-leading 200G/400G SerDes.

"In networking, we have at least one generation of technology and product leadership... We will now be taping out our next-generation 200-terabit switch this quarter." — Hock Tan, President & CEO

Assessment: Networking is the underappreciated moat. As clusters scale, the value of connecting XPUs and GPUs rises non-linearly, and Broadcom's roughly one-generation lead (Tomahawk 6 → 200T tape-out, CPO standard-setting) is hard to dislodge. Management expects networking to normalize from ~40% toward ~30% of AI revenue as XPU volume scales — a mix shift, not a weakness, since the absolute networking dollars keep growing.

6. Gross Margin Optics vs. Structural Margin Reality

The Q3 gross-margin guide of ~74% (down from 77.1%) drew questions, but management was emphatic it reflects mix (AI semis growing faster than 93%-margin software), not any structural erosion. Within semiconductors, ASICs/TPUs and wireless carry lower margins while AI networking carries "very rich" margins that partially offset. Crucially, operating margin is guided flat at ~67% despite the gross-margin decline — the operating leverage is the story.

"This decline in gross margin does not represent a structural change in semiconductor margin. Rather, it reflects product mix... We highly recommend that investors model semiconductor and infrastructure software margins separately." — Kirsten Spears, CFO

Assessment: The market consistently misreads Broadcom's blended gross margin. As AI semis mix up, blended GM mechanically declines, but operating margin holds at record levels because opex barely grows. Investors anchoring on headline gross margin will keep getting this wrong; the segment-level view (semi ~70% GM, software 93% GM, both stable) is the correct frame.

7. Revenue Per Gigawatt: Content Is Rising Generation-to-Generation

Management addressed the economics of compute scaling: dollars of content per gigawatt are relatively stable (chips draw more power individually, so fewer chips per gigawatt, but at higher ASP), while the number of gigawatts is accelerating sharply. Over time, Broadcom's content per gigawatt rises as XPUs add SRAM, embed CPU cores, go multi-die, and integrate more HBM.

"Our revenue, our content per gigawatt will increase... the compute chip, the XPU, will go up in price very dramatically — embedding CPU cores into the same XPUs and making those chips multi-die with lots of HBM. That trajectory of content increases will go on." — Hock Tan, President & CEO

Assessment: Two compounding vectors — more gigawatts and rising content per gigawatt (over generational cycles) — underpin the multi-year growth algorithm. The content-per-gigawatt rise is gradual (generation-to-generation, not quarter-to-quarter), but it means the >$100B FY27 number is built on both unit and ASP expansion, which is more durable than volume alone.

8. The "Insatiable" Demand Thesis and Where It Comes From

Pressed on whether enterprise AI adoption changes the demand structure, Hock made a subtle but important argument: most AI compute demand still flows from a handful of frontier-model labs (Anthropic, OpenAI, Gemini) via their productized APIs (Bedrock, Vertex, Azure, first-party), even as enterprises and consumers do the consuming. The token consumption from enterprise and consumer use ultimately routes back to the same few frontier labs that buy Broadcom's compute capacity.

"Most of the demand at the end of the day, in terms of compute capacity... comes from those few large frontier model developers and the products they generate to supply to consumers and enterprises globally." — Hock Tan, President & CEO

Assessment: This framing is both the bull case and the concentration risk in one. The bull reading: a structurally steepening, sustainable demand curve through 2028 as token consumption compounds. The bear reading: Broadcom's growth is levered to the capex of a small number of frontier labs whose own economics are still unproven. We side with the bull on a 2–3 year view, given the contracted, gigawatt-scale visibility, but the concentration is the thesis's central vulnerability.

9. Non-AI Cyclical Recovery and the Diversification It Brings

Beyond AI, the non-AI semiconductor recovery (bookings >$6B, Q3 guide +12%) and the accelerating VMware software business (Q3 +31%) broaden the growth base. Both are running well ahead of their recent trend, adding ballast to a story the market evaluates almost entirely on AI.

Assessment: A fully-recovering non-AI semi business plus a re-accelerating software segment means even a hypothetical AI pause would land on a healthier, more diversified base than the bears assume. This is an underweighted part of the bull case.

10. Leadership Transition: Kirsten Spears to Amy Teiner

The call marked CFO Kirsten Spears's final earnings report before her June 12 retirement after 12 years, with incoming CFO Amy Teiner introduced. Hock thanked Spears for her leadership; the handoff appears orderly and long-planned.

Assessment: CFO transitions at AI bellwethers warrant attention, but this one reads as a planned, well-telegraphed succession rather than a disruption. Continuity of the financial discipline (flat opex, record FCF conversion, segment-level margin transparency) is the thing to monitor; the framework Spears institutionalized is the standard Teiner inherits.

Guidance & Outlook

The Q3 guide is, on its own terms, spectacular: $29.4B of revenue (+84% YoY) beats consensus (~$28.6B) by nearly a billion dollars, with operating margin held at a record ~67% and software accelerating to +31%. The blemish that drove the tape is narrow: the $16B AI semiconductor component sits below the ~$17.2B whisper, and the full-year AI framework (~$56B FY26, >$100B FY27) was reiterated rather than raised.

Implied H2 ramp: Per Hock's own math, H1 AI shipped ~$19B; 2x in H2 gets to ~$56B for the year. With Q3 AI at $16B, Q4 AI is implied around ~$21B — a sequential acceleration, not the deceleration some feared. For 2027, ~10 gigawatts of shipment is planned (unchanged), back-half loaded, which sets up "a lot more gigawatts" and "substantial growth" in 2028.

Street at: Consensus will move up on the total-revenue guide beat but trim the most aggressive AI-semi models toward the $16B Q3 figure. The debate is whether the FY27 >$100B framework proves conservative (our view) or whether the lack of a raise signals the trajectory is flattening (the bear read).

Guidance style: Hock guides to what he can contract and supply, not to the whisper. The $30B bookings and 2028 visibility argue the guide is a floor — Broadcom has historically shipped the bulk of its disclosed backlog ahead of the linear assumption.

Analyst Q&A Highlights

The FY26 AI Math and the Trajectory Into a $200B Backlog

The opening question probed the half-over-half AI math (does $56B imply a sequential Q4 decline?) and whether the broadening customer base and multi-gigawatt deals could push the 18-month backlog toward $200B. Management reconciled the numbers — H1 ~$19B, 2x gets to ~$56B — and reiterated that 2027 will "very easily" exceed $100B on the same trajectory, while declining to provide a precise backlog figure.

Q: "On this fiscal year, AI 2x growth second half over first half would put AI revenues over $60 billion, but you gave us $56 billion... And given the strength of all your programs... is it fair to assume your 18-month backlog sits at $200 billion or better?"
— Harlan Sur, JPMorgan

A: "2x the first half, we shipped about $19 billion in total AI revenue... 2x that, the second half, gets to pretty much in the range of $56 billion... 2027 will exceed, very easily, $100 billion... it is on the same trajectory as we are seeing in the back half of 2026."
— Hock Tan, President & CEO

Assessment: Management confirmed the framework is intact and the trajectory unbroken, but pointedly would not be drawn into a higher headline number. For a stock that wanted a raise, "very much on track, if not stronger" was not enough — but it is exactly what a disciplined operator with multi-year contracts should say.

The Google Long-Term Agreement: Durability and Share

A question pressed on the recently-disclosed Google long-term agreement amid concern about Broadcom's share within that account. Management characterized the commitment as "very, very substantial" in dollars while candidly acknowledging that Google's enormous and growing AI compute consumption means some diversity of supply is to be expected.

Q: "On the long-term agreement with Google — there is a lot of concern about share within that customer. Could you speak to your confidence, and is there upside? Is it a fixed amount or are there shares?"
— Blayne Curtis, Jefferies

A: "It is a very, very strong agreement... a commitment that is very substantial in dollars. Now, we also accept the fact that... given the rate of growth of consumption of AI compute by our partner Google, we fully expect that there will be some diversity of sources for them. But our commitment from them is a very substantial dollar amount."
— Hock Tan, President & CEO

Assessment: A refreshingly honest answer. Rather than claim 100% share, Hock acknowledged Google will multi-source while emphasizing the contracted dollar commitment. This is the right way to think about the anchor customer — the absolute dollars matter more than the share percentage, and the LTA locks in multi-generation revenue regardless of marginal share shifts.

Gross Margin Drivers Within Semiconductors

An analyst dug into the gross-margin decline, asking whether the drivers were XPU-versus-networking mix and whether the trend continues. Management reiterated that ASICs/TPUs and wireless carry lower margins while AI networking is margin-rich, and that the blended decline is mix-driven and offset by operating leverage.

Q: "On gross margin going down due to mix... behind the scenes, what are the drivers within semis? Is that the XPU versus networking side, and is that trend likely to continue next year?"
— Ross Seymore, Deutsche Bank

A: "Within semiconductors, our ASICs, TPUs, some of the wireless business have lower margins. So as the TPUs continue to accelerate, there will be pressure overall on margins. But the connectivity side, the AI networking side, has very rich margins, so it will offset it somewhat... structurally, the semiconductor margins remain very stable and very solid."
— Kirsten Spears, CFO / Hock Tan, CEO

Assessment: The exchange clarified the single most-misunderstood part of the Broadcom model. TPU/XPU growth is gross-margin-dilutive but operating-margin-accretive (via leverage and the rich-margin networking attach). Investors should track operating margin and segment margins, not blended gross margin, which will keep optically declining as AI scales.

2027 Gigawatt Targets — Any Change Versus Last Quarter?

A pointed question asked whether the 2027 gigawatt shipment target had changed from the ~10 GW indicated last quarter, and how the year shapes up. Management confirmed ~10 GW for 2027 is "very much intact," back-half loaded, which sets up a much larger 2028.

Q: "You gave gigawatt shipment targets for next year... I think you said close to 10 gigawatts in 2027. Is there any change — more gigawatts, less, or the same versus last quarter? And is it more back-half loaded?"
— Stacy Rasgon, Bernstein Research

A: "For 2027, we indicated about 10 gigawatts. That is still very much intact... back-half loaded, to that extent, yes. And which really provides an interesting trajectory into 2028. So 2028, we expect a lot more gigawatts."
— Hock Tan, President & CEO

Assessment: Unchanged is the operative word — and for a stock that wanted "more gigawatts," unchanged read as a disappointment. But ~10 GW back-half-loaded in 2027 setting up "a lot more" in 2028 is precisely the multi-year compounding the thesis requires. The shape (back-half-loaded) also means 2027 estimates have upside as the ramp steepens.

Networking as a Share of AI Revenue — Sustainable at 40%?

An analyst asked whether networking (~40% of AI revenue this quarter) would fall back as custom XPU programs ramp, and when optical/CPO revenue becomes meaningful. Management called 40% a "stars-aligned" peak and guided to networking settling closer to ~30% of AI revenue as XPU volume scales.

Q: "On the profile of your networking business — about 40% of AI revenue this quarter — would you expect that to fall back as custom ramps ramp up, or stay at the upper end? And when does optical/CPO revenue become meaningful?"
— James Schneider, Goldman Sachs

A: "This 40%, I consider almost a situation where stars are aligned... I see that as probably as high as that percentage of total AI revenue would go... I expect the percentage as a share of total AI revenue for networking would be closer to around 30%."
— Hock Tan, President & CEO

Assessment: The mix shift from ~40% to ~30% is a function of XPU dollars growing faster, not networking shrinking — absolute networking revenue keeps rising. The honest framing (40% was a peak) is the kind of disclosure that gets misread as caution; it is actually a statement that the larger XPU pie is growing even faster.

$30B of AI Bookings — Why So Much Backlog Now?

An analyst flagged the unusual magnitude of $30B in AI bookings relative to shipments and asked what is driving the surge. Management explained that customers are ordering far ahead to secure not just wafers, HBM, and DRAM but the power and infrastructure that gate deployment — pushing visibility to 2028.

Q: "You talked about $30 billion of AI bookings in the quarter, a lot relative to this and next quarter's shipments. Why is there so much backlog now?"
— Joe Moore, Morgan Stanley

A: "There is huge demand for compute... they realize lead time to get compute, you need to be thoughtful... they need to align quite a few other things — particularly power — before they can deliver. But they are placing orders early and in fairly huge demand, which gives us a lot more visibility... our visibility runs all the way to 2028 right now."
— Hock Tan, President & CEO

Assessment: The most important exchange on the call. A ~3x book-to-bill and a year of added visibility is a profound de-risking of the out-year estimates. The orders are constrained by power and infrastructure (not demand), which is exactly why Broadcom is building the financing platform to remove those bottlenecks. This is the signal; the $16B Q3 guide is the noise.

The Apollo Financing Platform and the Anthropic Structure

An analyst asked whether the Anthropic deal was "backstopped by Broadcom chips" and whether more such financing structures would follow. Hock corrected the premise — Broadcom supplies its own TPU chips to provide compute capacity to Anthropic — and tied it to the broader Apollo/Blackstone XPU platform that funds chips for capital-constrained frontier labs.

Q: "Your most recent deal with Anthropic is being backstopped by Broadcom chips. Do you think you will see more deals done this way? And any comments on the future financing of deals with the large AI models?"
— Thomas O'Malley, Barclays

A: "I have to correct you on that. The deal we did with Anthropic — we use our TPU chips that we develop to provide compute capacity to Anthropic. It was not a backstop... We were the ones providing the chips, providing the compute capacity. What we are doing [with the XPU platform] is, in partnership with guys with the best balance sheets, a vehicle to have these chips funded for these LLM players who otherwise might have difficulty getting access to our technology."
— Hock Tan, President & CEO

Assessment: The correction matters: Broadcom is the silicon supplier, with Apollo/Blackstone (not Broadcom) providing the capital. That structure pulls demand forward and widens the addressable customer set to capital-constrained labs, while keeping the financing risk off Broadcom's balance sheet. It is innovative supply-chain engineering — and the part of the story most worth monitoring for circularity if AI funding conditions tighten.

What They're NOT Saying

1. A raised FY27 number: Management reiterated >$100B for FY27 but pointedly refused to raise it despite $30B of bookings and visibility extending to 2028 — leaving the market to wonder whether the restraint is conservatism (our read) or a ceiling. The unwillingness to quantify the backlog ("$200B?") fed the same uncertainty.
2. Per-customer revenue concentration: Broadcom discloses gigawatts and program structures but not the revenue split across the six core customers. With Google the anchor and the two consumer-facing labs (OpenAI, Anthropic) carrying the largest out-year gigawatt commitments, the concentration profile is opaque by design.
3. The economics of the frontier-lab customers: The "insatiable demand" thesis rests on the capex of a handful of labs whose own profitability is unproven. Management framed token consumption as structurally compounding but did not engage the question of what happens if AI-lab funding or unit economics deteriorate.
4. Margin trajectory of the XPU mix at scale: Management confirmed TPUs/ASICs are lower-margin and will pressure blended GM, but declined to quantify where consolidated gross margin settles in FY27 as AI semis approach ~60%+ of revenue. "Model the segments separately" is correct guidance, but it sidesteps a hard number.
5. Any detail on the Apollo platform's terms: The $35B first tranche was announced, but the structure, Broadcom's economic participation, and the risk-sharing terms were left undisclosed — a material new mechanism described only at the headline level.

Market Reaction

• Pre-print setup: AVGO closed Wednesday June 3 at $479.23, near all-time highs, having run up sharply through the spring on the April wave of custom-silicon deal announcements (Google TPU LTA, Meta MTIA, OpenAI/Anthropic gigawatt agreements). The stock was widely described as "priced for perfection," with the buy-side AI-semi whisper for Q3 sitting near ~$17.2B versus the eventual $16B guide.
• Day-of move: Shares fell ~12–14%, trading down toward ~$410 — one of the largest single-day declines in the stock's recent history despite a record quarter. The drop dragged down AI-semi peers (AMD, Intel) on read-through.

The selloff is a textbook expectations reset rather than a fundamental repricing. Three things were sold: a Q3 AI guide ($16B) below the most aggressive whisper, the absence of a raise to the ~$56B FY26 / >$100B FY27 framework, and the simple fact that a stock discounting a beat-and-raise on every line got a beat-and-confirm instead. What the tape ignored is the substance: records across revenue, operating margin, EBITDA and FCF; AI bookings of >$30B at a ~3x book-to-bill; visibility extended a full year to 2028; the non-AI cyclical recovery; a re-accelerating VMware; and a new $35B financing platform. When a fundamentally strengthening quarter is sold purely because expectations outran it, the resulting pullback is an opportunity for investors whose horizon is the multi-year compute build-out rather than the next quarter's whisper.

Street Perspective

Debate: Is the ~13% Selloff a Gift or the Start of a De-Rating?

Bull view: The fundamentals strengthened (3x book-to-bill, 2028 visibility, record margins/FCF), and the stock fell purely on expectations. A high-quality compounder de-rating ~13% on a beat-and-confirm is a textbook add-on-weakness setup, especially with the multiple now more reasonable.

Bear view: A stock priced for a perpetual beat-and-raise just delivered the first quarter that merely met its existing framework. If the AI capex cycle is approaching its peak rate of change, the lack of a raise is the early tell, and the de-rating has further to run as out-year estimates get marked to a non-accelerating trajectory.

Our take: The bull case is stronger. A 3x book-to-bill and a full year of added visibility are not the signatures of a decelerating cycle — they are the opposite. The selloff reset positioning, not the thesis. We would use the weakness to add, while respecting that a momentum stock can stay out of favor until a clean beat-and-raise re-anchors sentiment.

Debate: Customer Concentration — Durable Anchor or Fragile Dependence?

Bull view: Six core customers with contracted, multi-generation, gigawatt-scale programs is a far more durable base than the "design win" model of two years ago. The breadth (four named hyperscalers/labs plus two ramping) and the multi-year LTAs (notably Google) reduce single-point risk.

Bear view: Broadcom's growth is levered to the capex of a handful of frontier labs whose own economics are unproven, and management openly concedes the anchor customer (Google) will multi-source. A demand base concentrated in a few capital-hungry AI labs is fragile if funding conditions tighten.

Our take: Concentration is the thesis's genuine vulnerability, but the contracted, gigawatt-scale, multi-year nature of the commitments — and the diversification across six customers and two end-markets (custom silicon + networking) — makes the base materially more resilient than the bear case allows over a 2–3 year horizon. Beyond that, the durability tracks the AI-lab economics, which remain the key external risk.

Debate: The Apollo Financing Platform — Visionary or Circular?

Bull view: The XPU platform removes the capital bottleneck for cash-constrained frontier labs, locks in Broadcom silicon at the lowest power/cost, and is funded by Apollo/Blackstone — not Broadcom's balance sheet. It is demand-pull engineering that widens the customer set and accelerates deployment.

Bear view: Vendor-adjacent financing of your own demand is the kind of structure that flatters growth on the way up and amplifies the unwind on the way down. If AI capex wobbles, financed demand can reverse faster than organic demand, and the platform's risk-sharing terms are undisclosed.

Our take: We lean bullish because the capital sits with Apollo/Blackstone rather than Broadcom, limiting direct balance-sheet risk, and because it solves a real bottleneck (lab access to capital). But it is a new and material mechanism described only at the headline level — we want disclosure on the terms and Broadcom's economic participation before fully crediting it, and we flag it as the single most important thing to watch in the model.

Model Update Needed

Valuation impact: The fundamental estimates move slightly higher (total-revenue guide beat, software acceleration, non-AI recovery, extended visibility), while the AI framework is unchanged. The ~13% price reset lowers the multiple on an unchanged-to-stronger earnings stream, improving the risk/reward. On a record ~46% FCF margin and a multi-year, contracted AI ramp with visibility now to 2028, the post-selloff valuation is more attractive than at any point this calendar year. We raise no rating change but view the pullback as an accumulation opportunity.

Thesis Scorecard Post-Earnings

Overall: Thesis confirmed and, on the forward indicators (bookings, visibility, customer commitments), strengthened. The only thing that "broke" was the stock's expectation premium, not the business. A new watch item — the Apollo financing platform — was added.

Action: Maintaining Outperform. The quarter validated the custom-silicon and AI-networking franchise, extended visibility to 2028, and produced record margins and cash flow; the ~13% selloff is a positioning reset that improves the entry. We would use the weakness to add. Watch items: a quantified backlog or FY27 raise at the next print, the terms and economic participation of the Apollo XPU platform, the consolidated gross-margin trajectory as AI semis mix up, and any change in frontier-lab capex conditions.