Sixth XPU Customer (OpenAI in Volume 2027), AI Semis +106% to $8.4B, Q2 FY26 Guide $22B with AI Semis +140%, Line of Sight to AI Chip Revenue >$100B in 2027 — Maintaining Outperform

Rating Action: Walking the Arc to Today

Maintaining Outperform. The Q1 FY2026 print confirms the Q4 thesis on every fundamental axis we specified, fires the explicit "sixth XPU customer" optionality we flagged at Q4 (OpenAI), and crystallizes FY27 from a framing-not-numbers backlog narrative into a hard ">$100B AI chip revenue" floor with fully-secured supply through 2028. The post-print share-price reaction of +4.8% reflects an audience that is processing the magnitude of the 2027 disclosure but has not yet capitalized it into the forward multiple. We hold the rating; we walk the arc as follows.

Updates and supersedes nothing — this is the first Q1 FY26 publication on AVGO from Aardvark Labs. No flashcap was published into this print.

Rating-evolution arc, walked:

• Q2 FY2025 (Initiating at Hold, constructive bias) — Operational execution exemplary; Hold rating reflected a fully-priced multiple, narrow XPU customer concentration (3 customers, 4 prospects), and a non-AI semis floor that was taking longer than expected to firm. Required a fourth XPU customer at revenue, a sustained +60% AI growth rate into FY26, OR a non-AI cyclical inflection to upgrade.
• Q3 FY2025 (Upgrading to Outperform) — Two of the three Q2 upgrade triggers fired in a single quarter: a fourth XPU customer with $10B+ in qualified production orders (Anthropic), and an explicit upward re-rate of the FY26 AI growth trajectory ("fairly material improvement" / "accelerate") supported by a record $110B consolidated backlog. Hock-through-2030 management continuity was an unexpected positive. Upgraded to Outperform.
• Q4 FY2025 (Maintaining Outperform) — The Q3 thesis confirmed and extended. The fifth XPU customer fired ($1B order for late-2026, the explicit "more aggressive Outperform" trigger from Q3). Q1 FY26 AI semis guide of $8.2B (+~100% YoY) materially above where consensus sat heading in. Anthropic added $11B incremental for late-2026 ($21B cumulative). $73B AI backlog inside $162B consolidated. Rack-mix GM compression underwritten and on-track. Maintained Outperform; flagged the OpenAI XPU program at "very advanced stage" as the next path-(b) catalyst.
• Q1 FY2026 (Today — Maintaining Outperform) — The path-(b) catalyst from Q4 fired. OpenAI is now formally disclosed as the sixth XPU customer, deploying first-generation XPU in volume in 2027 at >1 gigawatt — the "very advanced stage / very, very quickly / committed element" Q4 framing has crystallized into a hard 2027 dollar contribution. AI semis printed +106% above the +100% guide; Q2 FY26 guide of $22B (AI semis +140%) is the steepest single-quarter sequential step in franchise history. The ">$100B AI chip revenue in 2027" framing — a number Hock has never previously put into a forward window — is anchored by ~10 gigawatts of XPU shipments across six customers and supply secured through 2028 in T-glass, HBM, leading-edge wafers, and substrates. Networking growing faster than XPUs in absolute terms (+60% YoY, advancing toward 40% of AI revenue) is the second-order positive that compounds the bull case. The thesis-impairment scenarios (one of the named XPU customers slips, OpenAI relationship cools, GM walk worsens materially) did not surface; the GM walk in particular came in better than feared with Hock and Kirsten both pushing back hard on the rack-margin bear thesis. Maintaining Outperform.

What gets us to a wider Outperform conviction (or wider price target): (i) FY27 AI chip revenue prints clearly above the $100B floor, with ~$110–120B looking achievable on the disclosed gigawatt math; (ii) a seventh XPU customer surfaces — Hock has been deliberately conservative about the count ("just 6"), but the deal velocity has compounded each of the last four quarters; (iii) Tomahawk 7 (2x Tomahawk 6 performance, 2027 launch) extends the networking lead into 400G SerDes, sustaining the +33–40% AI-networking mix-share floor; (iv) the non-AI cyclical recovery actually arrives in fiscal H2 FY26 as the Q2 +4% YoY framing suggests an early thaw is starting; (v) capital allocation continues to compound on $32B+ FY26 FCF (we are walking up to $34–36B post-print).

What would put us back on Hold: (i) any one of the six XPU customers signals a 2027 capex pause, design slip, or pull-out — with the FY27 number now anchored on six customers' commitments, even one slipping would matter materially; (ii) the OpenAI 2027 ramp slips (e.g., the ">1 gigawatt" first-gen deployment delays into 2028); (iii) the supply-chain "secured through 2028" disclosure proves leakier than presented (T-glass, HBM, or 2nm wafer allocation tightens at TSMC); (iv) the system-rack GM trajectory, despite Hock's pushback, materializes as the bears expect (i.e., 200–500bps of consolidated GM compression as the rack mix scales); (v) export-control regime tightens materially against the inferred customer set; (vi) the +106% Q1 print proves the AI growth peak rather than a stepping stone — this is the highest-vigilance scenario on a stock priced for continued acceleration.

Results vs. Consensus

Q1 FY2026 cleared consensus on revenue, EPS, EBITDA, AI semis, and the Q2 guide. The magnitude of the AI-semis acceleration to +106% YoY (from +74% in Q4) is the single most relevant number; Q2 FY26 AI semis guide of $10.7B (+140% YoY) is the operative anchor for the FY26 trajectory; and the FY27 ">$100B AI chip revenue" framing is the structural layer that re-rates the entire FY26–FY27 thesis.

Quality of Beat

The Q1 beat is high-quality on the same three dimensions we flagged at Q4 — AI-driven, software-flat-but-bookings-strong, operating-leverage-visible — with one new feature: the Q2 FY26 guide is the steepest sequential step the franchise has ever guided. $22B at +47% YoY is not a "marginal acceleration off Q1." It is a $2.7B sequential step in total revenue and a $2.3B sequential step in AI semis (from $8.4B to $10.7B), implying a sequential AI-semis growth rate of +27% — in a single quarter. That is not a model curve; that is the Anthropic / Google Ironwood / Meta MTIA / OpenAI prep cycle layering visibly on the ship rate.

The cosmetic flag from Q4 (rack-mix GM compression) appears to have softened. Q1 GM came in at 77% — in line with the prior framing — and Q2 guide is "flat sequentially at 77%." Hock and Kirsten both pushed back hard on Timothy Arcuri's (UBS) framing that rack revenue would compress consolidated GM by ~500bps. Hock's response was direct: yields and cost have hit a model where AI products run "fairly consistent with the models we have in the rest of the semiconductor business." Kirsten layered: "on further study relative to even comments that I did make last quarter, the impact relative to our overall mix is actually not going to be substantial at all." This is the second consecutive quarter where the rack-GM walk has come in better than the bear case modeled it; we treat the Q4 "structural compression" framing as resolved more favorably than underwritten.

Operating margin of 66.4% (+50bps YoY) is the operating-leverage disclosure; semi operating margin of 60% (+260bps YoY) and software operating margin of 78% (+190bps YoY) are both expanding faster than revenue. EBITDA margin at 68% beat the 67% guide. This is what "scale continues to drive significant operating leverage" looks like in the financials: revenue growing faster than opex on every line.

Segment Performance

Semiconductor Solutions: AI Acceleration to +106%, Custom XPU +140%, Sixth Customer (OpenAI) Disclosed

Semiconductor revenue of $12.5B was up 52% YoY (an acceleration from Q4's +35% pace) and represented 65% of total revenue. Semi operating margin of 60% expanded 260bps YoY. Inside this:

The OpenAI disclosure is the structural headline. In Q4, Hock distinguished between two parallel OpenAI relationships: (a) the 10GW system-build alignment for FY27–FY29, and (b) a separate OpenAI XPU development program at "very advanced stage" with "a committed element." On this Q1 call, Hock formally collapsed (b) into the customer count: "We also now have a sixth customer. We expect OpenAI deploying in volume their first-generation XPU in 2027 at over 1 gigawatt of compute capacity." The path-(b) optionality we flagged at Q4 is now a hard 2027 dollar contribution. The 1GW first-gen 2027 anchor is also consistent with the question Joshua Buchalter (TD Cowen) raised on whether the 10GW total OpenAI envelope implies a sharp 2028 inflection — Hock effectively confirmed that read.

Anthropic ramp confirmed and accelerated. Hock disclosed that Anthropic is "off to a very good start in 2026 for 1 gigawatt of TPU compute" with FY27 demand "expected to surge in excess of 3 gigawatts of compute." That is the Q4 $21B cumulative bookings translating directly into deployment cadence: 1GW landing in 2026, >3GW in 2027. The 3x sequential ramp from 2026 to 2027 is the largest disclosed customer ramp in the franchise's history and validates the FY27 ">$100B" framing without needing the other five customers to contribute disproportionately.

Meta MTIA: "alive and well, shipping now." Hock's preamble explicitly addressed the analyst reports that had been circulating questioning whether Meta was de-emphasizing MTIA in favor of more GPU spend: "Meta's custom accelerator MTIA road map is alive and well. We're shipping now. And in fact, for the next-generation XPUs, we will scale to multiple gigawatts in '27 and beyond." "Multiple gigawatts" is the operational disclosure; Stacy Rasgon (Bernstein) inferred at least 2GW in his gigawatt arithmetic. With Anthropic at 3GW, Meta at multiple, OpenAI at 1GW, customers 4 and 5 (presumed Google + one other) at "more than double" their 2026 base, and the seventh program (the "fifth customer" disclosed at Q4) showing strong 2026 shipments, Hock's "close to 10 gigawatts" 2027 framing is reconcilable to the customer-by-customer math.

Networking +60% YoY and accelerating to 40% of AI revenue. AI networking represented one-third of total AI revenue in Q1 (i.e., ~$2.8B of the $8.4B, growing +60% YoY) and is projected to grow to 40% in Q2. The drivers are explicit: Tomahawk 6 at 100Tb/s — "the only one out there" — and 200G SerDes capturing demand from hyperscalers regardless of whether the deployed accelerator is XPU or GPU. The framing matters: networking is the rare AI-tailwind line that grows on both XPU AND GPU deployments, which is why the absolute $-growth rate is currently outpacing XPU growth. Hock added the forward bridge: Tomahawk 7 in 2027 at 2x Tomahawk 6 performance, sustaining the lead. This is the single line item that has moved most favorably vs. our Q4 model.

Scale-up: copper through 400G, the lowest-cost / lowest-power moat. Tom O'Malley (Barclays) pressed Hock on why the preamble emphasized direct-attached copper through 400G SerDes "in particular, especially as a leading pioneer in CPO." Hock's answer was the most pointed defense of the networking moat: "In scaling out, we're past that. We use optical. That's fine. But I'm talking about scaling up in a rack, in a cluster domain. You really want to use direct-attached copper as long as you can…you don't need to go run into some bright shiny objects called CPO, even as we are the lead in CPOs." The structural read: AVGO's SerDes franchise ($-content per scale-up domain) compounds for at least one more architectural generation before optical interconnect (CPO) becomes necessary at the rack level. This is FY27–FY28 visibility, not just FY26.

Non-AI semis: stable but starting to thaw at the edges. Q1 non-AI revenue of $4.1B was flat YoY (in line with guide), with enterprise networking, broadband, and server/storage all up YoY, offset by a seasonal wireless decline. Q2 guide of $4.1B is +4% YoY — the first positive YoY non-AI print since the cycle began. This is the first sliver of evidence for the cyclical recovery we have been waiting on through Q3 and Q4. We continue to model non-AI as a net-flat FY26 contributor, but the Q2 +4% guide is the leading indicator that fiscal H2 FY26 may begin to deliver. We do not need this to underwrite the rating, but it is an additive call option.

Infrastructure Software: 1% Q1 But $9.2B TCV Booked, VMware ARR +19%

Software revenue of $6.8B was up just 1% YoY in Q1 (the lowest growth print since the VMware integration), but the bookings momentum tells a different story: $9.2B of total contract value booked in Q1, sustaining VMware ARR growth of 19% year-on-year. Q2 guide of $7.2B (+9% YoY) is a clear re-acceleration. Software operating margin reached 78% (+190bps YoY); gross margin sustained at 93%.

The Q1 +1% headline is a reported-revenue quirk — reflecting the timing of large multi-year contract recognitions that booked in Q4 and Q1 but ratify into reported revenue over the contract life — not a deceleration in underlying business momentum. The right metric is the $9.2B TCV print, which on a Q1 basis is the strongest software-bookings quarter the franchise has reported. Hock devoted unusual prepared-remarks time to defending the software thesis against AI-disruption framing: "VMware Cloud Foundation, VCF, is the essential software layer in data centers integrating CPUs, GPUs, storage, and networking into a common high-performance private cloud environment…VCF cannot be disintermediated or replaced. It allows enterprises to scale complex generative AI workloads effectively…" The Phase 2 (VCF deployment + advanced services) thesis we laid out at Q3 is on-track; the Q2 +9% guide is the first reported-revenue confirmation.

Key Topics & Management Commentary

The ">$100B AI Chip Revenue in 2027" Framing

The single most important number in the call — arguably the most important number AVGO has ever forward-guided — is Hock's calendar-year 2027 framing: "Today, in fact, we have line of sight to achieve AI revenue from chips, just chips, in excess of $100 billion in 2027. We have also secured the supply chain required to achieve this."

The framing matters in three dimensions:

• "From chips, just chips." Hock was explicit in clarifying to Blayne Curtis (Jefferies) that the >$100B is silicon content — XPUs, switch chips (Tomahawk family), DSPs — and excludes any system-rack revenue. The total addressable AVGO 2027 revenue, including racks where AVGO sells the system rather than the chip set, is materially higher than $100B. We treat >$100B as the chip floor and any system-rack contribution as additive.
• Gigawatt math reconciliation. Stacy Rasgon (Bernstein) walked through the implied gigawatt count: ~3GW Anthropic + multiple Meta + 1GW OpenAI + ~3GW Google + customers 4–5 — arriving at 8–9GW. Hock confirmed that the 2027 envelope is "close to 10 gigawatts" and that the chip-dollars-per-gigawatt varies by customer "sometimes quite dramatically" but isn't far from Stacy's $20B-per-GW range. The math is internally consistent: 10GW × $20B/GW = $200B aggregate AI capex content, of which AVGO captures the silicon ($100B+) and a portion of the rest in racks. This is the first time the franchise has been quantified by gigawatt-deployment math rather than backlog snapshots.
• "Supply chain secured." Hock disclosed that supply for 2026 through 2028 is fully locked across the four constraint vectors: leading-edge wafers (TSMC 2nm/3nm), HBM, T-glass (the substrate technology that has been the industry bottleneck through 2024–2025), and standard substrates. Charlie Kawwas added that these are multi-year secured commitments, sometimes co-developed with capacity partners. We treat this as the second leg of the 2027 framing — a $100B chip revenue plan without secured upstream supply is a hope; with secured supply, it is a forecast.

Reading the framing carefully: $100B in 2027 chips, secured supply through 2028, a "yes" to growth in 2028 from Charlie when Ben Reitzes (Melius) pressed on it. The forward-looking statement is genuinely multi-year at the deployment-capacity level, not just the bookings level — a structural visibility upgrade we have not seen in the franchise before.

OpenAI: The Sixth Customer, Now Formally In

The OpenAI disclosure path through the past two quarters has been carefully sequenced. At Q4, Hock distinguished between (a) the 10GW system-build alignment for FY27–FY29 and (b) a separate XPU development program at "very advanced stage" that "will happen very, very quickly" with "a committed element." On this Q1 call, (b) crystallized into the customer count:

"We also now have a sixth customer. We expect OpenAI deploying in volume their first-generation XPU in 2027 at over 1 gigawatt of compute capacity."
— Hock Tan, CEO

Three structural reads:

• 2027 timing, not 2026. "First-generation XPU in 2027" means OpenAI is not in the FY26 number — consistent with the Q4 framing that the 10GW system program runs FY27–FY29. The OpenAI contribution is a 2027 step-up on top of the already-strong 2027 trajectory implied by Anthropic's 3GW and Meta's multiple-GW.
• 1GW is a starting point. Joshua Buchalter (TD Cowen) framed the inflection: 1GW in 2027 inside a 10GW envelope through 2029 implies a sharp 2028 acceleration, which Hock effectively confirmed by characterizing the 1GW 2027 deployment as a first generation. The follow-on generations — not the first — are where the bulk of the 10GW lands.
• "Committed element" language. Hock's Q4 hedging on whether OpenAI was a binding XPU commitment is now resolved: a sixth customer count requires a real, named, dollar-committed program. The Q4 ambiguity reflected disclosure pacing, not deal uncertainty.

For modeling: we add OpenAI as the marginal 2027 incremental layer on top of our prior FY27 trajectory, but treat the FY26 number as unchanged by OpenAI specifically. OpenAI is a 2027 catalyst, not a 2026 catalyst — which is why the FY26 numbers have to stand on their own (they do, comfortably) and the >$100B 2027 number is where OpenAI most matters.

Customer-Owned Tooling: The Recurring Bear, Hock's Most Pointed Push-Back to Date

Harlan Sur (JPMorgan) opened the COT line of questioning — the bear thesis that some hyperscalers / LLM developers will move toward in-house silicon design and shrink AVGO's XPU TAM. Hock has fielded this question every quarter since Q2 FY25; on this call, his push-back was the most pointed it has ever been. The substance:

"You need the best silicon design team around. You need cutting-edge, really cutting-edge SerDes, very advanced packaging, and most—and just as much, you need to understand how to network clusters of them together…in this particular space today in generative AI, if you're trying to, as an LLM player, do your own chip, you cannot afford to have a chip that is just good enough. You need the best chips that is around because you're competing against other LLM players. And most of all, you're also competing against NVIDIA, who is by no means letting down their guard…we are by far way out there. And we will not see competition in COT for many years to come."
— Hock Tan, CEO

The framing is a sharper version of the Q4 "overblown hypothesis" line — with one new element: the time-to-market argument. Hock layered: "Anybody can design a chip in a lab that works well. Can you produce 100,000 of those chips quickly at yields that you can afford? And we don't see too many players in the world that can do that." The COT bear thesis has historically been a chip-design argument; Hock is reframing it as a yield-and-volume-execution argument, which is harder for an in-house team at an LLM company to replicate. We treat the COT thesis as still live but as a multi-year-out concern, not a 2026–2028 concern; for the rating window, this is a non-issue.

Networking Differentiation: Why the Mix Is Climbing to 40%

Ross Seymore (Deutsche Bank) asked the cleanest two-part networking question: what's driving Q2 networking to 40% of AI revenue, and is the mix shifting structurally inside the >$100B 2027 envelope? Hock's response framed networking as a 33–40% range of total AI revenue going forward, with the current quarter's growth driven by Tomahawk 6 (100Tb/s, "the only one out there") and 1.6Tb/s DSPs. The structural reason networking grows faster than XPUs in absolute terms: networking content scales with cluster size and bandwidth, both of which are growing faster than chip count in modern AI deployments. Hyperscalers and LLM developers want larger cluster domains, which means more switches per XPU, more SerDes per switch, more DSPs per optical link.

The forward bridge: Tomahawk 7 launches in 2027 at 2x Tomahawk 6 performance — "by far the first out there" — sustaining the networking-content-per-cluster lead. Combined with the 200G-to-400G SerDes road map for scale-up copper, AVGO has a 24+ month networking-architectural-lead window into 2028. This is the single thesis line that has improved most favorably vs. our Q4 modeling.

Disaggregation, Specialized XPU Designs, and the GPU-XPU Mix

Christopher Muse (Cantor Fitzgerald) raised a sharp question on architectural evolution: as workloads disaggregate — pre-fill on one accelerator, decode on another, MoE on a third — does that pressure the GPU-vs.-XPU mix? Hock's answer was the most explicit defense of the XPU model on architectural grounds we have heard:

"You start to see designs of XPUs become much more customized for particular workloads of particular LLM customers of ours. And the design starts to depart from what is the traditional standard GPU design…XPUs will eventually be more the choice simply because it will allow flexibility in making designs that work with particular workloads, one for training even and one for inference…we're seeing that road map in all our 5 customers."
— Hock Tan, CEO

Hock then layered (in response to Jim Schneider, Goldman Sachs) that mature XPU customers are now developing two chips per year — one specialized for training, one specialized for inference — in parallel. This is the second-derivative bull argument: XPU TAM expands faster than gigawatt deployment as customers move from one chip per generation to two specialized chips per generation. We do not formally model the second-derivative effect but flag it as call optionality on the FY27–FY28 trajectory.

Scale-Up Ethernet: The Industry Standardization AVGO Anticipated

Charlie Kawwas's response to Tom O'Malley (Barclays) on scale-up protocols was the most consequential under-the-hood disclosure of the call. The substance: 24 months ago, the industry was unclear on which protocol would win for back-end (scale-out) networking in AI clusters. AVGO bet on Ethernet early; the industry standardized on Ethernet roughly 12 months later. The same architectural debate is now playing out for scale-up (intra-cluster, intra-rack); the early signals across hyperscalers and the semiconductor peer group are converging on Ethernet again. AVGO is positioned as the de facto Ethernet-everywhere AI silicon vendor, both for scale-out and scale-up.

The implication for the FY27 >$100B chip framing: a meaningful portion of that envelope is networking content tied to Ethernet at every layer of the cluster. If scale-up Ethernet standardizes as cleanly as scale-out did, AVGO's networking $-content per cluster compounds again in 2027–2028. We treat this as a positive but unquantified addition to the FY27 thesis.

Q2 FY2026 Guidance & Outlook

The Q2 FY26 guide is the most aggressive single-quarter guide AVGO has ever provided, and the disclosed FY27 framing is the most aggressive multi-year framing. Both layer as follows:

The 2027 chip-revenue floor. The framing isn't a Q2 number, but it is the structural anchor that re-rates everything downstream. ">$100B AI chip revenue in 2027" with secured supply chain through 2028 implies a 2027 chip-segment growth rate of, conservatively, +130%+ off our updated FY26 AI semis range of $40–43B. That growth rate is reconcilable with the disclosed 10GW deployment math; it is also the first time AVGO has anchored a forward year on chip-content × deployment-capacity rather than backlog-snapshot mechanics.

The tax-rate step-up. The 16.5% non-GAAP tax-rate guide for Q2 FY26 (vs. the lower FY25 effective rate) is the single modeling watch-out. The driver is the global minimum tax regime layered on top of geographic mix-of-income. We model the full FY26 effective rate in the 16–17% range, which carves ~$0.30–0.40 off our prior FY26 non-GAAP EPS bridge from FY25. That puts our updated FY26 EPS range at ~$10.50–11.50, vs. the Q4 walk to $10.00–11.50; the AI-revenue acceleration more than offsets the tax step-up at the midpoint.

Analyst Q&A Highlights

The Q&A was concentrated on three threads: gigawatt math reconciliation for the >$100B 2027 framing, the COT bear thesis, and rack-mix gross-margin walk. The transcript reads as the tightest AVGO Q&A of the past four quarters — analysts came prepared with specific numerical pressure tests and Hock answered most of them in concrete terms.

• Blayne Curtis (Jefferies) opened with two questions: (i) clarification on whether the >$100B 2027 number includes networking and rack revenue, and (ii) the broader bear concern that AI semis growing 2x cloud capex means the hyperscalers must monetize within 1–2 years for the spending to be sustainable. Hock confirmed the >$100B is chip-only and addressed the monetization concern by reframing the customer set: only six players, all with strategic XPU programs, all of whom see inference (productization, not just training) driving a substantial and growing component of compute capacity. The "ROI must arrive in 2026" framing assumes a 2024-vintage GPU-deployment economics; the 2027 deployments are mostly inference, where the unit economics are demonstrably better.
• Harlan Sur (JPMorgan) pressed on the COT (customer-owned tooling) bear — whether hyperscalers' in-house silicon teams can take meaningful XPU share. Hock's answer (quoted in detail above) was the most pointed COT push-back of his tenure, layering the time-to-market and yield-execution argument on top of the design-team argument. We came away more confident, not less, that COT is a multi-year-out concern.
• Ross Seymore (Deutsche Bank) went deep on networking differentiation and asked whether the 33–40% AI-networking mix is structural inside the >$100B 2027 envelope. Hock confirmed the range and added the Tomahawk 7 / 400G SerDes road map — networking compounds for at least one more architectural cycle.
• Timothy Arcuri (UBS) challenged the GM walk, framing rack-margin compression at "500 basis points roughly" and asking for a floor. Hock pushed back hard ("you must be a bit hallucinating"); Kirsten reinforced ("the impact relative to our overall mix is actually not going to be substantial at all"). Two consecutive quarters of rack-GM walk coming in better than feared resolves the Q4 "structural compression" concern.
• Stacy Rasgon (Bernstein) walked the gigawatt arithmetic for 2027: 3GW Anthropic + 1GW OpenAI + multi-GW Meta + Google + customers 4–5 — ~8–9GW. Hock confirmed "close to 10 gigawatts" and that chip-$/GW vary "sometimes quite dramatically" but aren't far from Stacy's $20B/GW math. First time the franchise has been internally reconciled on gigawatt-deployment math.
• Ben Reitzes (Melius) asked how AVGO secured 2028 visibility on a four-component supply chain when peers have not, and whether AVGO can grow in 2028 vs. 2027. Hock credited early action (locking T-glass, substrates, HBM, and 2nm wafer commitments ahead of peers); Charlie added multi-year capacity-partner commitments. Both answered "yes" to growth in 2028.
• Vivek Arya (Bank of America) asked how AVGO maintains visibility across six customers in a fragmented world. Hock reframed the customer set as strategic, not transactional — each treats their XPU program as long-term, multi-year, with capacity commitments. The visibility is structural, not market-share-fight contingent.
• Tom O'Malley (Barclays), Christopher Muse (Cantor), Jim Schneider (Goldman), and Joshua Buchalter (TD Cowen) rounded out the Q&A on scale-up Ethernet / direct-attached copper through 400G, workload disaggregation driving specialized XPU designs, the inference-vs.-training dual-chip cadence, and the OpenAI 2027-to-2028 inflection. Each answer (substantively quoted in the Key Topics section above) reinforced the structural visibility upgrade.

What They're NOT Saying

Three meaningful absences from this call — each smaller than the equivalent absences at Q4, but worth flagging:

• No updated consolidated backlog number. Q4 disclosed $73B AI backlog inside $162B consolidated. This call did not refresh the snapshot. Reading: bookings velocity remains strong (the >$100B 2027 chip framing required it), but Hock chose not to update the backlog metric. We read the omission as deliberate — once you anchor on a multi-year deployment capacity (gigawatts × $/GW × secured supply), the rolling-backlog snapshot becomes a less informative metric. We do not interpret the omission as a backlog softness signal.
• No specific FY26 revenue framing. Hock walked through Q1 actuals and Q2 guide with concrete dollar numbers, then pivoted directly to 2027. There was no "FY26 will be approximately X" framing of the kind that some analysts pushed for. This is consistent with the AVGO disclosure convention (one quarter ahead, not a full year), but it leaves the FY26 number as a market-derived bridge: $19.3B + $22.0B + Q3 + Q4. We model FY26 at ~$87–90B (vs. our Q4 walk to $83–88B), with the upward revision driven by the Q2 +$22B floor and the AI-semis re-acceleration.
• No price disclosure on the racks-vs.-chips Anthropic mix. Vivek Arya (Bank of America) tried twice to extract from Hock how much of the Anthropic 1GW 2026 deployment is racks vs. chips. Hock declined: "I'd rather not answer that, but we're okay. As Kirsten said, we're good on our dollars and margin." The answer is essentially "trust us on the GM walk, the mix is what we said." Given the GM walk has now come in better than feared two quarters running, we let this one go — but we flag it as the one piece of operational opacity that remains.

Market Reaction

Three reads on the +4.8% reaction:

• The $100B framing is doing the heavy lifting. A pure beat-and-raise on Q1 results + Q2 guide would have moved the stock perhaps 2–3%. The incremental 2–3% that arrived in the after-hours / next-day price action reflects the >$100B 2027 chip number plus OpenAI plus secured-supply-through-2028 — a structural visibility upgrade that didn't fit cleanly into a Q1 beat-and-raise frame.
• Positioning is cleaner than at Q3 or Q4. The Q4 sell-off (which we flagged at the time as positioning unwind, not thesis impairment) cleared a meaningful share of the consensus-AI-long crowding. Q1's reaction is into a less-stretched positioning setup, which is a structural positive for forward returns even as the share price advance is more measured.
• The forward multiple has not yet capitalized the FY27 floor. On our updated FY26 EPS midpoint of ~$11, the post-print P/E is in the high 20s — meaningfully below the multiple at the Q3 print and below where we would underwrite a franchise with secured-supply 2028 visibility, six XPU customers, +$100B FY27 chip revenue floor, and $32–36B FY26 FCF. The structural multiple expansion path remains live.

Street Perspective

Sell-side commentary into and out of the print (paraphrased; direct quotes from external analysts are not provided in this report) clustered along three predictable axes:

• The bull case being made on the Street centers on the >$100B FY27 chip number as the new anchor for an EPS bridge meaningfully above existing consensus, and on the OpenAI confirmation as the long-awaited path-(b) catalyst. The bull-case framing is that with secured supply through 2028 and six XPU customers in the visibility window, AVGO has effectively taken the FY27 dispersion out of consensus and re-priced the stock as a multi-year-visible AI semis compounder rather than a backlog-snapshot story. Several sell-side teams walked up FY27 EPS estimates by 15–25% post-print.
• The bear case being made on the Street focuses on three pressure points: (i) the magnitude of the +106% Q1 / +140% Q2 AI-semis acceleration is unsustainable by 2027 and the path from Q4 FY26 to FY27 will look more like deceleration; (ii) the rack-mix GM walk is being pushed back on by management but the structural Q3–Q4 FY26 mix is not yet visible — the bear concern is that the GM compression simply hasn't shown up yet; (iii) customer concentration in the FY27 envelope is now meaningful (Anthropic 3GW, Google 3GW, the rest) and any single-customer slip would dent FY27 by 10–15%. We think the first concern is overblown given the +130% FY27 chip growth rate is reconcilable to the gigawatt math, the second is contradicted by two consecutive quarters of GM walk coming in better than feared, and the third is the legitimate watch-item but not yet a catalyst for downgrade.
• The neutral case pivots on whether the multiple has room to expand from its current level given the FY26 EPS bridge has now stepped up materially. We are constructive on this question — the post-print multiple is below where the franchise's structural visibility (secured supply through 2028, six customers, >$100B FY27 chip floor) would justify, and the FY26 EPS bridge has compounded faster than the share price.

Model Implications

Our prior FY26 framing (set at Q4) was: AI semis $36–40B (+80–100% YoY), non-AI $17B flat, software $30B (+11%), consolidated $83–88B (+30–38%), non-GAAP EPS $10.00–11.50. The Q1 print + Q2 guide require the following walks:

• FY26 AI semis. Walked up to $40–43B (+95–110% YoY) from prior $36–40B. The Q2 +140% guide is the explicit anchor; combining the Q1 +106% / Q2 +140% bridge with 1GW Anthropic 2026 + Meta MTIA shipping + customers 4–5 doubling implies $40B+ at the conservative end. Upside if Q3/Q4 FY26 sustain the Q2 +140% pace.
• FY26 non-AI semis. Walked up modestly to ~$17B reflecting the Q2 +4% guide thaw signal. FY26 software maintained at ~$30B (+11%); Q1 +1% reported masked $9.2B of TCV bookings. FY26 consolidated revenue walked up to ~$87–90B (+36–41% YoY) from prior $83–88B.
• FY26 op margin / EPS. Operating margin maintained at ~67%; non-GAAP EPS walked up to ~$10.50–11.50 from prior $10.00–11.50, net of the 16.5% tax-rate step-up. At a midpoint $11.00, the post-print forward P/E sits in the high-20s — below where we expect the franchise's structural visibility to settle.
• FY27 explicit framing. AI chip revenue >$100B as the new anchor, with secured supply through 2028. We model FY27 consolidated revenue at ~$135–150B with FY27 non-GAAP EPS bridge to ~$16–19. This is a structural EPS-floor disclosure that materially compresses the forward multiple at current price levels.
• Capital structure and FCF. $14.2B cash at quarter-end; $10B fresh buyback authorization through CY2026 layered on top of the $10.9B returned in Q1. FY26 FCF walked up to $32–36B. The OpenAI 10GW FY27–FY29 system program and the dual-chip-per-customer cadence are unmodeled additive call optionality.

Valuation framing. AVGO at the post-print level (~$332 close on March 5, +4.8% on the day) trades at a forward P/E in the high-20s on our updated FY26 EPS midpoint of $11.00, and in the high-teens on our FY27 EPS midpoint of $17.50. For a franchise printing 36–41% revenue growth into FY26 with an FY27 chip-revenue floor of $100B+, secured supply through 2028, six XPU customers, AI networking compounding alongside XPUs, and $32–36B of FY26 FCF, the multiple has not fully capitalized the structural visibility upgrade that this print delivered. We continue to underwrite the same fundamental story we underwrote at Q3, but with materially better forward visibility and a multiple that has not kept pace with the EPS walk.

Asymmetric scenarios that could change our rating again: (i) the FY27 AI chip revenue trajectory prints clearly above the $100B floor (e.g., $115B+) → supportive of widening Outperform conviction and a higher target multiple; (ii) one of the six XPU customers signals a 2027 capex pause, design slip, or pull-out → downgrade pressure; (iii) the FY26 AI growth rate sequentially decelerates materially below the Q2 +140% trajectory in Q3/Q4 prints → downgrade pressure; (iv) the system-rack GM walk through fiscal H2 FY26 actually compresses consolidated GM by 200–500bps despite Hock's pushback → modest downgrade pressure; (v) export-control regime tightens materially against the inferred customer set → bounded downside; (vi) supply-chain "secured through 2028" framing proves leakier than presented at TSMC 2nm or in advanced packaging → FY27 visibility impairs.

Thesis Scorecard Post-Earnings

Overall: Of the seven bull points (five from Q2 initiation + one from Q3 upgrade + one new from Q1 FY26), seven are now confirmed, with three materially upgraded again since Q4. Of the seven bear points (five carried forward + two new this quarter), one has improved further (concentration, now 6 customers), one has shown the first thaw signal (non-AI), one is better than feared two quarters running (rack GM), one is pushed back on most pointedly to date (COT), one is unbounded (export controls), and two are new this-quarter watch-items (FY26 acceleration sustainability and FY27 customer concentration). Four new fundamental positives (Hock through 2030 from Q3, system-rack TAM expansion from Q4, supply secured through 2028 from Q1, scale-up Ethernet standardization from Q1) compound. The composition is even more bull-skewed than at Q4.

Action: Maintaining Outperform. The Q1 print fires the explicit Q4 path-(b) catalyst (sixth XPU customer, OpenAI), confirms and extends the FY26 AI re-rate trajectory, crystallizes FY27 visibility into a hard ">$100B chip revenue" floor with secured supply through 2028, and resolves the rack-mix GM concern more favorably than underwritten. The post-print share-price reaction of +4.8% is measured relative to the magnitude of the structural disclosures and reflects an audience still processing the multi-year visibility upgrade rather than capitalizing it into the multiple. The fundamental case has compounded on every line; the asymmetry remains favorable. We will revisit on the Q2 FY26 print scheduled for after the close on Wednesday, June 3, 2026.