Broadcom's $164 Billion Order Book Is the Real Story the June Selloff Buried

Strip away the ticker and Broadcom is two businesses stapled together by cash flow. The larger one, Semiconductor Solutions, designs the custom AI accelerators — Broadcom calls them XPUs — that hyperscalers use instead of Nvidia GPUs, plus the Ethernet switches and optical interconnect that wire those chips into clusters. The smaller one, Infrastructure Software, is mostly VMware, the virtualization platform Broadcom bought in 2023 and has been re-pricing ever since. In fiscal 2025 the two split the company almost evenly — Semiconductor Solutions at $36.86B and Infrastructure Software at $27.03B of $63.89B total — but that snapshot is already stale, because the semiconductor half is now growing at a rate that will make the software half a rounding error inside three years.

What is at stake is whether the market is reading Broadcom's economics correctly. On June 3, 2026, the company reported a fiscal Q2 that beat on both lines, raised guidance, and disclosed a record order book. The stock fell roughly 13% the next session. The proximate cause was a single sentence from the CFO: as the fastest-growing product line scales, consolidated gross margin steps down to about 74% in Q3. My argument is that this was a mix optic misread as erosion, and that the tape briefly punished Broadcom for the arithmetic of winning the custom-silicon war. Underneath sits a contracted revenue backlog of a scale the semiconductor industry has essentially never carried, converting through a fabless model that turns almost half of revenue into free cash flow. The debate is not whether the AI number is big; everyone agrees it is. The debate is whether it is durable and cancellable — and that is where the primary filings, not the tape, do the work.

Why the ~74% Gross Margin Is the Sound of Winning, Not Cracking

The number that broke the stock deserves to be taken apart. In Q2 FY26 (ended May 3, 2026), consolidated gross margin was 77.1%, down 230 basis points year-over-year, and management guided Q3 to approximately 74%. CFO Kirsten Spears was explicit that this "does not represent a structural change in semiconductor margin" but "reflects product mix between semiconductors and infrastructure software" . The mechanism is simple and benign: software carries a 93% gross margin and 79% operating margin , while the custom AI accelerators diluting the blend carry lower gross margins but scale to enormous volume. When your fastest-growing line is your lower-margin line, the consolidated ratio drifts down by construction — even as every dollar of that growth is wildly accretive to operating income.

The honest line to watch is not gross margin but operating margin, and there the picture inverts. Q2 operating income was a record $14.9B, up 52% year-over-year, on consolidated opex of just $2.2B, of which $1.6B was R&D . Management guided Q3 operating margin to approximately 67% — flat quarter-over-quarter despite the gross-margin step . The semiconductor segment itself ran a 62% operating margin in Q2, up 460 basis points year-over-year . A business whose gross margin is "compressing" while its operating margin expands is not deteriorating; it is demonstrating that opex leverage is swamping mix. The market spent a day trading the wrong line of the income statement.

The Backlog That Has No Precedent in Semiconductors

Here is the fact I think the June selloff most clearly buried. As of May 3, 2026, Broadcom's remaining performance obligations under firmly committed multi-year contracts — including arrangements where customers have no termination rights — stood at approximately $164.6 billion, a figure that includes obligations under a long-term custom-AI-accelerator contract signed in the quarter itself . Against a company doing roughly $105B of revenue in FY26, that is more than a full year of forward, contracted demand sitting on the balance sheet as a disclosed obligation. On the flow side, AI semiconductor bookings in Q2 exceeded $30 billion against $10.8 billion shipped — coverage of nearly three times what went out the door.

That backlog is why management can guide with unusual specificity. Q3 AI semiconductor revenue is guided to $16 billion, up over 200% year-over-year, inside consolidated revenue of $29.4 billion, up 84% . Full-year FY26 AI is targeted at $56 billion, up roughly 180% , and Hock Tan reiterated FY27 AI "in excess of $100 billion," framed as roughly 2x the 2026 number, anchored on a plan to ship 10 gigawatts of AI compute in 2027 . The distinction that matters is between an order book that looks full at the top of a cycle and one that is contracted to a multi-year capacity roadmap. Gigawatt commitments from frontier labs building datacenters to a schedule are a different animal from a distributor's cancellable purchase order — and the $164.6B figure is management's disclosure that a large slice of the book carries no termination right at all.

The Cleanest Cash Machine in the AI Trade

The reason to value Broadcom on free cash flow rather than the EV/EBITDA multiple the screens quote is that the standard heavy-reinvestor distortion does not apply here. Broadcom is fabless. Q2 free cash flow was a record $10.3 billion — 46% of revenue — on capital expenditure of just $231 million . Across the first half of FY26, operating cash flow was $18.75B against PP&E purchases of $481M , so capex runs well under 2% of sales. The AI reinvestment happens through R&D (which sits in opex) and inventory to secure wafer supply, not through fabs. That makes free cash flow a clean read on earning power, because there is almost no growth capex to strip out — the opposite of a hyperscaler whose reported FCF is depressed by the very buildout that drives the thesis.

The incremental economics are the whole case. AI revenue carries roughly 62% segment operating margin at scale against a few hundred million of capex and some working capital. On that spread, the marginal return on invested capital sits far above any reasonable cost of capital near 10% — which is why, on a heavy reinvestor, the magnitude of the spend is a non-issue and the only question that matters is whether the revenue is real. The $164.6B contracted backlog is Broadcom's answer to that question. Meanwhile the balance sheet has de-risked as the cash has compounded: cash rose to $19.6 billion at quarter-end from $14.2 billion the prior quarter , against total debt principal of $66.7 billion — net leverage under 1x EBITDA, and falling.

The Rest of the Business, Placed

The non-AI parts are not the thesis, but a careful reader wants them accounted for. Within Semiconductor Solutions, the non-AI lines — broadband, wireless (the roughly $5B/year Apple relationship), enterprise storage, industrial — are in a shallow cyclical recovery, growing only about 6% year-over-year off a trough; they are ballast, not a growth engine, and their recovery is embedded in the flat-margin base case rather than driving it. Infrastructure Software delivered $7.2 billion in Q2, up 9%, with ARR growth of 17%, and management guided Q3 to approximately $8.9 billion — a striking reacceleration to 31% growth . The mechanism is real: VMware Cloud Foundation 9.1 now supports heterogeneous compute across AMD, Intel and Nvidia platforms, so as hyperscalers and enterprises deploy CPUs alongside GPUs, VMware attach rides along . It is the anti-cyclical, 79%-operating-margin counterweight to the lower-margin silicon — the segment that keeps consolidated operating margin defended even as gross margin drifts.

The Concentration That Is the Bear Case, in Three Costumes

The counter-argument I take most seriously is not valuation — it is that the entire incremental AI number rests on a handful of customers, and the filings quantify exactly how fragile that is. Direct sales to a single semiconductor distributor accounted for 42% of net revenue in the two quarters ended May 3, 2026, up sharply from 29% a year earlier, and Broadcom's top five end customers accounted for roughly 45% of revenue . The 10-K-level risk language is blunt: the loss of, or substantial reduction in sales to, any top customer for its custom AI accelerators "could have a material adverse effect" . Layered on top is manufacturing concentration — approximately 95% of Broadcom's wafers in the period were produced by TSMC — so a Taiwan disruption is a first-order tail risk, not a footnote.

Then there is the quality of the backlog itself, which the bulls quoting "$30B of bookings" tend to skip: as of May 3, 2026, approximately 64% of contract liabilities related to contracts subject to termination-for-convenience provisions . The $164.6B of firmly committed RPO is genuinely sturdy, but a meaningful chunk of the nearer-term deferred balance can be walked — which is the channel through which a hyperscaler capex blink would transmit fast. And a newer, subtler exposure surfaced in the same filing: on June 8, 2026, Broadcom arranged for an investor partner to take on agreements to purchase AI racks and the related customer lease obligations, backstopping those obligations over five-year terms with a maximum exposure of $29 billion . That backstop is not operational today — it scales up only as racks deploy and down as the customer pays — but it is a contingent liability that grows precisely with the AI ramp the thesis depends on, and it deserves to be watched, not waved away.

The last thread is VMware in Europe. Five cloud and CIO trade associations, led by CISPE, urged the European Commission on July 10, 2026 to impose interim measures against Broadcom's VMware licensing changes and to enforce a three-year transition period while the antitrust probe continues . Broadcom disputes the allegations, characterizing CISPE as hyperscaler-funded . This is a genuine overhang on the software reacceleration — but it is a tail on the smaller, slower-growing segment, not on the silicon engine that drives the valuation.

I discount the concentration case not because it is wrong but because it is priced and, more importantly, because its worst versions are multi-year, not next-quarter. The market's first taste of the fragility was Google reportedly diversifying part of its TPU program to a second supplier — a real dilution of Broadcom's leverage at its flagship account, and Tan himself has conceded to expect "some diversity of sources." But that is a 3-to-5-year erosion of a near-monopoly design-win position, not a 2027 revenue event, and it is exactly the kind of thing a rich multiple should — and I think partly did — begin to reprice in June. What the concentration case cannot easily explain away is $164.6B of firmly committed, largely non-cancellable obligations. That number is the difference between "this looks like a cycle top" and "this is contracted."

What the Stock Already Pays For

At $389.28, Broadcom carries a market capitalization of about $1.85 trillion . Netting the $66.7B of debt against $19.6B of cash puts enterprise value near $1.90 trillion, which against Tenzing's FY27E free cash flow of $73.3B is roughly 26x EV/FCF — and against consensus FY27E EPS near $19.55 (the Financial Modeling Prep analyst-estimates feed, 27 analysts) or Tenzing's more conservative $17.93, the stock trades at about 20–22x forward earnings. Neither is a cyclical-trough multiple, and neither is a bubble multiple for a business compounding free cash flow above 50%. For calibration, Nvidia — the growthier merchant-silicon comparable — trades materially richer on the same FCF basis, and mature semiconductors sit at 12–18x; Broadcom's ~26x sits between them, which is where a contracted-visibility AI compounder with a 79%-margin software annuity attached belongs.

The honest conclusion is that the multiple already discounts the AI double. On Tenzing's build — FY27 revenue of $166.5B, held below the Street's $174B, on a deliberately flat 64.5% operating margin that assumes no scale expansion — a base case holding today's ~26x EV/FCF lands close to fair, near $379, because the twelve-month return has to come from FCF actually compounding into the number rather than from any re-rating. The sell-side's ~$508 median target effectively underwrites both the full AI trajectory and a sustained premium multiple; the flat-margin assumption is the wedge between that view and mine, and every gross-margin print will re-underwrite it. The asymmetry is genuinely two-sided: if hyperscaler capex retrenches and FY27 AI lands nearer $90B than $112B, the contracted-visibility premium evaporates and the stock has meaningful downside toward the low-$200s; if the gigawatt commitments convert and networking mix holds margins richer, the durability earns a re-rate toward $500.

The verdict: the market spent June trading Broadcom's gross-margin line when the load-bearing disclosure was the $164.6 billion of firmly committed backlog and the 46%-of-revenue free cash flow that converts it. That mispricing has largely closed as the stock recovered from the high-$370s, so this is no longer a cheap stock — but it is a correctly-cheap-enough one. At roughly 26x forward free cash flow for a contracted AI franchise with a fabless cost structure and a software annuity behind it, Broadcom is worth owning for the compounding, with the clear-eyed understanding that six customers and 95% wafer concentration make the bear case real and the Q3 print — which management plans for after the close on September 2, 2026 — the next place the debate gets settled.

Disclaimer: Tenzing Analytics, Inc. is not a registered investment adviser, broker-dealer, or financial institution, and all content provided by Tenzing Memo is AI-generated for informational purposes only. It does not constitute investment advice, a recommendation, or an offer or solicitation to buy, sell, or hold any security or financial instrument. Content is general in nature, not tailored to any individual's financial situation or objectives, and may contain errors, omissions, or inaccuracies; it should not be relied upon as a primary basis for any investment decision. Any forward-looking statements are subject to risks and uncertainties and may not reflect actual future results. Users should conduct their own research and consult qualified financial, legal, or tax advisors before making investment decisions, and by accessing this content, you acknowledge and accept these limitations.