Bear Case for Broadcom: Alphabet Insourcing Threatens AI Revenue

In any great industrial transformation, the decisive advantage lies not in the finished product, but in the command of the productive assets that create it. The steel baron who owned the ore, the furnaces, and the rail lines dictated terms to the entire market. Today, the AI era’s foundational infrastructure—custom silicon accelerators, high‑bandwidth networking, and the vast data centers that house them—recapitulates that structure. The recent turbulence around Broadcom Inc.’s earnings is not merely a quarterly miss; it is a signal that Alphabet Inc., the marquee customer for Broadcom’s custom TPU silicon, is renegotiating the balance of power in its supply chain, much as a great steelmaker might eventually bring its ore mining and transport in‑house to capture margin and assure supply.

The Industrial Logic: From Proprietary Partnership to Multi‑Sourcing

Broadcom has been the primary design and integration partner for Alphabet’s custom Tensor Processing Units, a relationship solidified by a five‑year agreement extending through 2031 that covers development of the TPU v8 architecture ^{1,13,42,63,67,68,79}. This arrangement turned Broadcom into a critical supplier—its AI revenue surged 106% year‑over‑year in fiscal Q2, reaching $43 billion with a 140% growth rate ^{1,2,4,5,6,7,8,9,10,11,12,14,18,26,28,31,32,33,34,35,36,37,38,41,43,44,45,46,47,48,49,51,54,56,60,61,76,77}. Yet, like any wise industrialist controlling a strategic input, Alphabet is actively pursuing a multi‑sourcing and insourcing strategy. Broadcom’s own management acknowledged that “Google will multi‑source its technology needs” ⁷², and Macquarie explicitly downgraded Broadcom on the risk that Alphabet’s insourcing will meaningfully erode Broadcom’s share by 2027 ^70,71. The era of sole‑source dependency is ending; Alphabet is qualifying alternative ASIC vendors and investing in its own chip designs ^30,70,74. This is the classic hyperscaler move to commoditize its supply base, strengthening bargaining power and eventually internalizing more of the value stack.

The Earnings Flashpoint: Revenue Miss, Valuation Reckoning

The market’s abrupt repricing of Broadcom laid bare the concentration risk. Broadcom reported fiscal Q2 revenue of $22.19 billion, slightly below the $22.27 billion consensus, and adjusted EPS of $2.44, a modest beat ^{15,16,17,22,23,24,25,29,48}. The real fracture came in the forward outlook: Q3 AI chip revenue guidance of $16 billion missed the $17.2 billion consensus and a whisper number near $17 billion ^{21,31,41,46,47,48,49,61,65,77}. Management did not raise its full‑year 2026 forecast and merely reiterated its fiscal 2027 target of over $100 billion, without an upward revision ^{27,31,36,69,72,78}. The market interpreted this as softening demand visibility or supply constraints capping upside ³¹. The stock plummeted as much as 15% in a single session, losing over $300 billion in market capitalization and retreating 20–30% from its pre‑earnings peak ^{17,19,20,31,35,39,40,50,52,53,55,57,62,73,78}. The selloff was compounded by Alphabet’s multi‑sourcing remarks, fanning fears of a broader AI valuation de‑rating ^70,75. In Carnegie terms, this was the market realizing that the owner of the rail lines (Alphabet) was building its own locomotives, threatening the foremost locomotive builder’s future.

Alphabet’s Calculus: Command of the Value Chain

For Alphabet, the strategic logic is unmistakable. An announced $80 billion capital raise dedicated to data centers and chip development is a direct tailwind for suppliers like Broadcom in the near term, but it also signals Alphabet’s intention to increasingly self‑supply the most critical components ¹⁷. The dual‑track approach—maintaining the Broadcom partnership for near‑term continuity while accelerating internal design and alternative vendor qualification—aims to bring more of the AI infrastructure cost curve under Alphabet’s control. If successful, insourcing could improve Cloud margins and differentiate Google’s offerings through proprietary silicon, much as Amazon’s Graviton did. However, the complexity of cutting‑edge chip design, advanced packaging, and networking integration means that Broadcom’s role will not vanish overnight; the 2031 agreement and ongoing TPU v8 development confirm a deep, if evolving, technological interdependence ⁶⁷. The risk is that execution delays could slow Alphabet’s next‑generation infrastructure deployment, ceding ground to rivals.

The Broader AI Infrastructure Contest

The cluster illuminates forces beyond the chip supply chain. Broadcom’s research points to enterprise AI workloads shifting from public clouds to private cloud environments, driven by cost, compliance, and control ^58,59,64. VMware Cloud Foundation is positioned as the platform for this transition, which could capture spending that might otherwise flow to public cloud providers like Google Cloud ^59,69. This trend underscores that hybrid and private AI deployments represent a growing share of the addressable market—a dynamic Alphabet must counter with its own Anthos and hybrid cloud solutions. Moreover, Broadcom’s entrenched networking leadership (e.g., Tomahawk 6 and upcoming Tomahawk 7 switches) and its long‑term contracts with multiple frontier AI labs—including OpenAI’s 10‑gigawatt compute agreement through 2029 and an Anthropic engagement potentially worth $42 billion in revenue by 2027—show that hyperscaler demand for custom silicon and high‑bandwidth connectivity is durable, even as individual giants seek to internalize design ^{3,30,32,66,67,69,76,77}. Alphabet will both fuel and navigate this infrastructure buildout; its own capex decisions and architectural choices will have outsized influence across the ecosystem.

Strategic Implications and Prescriptions

Alphabet stands at a classic industrial crossroads. The multi‑sourcing pivot is a necessary step toward commanding the AI stack, but it must be executed with the discipline of capital that the steel magnates would recognize: avoid overreliance on any single supplier, but manage transition risk meticulously. The $80 billion capex program should be directed not only to raw capacity but to accelerating in‑house design capabilities and qualifying redundant suppliers, so that the shift away from Broadcom does not create a bottleneck. At the same time, Alphabet must reinforce its hybrid cloud and edge AI offerings to address the enterprise migration toward private cloud AI workloads; otherwise, Broadcom’s VMware‑led ecosystem could intercept demand that Google Cloud needs to capture. The ultimate prize is clear: whoever controls the accelerators, the networking fabric, and the software‑defined infrastructure will set the terms for the next decade of AI. Alphabet has the resources and the strategic imperative to become that master resource—provided it integrates as ruthlessly as it did when it first built the world’s information rails.