Broadcom's AI Revenue Guidance Soars While Stock Tumbles: The Expectation Gap Widens

Broadcom guided to $230 billion in AI revenue by 2028—four times this year's level—yet the stock fell sharply after Q3 earnings as investors demanded proof of near-term execution, while Cathie Wood's $72.8 million AMD exit signals skepticism about challenger chips' near-term competitive position.

What changed

Broadcom's CEO Hock Tan provided aggressive forward guidance in early September 2026, projecting $230 billion in AI revenue by fiscal 2028, representing a fourfold increase from the current year's run rate. The company also guided to more than $100 billion in AI revenue for fiscal 2027. Despite this bullish long-term commentary, Broadcom's stock fell sharply following the Q3 2026 earnings release: the shares dropped 5.8% in the morning session after the company reported results that fell below Wall Street's expectations, with Q4 sales guidance coming in soft relative to consensus estimates. Additional after-hours declines of 4% to 6% were reported as investors digested the earnings call.

Concurrently, Cathie Wood's ARK Invest executed a notable portfolio rebalancing on August 28, 2026, selling 156,286 shares of AMD (worth approximately $72.8 million) while purchasing 243,707 shares of Nvidia (worth approximately $53 million). This transaction was followed by an additional $18 million AMD sale in late August, signaling a strategic shift away from the challenger chip narrative toward Nvidia. Wood also sold approximately $74 million of AMD stock after the company's "explosive earnings," according to separate reporting.

On the AMD side, the company reported strong data center sales growth in its latest earnings, yet the stock declined 5.2% since the earnings report, suggesting that even positive results are not translating into sustained investor enthusiasm for the challenger thesis.

Why it matters

AMD's 5.2% post-earnings decline, despite strong data center sales, mirrors Broadcom's experience: operational success is being discounted by a market that is reassessing the challenger thesis's viability. This suggests that the market is pricing in either slower-than-expected hyperscaler adoption of custom silicon, or a realization that Nvidia's diversification into Vera CPUs and financing arrangements (including the reported $500 billion Wall Street financing pact and $105 billion support for an OpenAI data center) is creating a more durable competitive moat than previously assumed.

Opposing sources and risks

Multiple sources present material headwinds to the thesis:

Nvidia's strengthening competitive position. Nvidia reported revenue of $96.2 billion in its latest quarter, up 106% year over year, with management guiding to roughly 70% revenue growth for fiscal 2028—a figure the company attributes to supply constraints rather than demand weakness. This sustained growth trajectory, combined with reports that Nvidia is in talks to acquire Hugging Face for over $13 billion and has secured $500 billion in Wall Street financing, suggests that Nvidia is not losing share to challengers but rather deepening its control of the entire AI infrastructure stack (chips, software, financing, and model infrastructure). Wolfe Research named Nvidia its top AI pick, and analyst commentary suggests Nvidia's valuation is justified by its growth trajectory relative to Broadcom.

Hyperscaler design-aways and competitive tightening. Google has deepened its partnership with Marvell for AI chip development and is reportedly working with AMD on its new TPU, signaling that hyperscalers are diversifying away from Broadcom specifically. Broadcom's largest customer (widely understood to be Google) has reportedly dropped a key custom chip, raising design-away risk. While Broadcom's historical precedent suggests a 500% gain followed such scares, the current competitive environment is materially different: Nvidia is no longer the only alternative, and hyperscalers now have multiple custom silicon options (Google TPU, AMD, Marvell, OpenAI's chips). This fragmentation reduces Broadcom's leverage and increases the risk that its custom ASIC business becomes a niche player rather than a primary alternative.

OpenAI's chip strategy. OpenAI claimed in late August that its new chips outperform Nvidia processors in tests, signaling a new entrant to the custom silicon market. If OpenAI's chips prove viable at scale, they could cannibalize demand for both Broadcom and AMD custom silicon, as hyperscalers may prefer to build proprietary solutions rather than rely on third-party ASICs. This introduces a new competitive vector that the thesis did not originally account for.

Anthropic's Nvidia pivot. Despite having bet tens of billions on AMD infrastructure, Anthropic signed a $45 billion, six-year lease for Nvidia Vera Rubin computing capacity from Nscale in late August 2026. This high-profile reversal suggests that even AI labs that committed to AMD are finding Nvidia's ecosystem and performance advantages compelling enough to justify the cost premium. The deal's scale ($45 billion over six years) implies that Nvidia's supply constraints may be easing, reducing the urgency for hyperscalers to adopt custom silicon.

Broadcom's credit and financial risks. Broadcom is reportedly exploring a $100 billion debt-financing package to back AI chip deals for Anthropic and others. While this underscores the company's centrality to AI infrastructure, it also raises financial risk: if custom ASIC adoption slows or hyperscalers shift to alternatives, Broadcom could face a debt servicing crisis. Multiple sources flagged that Broadcom's AI spending is pushing credit risks higher, suggesting that the market is pricing in execution risk on the financing side.

What to watch

Broadcom's next earnings report for custom AI ASIC revenue trends, Samsung partnership revenue recognition, and management commentary on customer concentration risk and the Google design-away. The company's ability to demonstrate sequential revenue growth and margin expansion in custom silicon will be critical to restoring investor confidence.

AMD's Q3 2026 earnings for AI accelerator revenue growth, gross margins, and guidance—particularly whether the company is gaining or losing share in hyperscaler deployments relative to Nvidia and custom silicon alternatives.

Hyperscaler earnings commentary (Microsoft, Meta, Google, Amazon) on AI capex allocation and whether custom silicon adoption is accelerating or stalling. Specific disclosure of spending on Broadcom, AMD, and proprietary custom silicon will be key to assessing the thesis's viability.

Nvidia's next earnings for Vera CPU revenue, gross margins, and customer concentration—to confirm whether hyperscalers are truly diversifying or simply supplementing Nvidia with small custom silicon allocations.

Broadcom's financing vehicle updates and credit rating actions, as the $100 billion debt estimate becomes more concrete. Any downgrade or covenant tightening would signal that the market is losing confidence in the custom ASIC thesis.

Marvell's Google deal revenue recognition timeline, which is now pushed to fiscal 2029, suggesting a slower ramp than initially expected and raising questions about the pace of hyperscaler custom silicon adoption more broadly.

OpenAI's chip strategy, given its August 25 claim that its new chips outperform Nvidia processors in tests. Any production ramp or customer wins would signal a new entrant to the custom silicon market that competes with both Broadcom and AMD.

TensorWave's progress (the AMD-backed AI cloud provider) on customer acquisition and revenue growth. If TensorWave fails to gain traction, it would suggest that AMD's ecosystem play is not resonating with hyperscalers.

Related Arbora context

This thesis is closely linked to the broader AI infrastructure and data center build-out thesis, which focuses on the entire stack of compute, power, and networking required for AI workloads. The challenger silicon narrative is a subset of that thesis, focused specifically on the compute layer's competitive dynamics. The CPU renaissance and advanced process node competition thesis also overlaps, as AMD's AI accelerator gains are part of a broader CPU/advanced-node story. Additionally, the Micron memory chip supercycle thesis is complementary, as custom silicon adoption drives demand for high-bandwidth memory and DRAM that Micron supplies.

Sources

This article is research notes, not financial advice.