AVGO NVDA AI Stack Competitive Position: How Broadcom Holds Custom ASIC While Network Silicon Compresses

AVGO NVDA AI stack competitive position is the structural question Broadcom investors should be thinking about after Jensen Huang's June 1-2 Computex 2026 keynote signaled NVDA's intent to "win every layer of the AI stack" ¹. Broadcom — a $2.18 trillion company ² with the deepest ASIC and networking silicon franchise outside NVDA — has historically benefited from NVDA's vertical focus on accelerator GPUs by capturing the adjacent layers (custom ASICs for hyperscalers, switch silicon, optical components). The Computex keynote signals a strategic shift: NVDA increasingly wants to monetize the network layer (Spectrum-X), the PC layer (Arm-based desktop chips), and the physical AI layer (robotics infrastructure) — three layers in which AVGO has been the established incumbent. The competitive dynamics through 2027-2028 will determine whether AVGO's $2.18T valuation absorbs structural margin compression or whether the company's existing customer relationships create a durable moat.

What Huang's Computex 2026 Keynote Signals for AVGO

Jensen Huang's Computex 2026 keynote disclosed two strategic positionings that have direct competitive implications for AVGO ¹.

First, NVDA's positioning that they intend to win at every layer of the AI infrastructure stack. The specific articulation was that NVDA's value proposition is no longer just GPU accelerators — it now extends to GPU+CPU integration (Grace Blackwell), networking (Spectrum-X Ethernet platform), client/PC (Arm-based PC chips with Microsoft Surface and Dell), and physical AI (ABB Robotics, Advantech systems integration partnerships).

Second, the public "anointing" of Marvell as the next $1 trillion company. From the AVGO perspective, this was notable: Huang did not similarly anoint Broadcom. AVGO is already a $2.18T company ² — there is no "next $1 trillion" framing for it. But the absence of AVGO mention in Huang's strategic compliments paragraph is itself a signal that NVDA's near-term strategic alliances are oriented toward MRVL (in optical and ASIC partnership zones), not toward AVGO.

The structural read: NVDA is establishing the framework where it owns the GPU + networking + physical AI infrastructure, MRVL is its primary ASIC + optical partner, and AVGO is the established incumbent that NVDA will increasingly compete against in some segments while continuing to serve some hyperscaler customer relationships that NVDA cannot serve directly.

Why AVGO's Position Is Both Strong and Vulnerable

Three structural reads on AVGO's competitive position in the AI stack vertical.

First, AVGO's customer relationships in custom ASIC are the deepest in the industry. Google has used AVGO for TPU silicon since 2015 (now in v5/v6 generations) — a 10+ year incumbency that would take 3-4 years to displace. Meta's MTIA platform also runs on AVGO designs. Apple, Cisco, and other major customers maintain decade-plus AVGO relationships. The replacement cost for these customers to migrate to alternative ASIC partners is high — measured in 2-3 years of engineering time per replacement.

Second, the network silicon market — Broadcom's Tomahawk and Jericho switch silicon families — is where direct NVDA competition is now real. NVDA's Spectrum-X platform positions a fully integrated Ethernet networking solution that competes with AVGO's switch silicon in the hyperscale data center market. The 2026-2028 capex cycle will determine how much of the network silicon TAM AVGO retains vs how much NVDA captures through Spectrum-X. The early signal: NVDA Spectrum-X has already won design slots at AWS, Microsoft Azure, and Google Cloud for next-generation AI clusters.

Third, AVGO's diversified business structure — semiconductor + software (VMware, CA, Symantec acquisitions) — provides cushion against pure-semiconductor competition. TTM gross margin of 67.1% ² is significantly higher than MRVL's 50.6% or pure-semi peers, driven by the software segment's 80%+ gross margin contribution. This diversification gives AVGO operating margin resilience even if certain semiconductor segments compress.

Data Points: AVGO vs NVDA Comparative Position

Table 1: AVGO standalone profile at June 2, 2026 ²

Table 2: AVGO quarterly revenue trajectory through Q1 FY26 ³

Quarterly revenue grew from $14.05B (Q4 FY24) to $19.31B (Q1 FY26) — a 37.4% expansion over 5 quarters. The acceleration is concentrated in the AI semiconductor segment, which now represents roughly 35-40% of total revenue per AVGO's own disclosures.

Table 3: AVGO vs NVDA vs MRVL at June 2, 2026 ²

The structural read: AVGO is positioned between NVDA (vertical AI infrastructure leader) and MRVL (focused ASIC and optical specialist). AVGO captures the wider TAM but at lower margin than NVDA. The strategic question through 2026-2028 is whether AVGO can maintain its 35-40% AI semiconductor revenue share against NVDA's vertical expansion.

Analysis: AVGO Competitive Position vs NVDA Through 2027-2028

Three structural reads on the AVGO NVDA AI stack competitive position.

(1) Custom ASIC business has the longest-term durability. AVGO's Google TPU and Meta MTIA relationships are difficult for NVDA to penetrate because the entire chip design IP belongs to the customers. NVDA would need to convince Google and Meta to abandon their proprietary architectures — economically irrational. AVGO retains this segment with high probability through 2028+.

(2) Network silicon faces structural compression. Broadcom's Tomahawk and Jericho switch silicon directly compete with NVDA Spectrum-X Ethernet platform. The 2025-2026 design wins at hyperscalers favored Spectrum-X in several cases. The forward read: AVGO's network silicon segment may face 5-10 percentage point margin compression as it competes more aggressively to retain share.

(3) Software business provides margin resilience. AVGO's VMware + CA + Symantec software stack contributes 22% of revenue at 80%+ gross margin. This diversification provides operating margin floor even if specific semiconductor segments compress. The bull case for AVGO's forward P/E remaining 35x is that the software cushion absorbs semiconductor volatility.

The valuation tension: AVGO at 35.5x forward P/E ² is currently priced at a discount to MRVL (59.5x) but a premium to legacy semiconductor peers (Intel, Texas Instruments at 20-25x). Multiple expansion to 40-45x requires AVGO to demonstrate clear AI semiconductor revenue acceleration beyond the 25% TTM growth. Multiple compression to 28-32x is the bear case if NVDA's vertical expansion captures 10-15 percentage points of switch silicon share.

What to Watch Through Q4 2026

Three near-term catalysts for the AVGO NVDA AI stack competitive position story:

1. September 2026 — AVGO Q3 FY26 earnings: First quarter that may explicitly address NVDA Spectrum-X competition. Watch for (a) AI semiconductor segment revenue acceleration, (b) network silicon segment commentary, (c) custom ASIC customer pipeline disclosure.
2. Q4 2026 — Hyperscaler 2027 capex planning: AWS, Google Cloud, Microsoft Azure, Meta will publicly outline 2027 AI infrastructure capex commitments. The mix between AVGO ASIC, NVDA Spectrum-X, and MRVL optical determines the second-wave AVGO trajectory.
3. Q1 2027 — NVDA Spectrum-X v2 launch: NVDA's next-generation networking platform announcement will signal whether NVDA accelerates vertical capture or maintains an ecosystem-friendly approach.

For paying readers, drillr terminal tracks AI semiconductor segment-level revenue trajectory, hyperscaler design-win commentary, and the multi-vendor competitive positioning in real time.

---

Footnotes

1. CNBC, "Nvidia's new PC chips represent CEO Huang's bid to win at every layer of AI stack," June 2, 2026; Bloomberg, "Marvell Surges After Huang Calls It the Next $1 Trillion Company," June 2, 2026. ↩ ↩²

2. Broadcom (AVGO), NVIDIA (NVDA), Marvell (MRVL) company snapshot via drillr terminal, TTM metrics, valuation, and intraday pricing as of 2026-06-02. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

3. AVGO quarterly financial statements (Q4 FY24 through Q1 FY26) via drillr terminal, accessed 2026-06-02. ↩