BROADCOM: NAVIGATING THE INFRASTRUCTURE TRANSITION

Strategic Memo to Leadership

June 2030 | CEO Edition

TO: Broadcom Executive Team

FROM: Office of the Chief Strategy Officer

RE: Strategic Repositioning for the Post-AI-Boom Era

The past two years have been the most turbulent in Broadcom's modern history. Between June 2027 (peak valuation) and June 2030, we've experienced: - A 58% drawdown in stock price - A 34% compression in forward earnings multiples - The departure of 18% of institutional investors - Activist hedge fund pressure to accelerate share buybacks

Yet on deeper inspection, our fundamental business position is stronger than the market assumes. This memo outlines the strategic reality and our path forward.

THE MARKET'S MISDIAGNOSIS

Wall Street believes Broadcom is suffering from "structural AI infrastructure decline." This is imprecise. The market is confusing training infrastructure demand (which has clearly plateaued) with overall infrastructure demand (which continues to grow, albeit more slowly).

Consider the facts: - Q1 2030 networking segment revenue: $2.3B (up 8% YoY) - Q1 2030 infrastructure software revenue: $1.7B (up 19% YoY) - Operating margin in core segments: 56-62% (essentially flat vs. 2027)

We are not a company in decline. We are a company that was priced for 45% CAGR and is being valued for 8% CAGR. Both assumptions are wrong—the truth is probably 12-14%.

THE REAL INFLECTION POINT: 2028-2029

Between August 2028 and March 2029, the market experienced what some research organizations called the "2028 Intelligence Crisis"—a moment when the exponential scaling of AI hit economic boundaries. Model training costs exceeded $50 million per frontier model. Inference became the focus. Efficiency became the religion.

What the market missed is that this transition doesn't hurt Broadcom as much as it hurts our competitors. Here's why:

Custom Silicon is a Distraction: Yes, AWS, Google, Microsoft, and Meta are designing their own chips. But they still need: - The networking fabric to connect these chips - The switching infrastructure to route data - The orchestration software to manage them - The high-bandwidth memory interconnects to feed them

We don't lose the entire transaction; we lose the custom silicon margin and keep the infrastructure margin. That's actually a better business model long-term (lower cap intensity, predictable, recurring revenue).

The Real Competition is Economics: Our competitors are not Intel or NVIDIA. Our competitors are hyperscalers deciding whether to spend $1.2 billion on a data center network upgrade or patch-work their existing infrastructure. The VMware acquisition gave us the software layer to make the upgrade case much stronger than our competitors can.

OUR STRATEGIC POSITIONING: THREE PILLARS

Pillar One: The Networking Fabric Becomes THE Bottleneck

By 2030, we've observed a clear pattern: companies optimizing AI inference workloads discover that their bottleneck is not compute, not memory, but network latency. Distributed inference across multiple edge locations requires sub-100-microsecond latency. That's a Broadcom domain.

Our action plan: - Increase R&D spend by 18% in fiscal 2031, focused on ultra-low-latency switching architectures - Target 12% margin expansion in networking through design-in wins with hyperscalers - Price this as a "strategic necessity," not a commodity

Pillar Two: VMware Becomes the Multi-Vendor Operating System

The VMware acquisition was strategically sound but operationally botched. We paid $61 billion for a company with 4% organic growth. By 2030, we're seeing why that was actually fair.

When a Fortune 500 company wants to deploy AI across on-premises systems, AWS, Azure, and custom silicon (because some workload requires specific hardware), they need orchestration software. VMware is the only player that can credibly manage that heterogeneity.

Our action plan: - Decouple VMware's AI/ML workload orchestration features into a stand-alone product (tentative name: VMware Atlas) - Price it at 3.2x the margin of our legacy VMware products (customers will pay this for multi-cloud coherence) - Target $400M in new atlas revenue by fiscal 2032 - This alone could drive 8-12% growth in VMware segment

Pillar Three: HBM Interconnect Becomes a Secular Growth Driver

High Bandwidth Memory demand increased 156% between 2027-2029. Projection: 89% CAGR through 2032.

Why? Because AI models are simultaneously getting more memory-efficient (smaller) and more memory-hungry (because they're deployed at scale). The memory hierarchy problem becomes acute.

Broadcom controls approximately 34% of the HBM interconnect market. Our margin on this product: 67%.

Our action plan: - Invest $180M in HBM interconnect R&D over next three years - Target 8% market share gains (from 34% to 42%) by 2032 - This segment could represent 11% of total revenue and 15% of gross profit by 2032

THE HARD TRUTH: EARNINGS GUIDANCE

The market expects Broadcom to return to 25%+ revenue growth. This is not happening.

Realistic forecast: - FY2031: 9% revenue growth, 58% gross margin - FY2032: 11% revenue growth, 59% gross margin - FY2033: 13% revenue growth, 61% gross margin (as new VMware and HBM initiatives scale)

This implies $11.20 EPS in FY2032 (vs. consensus estimate of $13.50). The stock is rationally priced if we deliver on this plan.

ORGANIZATIONAL IMPLICATIONS

1. Accelerate VMware Integration: We've been running VMware as a standalone. By Q4 2030, they need to be fully integrated into our go-to-market organization. The AI/ML orchestration story cannot be told by VMware alone—it must be told by Broadcom Infrastructure.

2. Hire Aggressively in Analog/Mixed-Signal Design: HBM interconnect requires world-class analog designers. Poach from NVIDIA, AMD, and Intel. Offer +40% premium on market salary for the best talent. This is mission-critical.

3. Restructure Sales Incentives: Currently, salespeople are incentivized on total deal size. We need to change incentives to focus on long-term relationship value and recurring revenue. A networking customer with a 5-year relationship is worth 4x more than a custom silicon sale.

4. Divest or Integrate Struggling Assets: We own several smaller software businesses acquired in the last 10 years that are growing <3% and have margins below 30%. Divest these by Q2 2031. We need to be seen as a "focused infrastructure company," not a conglomerate.

THE CAPITAL ALLOCATION QUESTION

Our stock is down 58% from peak. The board is asking: Should we increase buybacks?

My recommendation: NO (yet).

Here's the logic: - Buyback at $118 stock price assumes 12% IRR on the reinvested capital - Our internal ROIC on infrastructure R&D and VMware integration is 18-21% - Our internal ROIC on HBM interconnect buildout is 15-17%

We should redeploy half of what we would normally spend on buybacks into R&D and M&A. In 2032, when we've proven the VMware Atlas and HBM platforms are working, we can return to normal capital allocation.

This is not popular with activists. It's correct.

THE 18-MONTH MILESTONES

To reset market expectations and restore credibility:

By Q4 2030: - VMware Atlas v1.0 shipped, >50 customers in beta - HBM interconnect revenue >$450M (up from $280M in Q1 2030) - Networking segment margin at 58% (vs. 55% in Q1 2030)

By Q2 2031: - VMware Atlas revenue run rate >$80M annualized - Announce $500M strategic partnership with one of "the three" (AWS/Azure/GCP) for multi-cloud orchestration - Introduce next-gen switching architecture with <50 nanosecond latency

By Q4 2031: - Prove FY2032 guidance is realistic (9% revenue growth on a run-rate basis) - Stock price recovers to $155-170

CLOSING THOUGHTS

Broadcom will not return to the valuation multiples of 2027. That was a bubble. But we can build a highly profitable, slowly growing infrastructure company worth $220-240 billion by 2033.

The next 18 months will determine whether we're seen as a "fading infrastructure company" or "the platform that makes multi-vendor AI deployment work."

We control that narrative.