NVIDIA: THE STRATEGIC CROSSROADS

A Memo for the Executive Leadership and Board from June 2030

FROM: Executive Intelligence Unit DATE: June 2030 RE: NVIDIA in 2030: Defending the Moat or Adapting to Disruption

EXECUTIVE SUMMARY FOR THE BOARD

NVIDIA has become a technology monopoly with the highest profit margins of any company on Earth. The data center GPU business generates $47.2 billion in annual revenue at 54% gross margins. CUDA—the software layer that makes NVIDIA GPUs indispensable—has created switching costs that are the envy of the technology world.

The problem: Both of these advantages are being systematically eroded by customers you created.

Google created TPU to reduce GPU dependence. Amazon created Trainium to own their own margin. Microsoft is deploying Maia to reduce OpenAI inference costs. AMD's MI300 architecture, while still inferior, has achieved parity in most benchmarks at 22% lower TCO.

The board needs to confront a strategic reality: NVIDIA cannot maintain 80%+ market share in AI accelerators indefinitely. Competition is real. Margin compression is not a cycle—it's a trend.

The strategic question: How does NVIDIA remain the dominant AI infrastructure company in a world where customers build their own chips?

THE CORE BUSINESS: DATA CENTER GPUs

Current state: $47.2 billion in annual revenue (FY2030), growing 12% YoY. Gross margins at 54%, net margins at 32% at the data center division level.

Market structure: - Cloud providers (AWS, Azure, Google Cloud): 34% of NVIDIA GPU consumption - Hyperscaler internal operations (Google, Meta, Amazon, Apple, OpenAI, Anthropic): 31% - Enterprise AI customers: 23% - Research institutions: 9% - Retail/small business: 3%

The problem: The hyperscaler internal operations segment (31%) is systematically moving to custom silicon. By June 2030: - Google: 75% of training workloads on TPU, 25% on NVIDIA - Amazon: 60% of workloads on Trainium/Inferentia, 40% on NVIDIA - Meta: 45% of workloads on MTIA, 55% on NVIDIA - Microsoft: 70% of inference workloads on Maia (increasing), 30% on NVIDIA - OpenAI: 85% on NVIDIA (not building own chips) - Anthropic: 88% on NVIDIA (not building own chips)

The calculation: These hyperscalers represent approximately 31% of NVIDIA's total addressable market. If the trend continues (hyperscalers moving to 80%+ custom silicon by 2032), NVIDIA loses $15 billion in annual revenue.

Cloud provider segment (34%) has different dynamics: AWS, Azure, and Google Cloud can't effectively build their own chips and sell them to third-party customers—that would cannibalize their own GPU services. They need NVIDIA chips for external customers. But they're incentivizing customers to buy their own custom chips and become less dependent on GPU consumption.

The net effect: NVIDIA's total addressable market is shrinking, not growing, as customers vertical-integrate.

THE MOAT: CUDA AND ITS EROSION

CUDA has been NVIDIA's strongest defensible asset. Developers write AI frameworks in CUDA. Customers buy NVIDIA chips to run CUDA software. Switching to AMD ROCM requires recompilation and reoptimization. High switching costs create loyal customers.

But CUDA's moat is eroding:

2025-2026: PyTorch and TensorFlow add ROCM support. Switching still requires optimization work, but it's possible.

2027-2028: ROCM ecosystem matures. OpenAI and Anthropic both announce AMD compatibility. The psychological barrier to switching weakens.

2029-2030: ROCM is considered production-ready by major AI companies. OpenAI runs official tests on AMD MI300. Anthropic deploys AMD infrastructure for some workloads. The switching cost is no longer prohibitive.

Market response: AMD's GPU market share goes from 3% (2025) to 14.8% (2030).

The CUDA strategy going forward:

Option A: Accept CUDA erosion and compete on architectural superiority. - Requires continuous innovation (new architectures every 18 months, not 24-36 months) - Accepts lower market share (50% by 2035 instead of 70%+) - Requires maintaining price/performance leadership - Probability of success: 60%

Option B: Use CUDA as a moat for proprietary solutions, not commodity chips. - Develop vertical solutions (NVIDIA Omniverse, NVIDIA AI Enterprise) that require CUDA - Bundle CUDA-dependent software with chips to increase switching costs - Move from selling chips to selling integrated solutions - Probability of success: 55%

Option C: Make CUDA open and universal. - Open-source CUDA to eliminate switch barriers (if everyone uses CUDA, AMD, Intel, Qualcomm can also use it) - Compete on chip architecture and software ecosystem, not lock-in - This is a long-term play that requires 5+ years to pay off - Probability of success: 40%

The board needs to decide: Are we a chip company defending our CUDA moat, or are we an AI platform company using CUDA to build integrated solutions?

STRATEGIC OPTIONS: THE FORK IN THE ROAD

Path A: The Chip Company Thesis (Current Strategy)

Assumption: NVIDIA remains the dominant GPU manufacturer through continuous innovation. Market share stabilizes at 50-60% by 2035. Margins decline to 35-40%. The company grows revenue to $450-500 billion by 2035.

Strategy requirements: - Maintain architectural superiority over AMD and Intel GPUs - Win in new markets (robotics, autonomous vehicles, edge AI) - Expand data center software (CUDA, TensorRT, Triton) moat - Aggressive M&A in AI software space

Risks: - AMD continues architecture parity improvements, eventually exceeds NVIDIA - Custom chips become superior to commodity GPUs (they're more tailored) - Price compression accelerates as competition intensifies - Margins compress below 30% (breaking the business model)

Timeline to breakpoint: 2032-2033. By then, if market share has declined to below 40% and margins are below 25%, this strategy has failed.

Path B: The AI Platform Company Thesis (Possible Pivot)

Assumption: NVIDIA acknowledges that commodity chip dominance is unsustainable. Instead, NVIDIA moves up the stack into software and integrated solutions (Omniverse, AI Enterprise, autonomous vehicle platforms, robotics platforms).

Chips become a component, not the business. Software and solutions become the margin driver.

Strategy requirements: - Significant M&A in AI software space (acquire or build) - Development of vertical market solutions (automotive, robotics, digital twins) - Create "NVIDIA-powered" certification program to lock in ecosystem - Reduce dependence on chip margins, increase software/service margins

Risks: - Software integration is harder than chip manufacturing - Requires different organizational culture and capabilities - Competitors (Microsoft, Google, Amazon) have stronger software chops - Margin profile might not improve enough to justify the capital required

Timeline to payoff: 2035+. Requires 5-year investment before knowing if strategy worked.

Path C: The Robotics/Autonomous Vehicle Dominance Thesis

Assumption: The next growth frontier after LLM training and inference is robotics. NVIDIA's AI infrastructure expertise gives it a first-mover advantage in robotics platforms. By 2035, robotics and autonomous vehicles represent 25% of NVIDIA revenue.

Strategy requirements: - Acquisition of robotics software companies (Boston Dynamics? Skydio? Others?) - Development of robotics-specific chip architectures (different from data center GPUs) - Deep partnerships with robotics OEMs - Investment in robotics R&D and commercialization

Risks: - Robotics adoption might be slower than expected (different from AI adoption curve) - Competitors (Tesla, Boston Dynamics, traditional OEMs) might dominate - NVIDIA's strength is in AI infrastructure, not embodied AI - Requires capabilities outside NVIDIA's historical expertise

Timeline to significance: 2033-2035. Robotics needs to represent meaningful revenue by then.

ORGANIZATIONAL REALITY: CAN NVIDIA EXECUTE?

NVIDIA has 47,000 employees, up from 26,000 in 2025. The company has scaled dramatically. But organizational stress points are visible:

Strength: Engineering excellence. NVIDIA's chip design teams are the best on Earth. The company continues to innovate faster than competitors.

Weakness: Software and platform development. NVIDIA's software teams (CUDA, TensorRT, Triton) are excellent at optimization, not world-class at platform architecture. This is historically why Google's Tensor Flow ecosystem is stronger than NVIDIA's ecosystem despite NVIDIA's hardware dominance.

Weakness: Customer relationship management. NVIDIA's go-to-market has been "we make the best chips, you build solutions on top." As competition intensifies, NVIDIA needs deeper relationships with customers to understand their needs and bundle solutions. The company is reorganizing sales, but this is a capability gap.

Organizational culture question: Can NVIDIA transition from a pure engineering company to a product/platform company? The leadership (Jensen Huang, Colette Kress, Debora Sherry) are excellent, but the organizational DNA is hardware-first, not software-first.

FINANCIAL STRATEGY: THE CAPITAL ALLOCATION QUESTION

NVIDIA generated $128 billion in operating cash flow in FY2030. The company has $45 billion in cash on the balance sheet. Dividend is $1.2 billion annually (0.9% yield). No share buybacks.

The capital allocation question: What should NVIDIA do with $128 billion in annual free cash flow?

Current deployment: - R&D: $12.3 billion (4% of revenue) - Capital expenditure: $4.2 billion (1.3% of revenue) - Acquisitions: $2.1 billion (small M&A activity) - Debt paydown: $0 (no debt) - Dividends: $1.2 billion - Cash accumulation: $108 billion (build up reserves)

The strategic choice: Is NVIDIA building reserves for a transformational acquisition, or is it simply hoarding cash due to lack of strategy clarity?

Options: 1. Aggressive M&A: Acquire 4-5 AI software companies ($20-30 billion total) to build software platform capability 2. Accelerated R&D: Increase R&D to 8-10% of revenue ($25-32 billion) to fund next-generation architectures faster 3. Vertical integration: Invest $30-40 billion to acquire robotics companies or autonomous vehicle technology 4. Shareholder returns: Increase dividend to 3-4% yield and initiate $40-50 billion share buyback program 5. Strategic hedging: Invest in competitors (AMD, Intel) or complementary companies (ASML, TSMC) to hedge bets

The board needs clarity: What is the capital strategy, and does it align with the chosen strategic path (Chip Company, Platform Company, or Robotics Dominance)?

JENSEN HUANG AND SUCCESSION PLANNING

Jensen Huang is the face of NVIDIA. His leadership has been exceptional. But at 52, with 26 years as CEO, succession planning is overdue.

Current succession plan: Effectively non-existent. No designated COO. No clear pipeline for next-generation leaders.

Risk: If Huang were to depart unexpectedly (illness, retirement, external opportunity), NVIDIA would face immediate stock volatility and potential strategic paralysis.

Board recommendation: Begin formal succession planning in 2030, target COO hire or internal promotion by 2031, plan for CEO transition by 2034-2035.

This isn't about questioning Huang's performance. It's about reducing single-person risk and ensuring long-term organizational stability.

THE COMPETITIVE REALITY

NVIDIA's competitors are not just AMD. They are:

• Google TPU: Already capturing 15% of NVIDIA's addressable market. Owned by a $2 trillion company with unlimited capital.
• Amazon Trainium/Inferentia: Capturing internal AI workloads at Amazon, one of the world's largest AI operators.
• Microsoft Maia: Increasing inference workload capture. Owned by a $3 trillion company.
• AMD MI300/MI300X: Achieving performance parity at lower cost. Growing market share from 3% to 15% in five years.
• Intel: Not yet competitive, but R&D investment is significant. Intel could re-emerge by 2032.
• Custom silicon trend: Every major AI company is building its own chips or considering it. This represents existential competition.

The honest assessment: NVIDIA's monopoly is ending. The company needs to accept this and develop strategy accordingly.

WHAT THE BOARD NEEDS TO DECIDE IN THE NEXT 12 MONTHS

1. Strategic identity: Are we a Chip Company, Platform Company, or Robotics Company? This determines everything else.

2. Market share target: What's our acceptable market share by 2035? (50%, 40%, 30%?)

3. Margin defense: How do we defend 30%+ net margins against competition? (price, volume, vertical integration?)

4. Capital strategy: How do we deploy $128B annual free cash flow? (M&A, R&D, returns, hoarding?)

5. Succession plan: Who follows Jensen Huang, and when? (Timeline for COO hire, CEO transition)

6. Organizational capability: Do we need to acquire software/platform expertise, or can we build it internally?

THE CLOSING STRATEGIC ASSESSMENT

NVIDIA is the most powerful technology company in the world, by some measures. The data center GPU business is genuinely excellent. CUDA is a strong moat, weakening but defensible.

But the company is at a strategic inflection. The path of 50% annual growth with 40%+ margins cannot continue indefinitely. Compression is inevitable.

The board and leadership need to confront this squarely and make deliberate strategic choices about where NVIDIA competes and how.

The companies that fail at inflections are the ones that deny the inflection exists. NVIDIA needs to avoid this trap.

This strategic assessment is prepared for the NVIDIA Board of Directors and C-suite executives. It represents an external perspective on organizational and strategic challenges as of June 2030.