Invest in NVIDIA (NVDA) and Meta Platforms (META) to capture direct upside as annual AI infrastructure CapEx expands from $1 trillion to over $2 trillion by 2028.

Buy ASML Holding (ASML) for long-term exposure to critical photolithography bottlenecks that control the physical scaling of global computing capacity through 2030.

Capitalize on deeply discounted 2x to 3x earnings valuations across memory manufacturers like Micron Technology (MU), SK Hynix, and Samsung, which maintain immense pricing power driven by structural High Bandwidth Memory (HBM) shortages.

Maintain exposure to Alphabet (GOOGL) and Amazon (AMZN) as they roll out proprietary custom silicon to defend enterprise cloud margins amid $11 trillion in cumulative infrastructure spending through 2029.

Trim traditional fixed income and low-growth dividend-paying defensive equities to avoid valuation compression as over $5 trillion in new tech bond issuances drives global borrowing yields higher.

NVIDIA Corporation (NVDA)

• NVIDIA continues to dominate the AI hardware ecosystem, powering leading models across frontier labs with hardware such as Hopper, GB300, and the upcoming Vera Rubin architecture.
  • Next-generation chips are projected to deliver 3x to 5x more performance per watt, driving significant compute efficiency gains.
  • While hardware margins were historically captured heavily at the chip level, value is shifting as frontier labs monetize models at rates between $50 million to $100 million per megawatt, giving hardware vendors strong pricing leverage.
  • Overall global AI capital expenditure (CapEx) is projected to grow from $1 trillion currently to well over $2 trillion annually by 2028.

Takeaways

• NVIDIA remains the primary hardware beneficiary of the massive CapEx expansion through 2028, supported by rapid efficiency advancements in its chip roadmap.
• Investors should monitor shifting value capture across the supply chain, as pricing power dynamically moves between chip designers, memory suppliers, and model creators.

Meta Platforms, Inc. (META)

• Meta is aggressively building and hoarding massive compute infrastructure using its strong balance sheet, without needing upfront end-customer commitments.
  • The company benefits from internal efficiency gains, such as optimizing ad algorithms to increase engagement time by 5%, while generating far more direct value from AI models than the base cost of compute.
  • Meta possesses high optionality: it can either deploy its compute internally or rent it out to frontier labs at high premiums ranging from $25 million to $50 million+ per megawatt.
  • To finance massive infrastructure, Meta has raised debt at 5% to 6%, with willingness to borrow at higher rates (up to 8%) due to the immense return on invested capital in AI compute.

Takeaways

• Meta is positioned as a uniquely capitalized player with dual upside: deploying proprietary AI to enhance core advertising revenue and acting as a secondary compute lessor at high gross margins.

ASML Holding N.V. (ASML) & Semiconductor Equipment

• Tooling and photolithography equipment represent one of the tightest physical bottlenecks in scaling global compute capacity.
  • Approximately $6 billion in semiconductor fabrication CapEx can produce 1 gigawatt of annual compute, which in turn generates over $100 billion in downstream AI revenue.
  • Extreme Ultraviolet (EUV) tools face multi-year supply chain lags; suppliers like Carl Zeiss are expanding production toward 100 EUV tools per year by 2030, which may still fall short of economic demand.

Takeaways

• Semiconductor equipment manufacturers hold an essential monopoly on the physical expansion of AI compute, making ASML a critical long-term bottleneck asset despite elongated supply cycle lead times.

Memory Manufacturers: Micron Technology (MU), SK Hynix, and Samsung

• High Bandwidth Memory (HBM) and advanced DRAM are capturing an increasing share of gross margins within the data center supply chain.
  • A single gigawatt of compute requires roughly 170,000 DRAM wafers, making memory a key structural constraint alongside logic chips.
  • Memory suppliers have gained substantial pricing power compared to contract foundries, rapidly raising prices during supply crunches.
  • Memory companies frequently trade at low multiples of 2x to 3x earnings, reflecting broader market discount rate pressures and cyclical concerns despite record demand.

Takeaways

• Memory producers are primed for elevated pricing power and revenue growth driven by HBM demand, though investors should account for valuation compression common across cyclical semiconductor sub-sectors.

Big Tech Hyperscalers: Alphabet Inc. (GOOGL) & Amazon.com, Inc. (AMZN)

• Hyperscalers are transitioning from funding infrastructure entirely out of free cash flow and stock buybacks toward raising substantial corporate debt.
  • Total cumulative CapEx across data center and energy infrastructure is estimated to reach $11 trillion between 2024 and 2029, requiring an estimated $5 trillion in credit issuance.
  • Both companies are investing heavily in proprietary custom silicon, including Google's TPU v7 and Amazon's Trainium 3, to reduce reliance on third-party merchant chips and improve power efficiency.

Takeaways

• Expect lower share buyback activity and higher balance sheet leverage as hyperscalers prioritize debt-financed infrastructure buildouts to capture enterprise cloud and AI hosting demand.

Frontier AI Labs: Anthropic & OpenAI (Private)

• The leading AI labs have crossed into positive unit economics, shifting away from pure venture-funded operating losses.
  • Anthropic became profitable in Q2, with OpenAI expected to follow in Q3; revenue generation has surged from negative margins on early models up to $50 million to $100 million per megawatt.
  • Combined, OpenAI and Anthropic are projected to consume 40% to 50% of all incremental worldwide compute capacity next year, with the trajectory pointing toward controlling the majority of global usable compute by 2028.
  • Labs are increasingly reallocating marginal compute capacity from external commercial inference toward internal research and development (R&D) and model training to accelerate capabilities.

Takeaways

• Frontier labs are achieving unprecedented revenue density per megawatt, creating a self-reinforcing flywheel where profits are immediately reinvested into expanding training clusters.
• Near-term revenue velocity could face headwinds if regulatory restrictions or voluntary safety pauses delay the public release of frontier models.

SpaceX (Private)

• SpaceX is emerging as a major infrastructure player by deploying massive compute capacity and leasing it to frontier labs.
  • The company has engaged in high-margin compute arbitrage, securing deals to sell capacity to labs like Google and Anthropic for $25 billion to $40 billion per gigawatt ($25 million to $40 million per megawatt).
  • This dynamic enables SpaceX to recoup infrastructure CapEx in short timelines by monetizing the scarcity of immediately available data center power.

Takeaways

• Private market investors in SpaceX gain indirect exposure to high-margin compute leasing and AI infrastructure monetization outside of its core aerospace operations.

Traditional Equities, Fixed Income & Global Macro

• The massive capital requirements of AI infrastructure threaten to crowd out other sectors in the credit and debt markets.
  • Financing $5 trillion+ in data center debt could push corporate borrowing spreads higher across the broader economy by 200 to 250+ basis points.
  • Higher prevailing interest rates raise corporate discount rates to 8% to 10%+, severely compressing the valuation multiples of traditional dividend-paying, long-duration "stable" stocks (e.g., consumer goods, utilities, and railroads).
  • Highly indebted developing nations (e.g., Pakistan, Nigeria) face acute default risks as global capital is redirected toward high-yielding US AI infrastructure.

Takeaways

• Higher systemic cost of capital poses persistent valuation headwinds for capital-intensive, low-growth defensive equities and sovereign debt.
• Fixed income investors should prepare for higher long-term yields as massive private-sector data center bond issuances compete directly with government debt.