Investing.ad
news

AI Chips Weekly: Rotation From Hype To ‘Picks-and-Shovels’

SoftBank exits Nvidia to fund OpenAI, hedge funds add AI exposure, and new data on GPU depreciation extends the ROI window—setting up value in packaging, substrates, memory, and foundry.

AI Chips Weekly: Rotation From Hype To ‘Picks-and-Shovels’
#AI chips #GPUs #Accelerators #HBM memory #Packaging #Substrates #Foundry #Nvidia #AMD #TSMC

AI Chips Weekly: Rotation From Hype To ‘Picks-and-Shovels’

Executive Summary

  • Sentiment: Mixed-to-cautious near term. Wells Fargo advises profit-taking as AI capex payoffs get questioned, but hedge funds and Berkshire add AI-adjacent exposure.
  • Risks: Multiple compression after a strong run, capex ROI uncertainty, and financing costs; supply tightness (HBM, advanced packaging) and geopolitics remain.
  • Catalysts: Nvidia earnings (Nov 19), AMD’s multi-year AI revenue growth guide, and evidence that GPU assets depreciate more slowly than feared—supporting long-tailed cash flows.
  • Positioning: Favor “picks-and-shovels” beneficiaries—packaging, substrates, memory, foundry, and interface IP—over crowded pure-play GPU names into earnings volatility.

1) Key Value Signals

  • Rotation within AI stack: SoftBank sold its entire Nvidia stake to fund OpenAI—signal of capital migrating from hardware winners to model/inference layers and infra adjacencies. CNBC Daily Open, Startup Ecosystem Canada
  • Hedge funds add AI exposure on dips: Dan Sundheim’s D1 Capital initiated stakes in Nvidia and Meta—buying strength into weakness. CNBC
  • Cash flow durability: CoreWeave highlights six-year depreciation for GPUs, implying higher residual values and longer monetization windows—supportive of ROI for GPU lessors and server OEMs. CNBC
  • Packaging/substrate bottleneck remains a moat: TSMC and the OSAT ecosystem (Amkor, ASE) and substrate providers (Ibiden, AT&S) remain critical to AI GPU/HBM supply—favorable economics and utilization ahead.
  • Fundamental quality outside the hype: TSMC flagged as a top AI infrastructure play with strong growth and elevated free cash generation. Forbes
  • Sentiment reset: Short-term caution with Wells Fargo advising to trim tech profits amid capex/debt concerns—creating opportunity for value entries in under-owned enablers. CNBC

2) Stocks or Startups to Watch

Note: Valuation metrics are approximate “latest available/forward” ranges as of Nov 2025; please verify before investing.

  • Amkor Technology (AMKR) — OSAT leader in advanced packaging

    • Rationale: CoWoS/advanced packaging scarcity leverages pricing power; diversified customer base across AI accelerators and HBM. High operating leverage into AI back-end demand.
    • Approx metrics: P/E 20–23; P/B 3.0–3.5; Debt/Equity 0.3–0.5; FCF TTM ~$0.4–0.6B; PEG 1.2–1.6
    • Value angle: Picks-and-shovels with cyclical upside and relatively modest multiple.

  • Kulicke & Soffa (KLIC) — Advanced packaging equipment

    • Rationale: Beneficiary of HBM/2.5D-3D packaging tool demand; asset-light, net-cash balance sheet; operating leverage as AI back-end capacity expands.
    • Approx metrics: P/E 17–20; P/B 2.2–2.8; Debt/Equity ~0 (net cash); FCF TTM ~$0.15–0.30B; PEG 1.0–1.4
    • Value angle: Low-leverage cyclical with AI-specific tooling tailwinds.

  • Taiwan Semiconductor (TSM) — AI foundry backbone

    • Rationale: Dominant in leading-edge N3/N5 nodes and CoWoS; structural AI demand plus pricing power; large, durable FCF.
    • Approx metrics: P/E 22–26; P/B 5–6; Debt/Equity ~0.3; FCF TTM >$20B; PEG 1.1–1.3
    • Value angle: Quality compounder at a discount to many AI darlings, with superior moat.

  • Micron Technology (MU) — HBM and high-bandwidth memory

    • Rationale: HBM attach rates to accelerators drive a multi-year upcycle; pricing improves mix and margin; operating leverage as utilization tightens.
    • Approx metrics: P/E N/M TTM; forward P/E 14–18; P/B 1.6–2.0; Debt/Equity 0.3–0.4; FCF FY25E ~$3–6B; PEG 1.0–1.5
    • Value angle: Classic memory cycle recovery with secular AI kicker.

  • Ibiden (4062.T) — High-end substrates for AI GPUs/HBM

    • Rationale: BT substrates critical to GPU/HBM packaging; supply remains tight; pricing power and utilization support margins.
    • Approx metrics: P/E 14–18; P/B 1.2–1.8; Debt/Equity 0.1–0.3; FCF positive; PEG 1.1–1.4
    • Value angle: Under-followed enabler with structural demand visibility.

  • Rambus (RMBS) — Memory/PHY IP leveraged to HBM

    • Rationale: High-margin IP model with exposure to HBM interfaces; net cash; expanding content as memory bandwidth scales.
    • Approx metrics: P/E 35–45; P/B 6–8; Debt/Equity ~0; FCF TTM ~$0.3–0.5B; PEG 1.3–1.8
    • Value angle: Not “cheap,” but quality cash generator with secular royalty tailwinds.

Optional watchlist: Alphawave IP (AWE.L) for SerDes/chiplet connectivity (P/E 25–35; P/B 2.5–3.5; low leverage; modest FCF), and AT&S (ATS.VI) for higher-risk/higher-reward substrate capacity ramp (higher leverage).

3) What Smart Money Might Be Acting On

  • Layer rotation: SoftBank’s Nvidia exit to finance OpenAI suggests a maturing trade—capital shifting from crowded GPU leaders into model/inference and infra enablers. CNBC Daily Open
  • Buy-the-dip in platform winners: D1 Capital’s new positions in Nvidia and Meta indicate continued confidence in AI monetization breadth. CNBC
  • Hyperscaler moat reinforcement: Berkshire adding Alphabet underscores durable AI infrastructure economics at the cloud layer—demand driver for accelerators and memory. Fortune
  • Depreciation alpha: GPU residual values (CoreWeave’s 6-year depreciation) extend cash-on-cash payback periods for lessors and server OEMs; investors may rotate to businesses with recurring utilization revenue and slower asset obsolescence. CNBC
  • Picks-and-shovels: Tightness in HBM, substrates, and advanced packaging suggests better pricing/margins and less headline risk than the most crowded AI names; this theme aligns with value discipline.

4) References

5) Investment Hypothesis

  • Overweight: Picks-and-shovels in the AI supply chain (AMKR, KLIC, TSM, MU, IBIDEN). These names combine structural AI demand with more reasonable valuations and cleaner balance sheets than the headline GPU leaders. Residual value evidence for GPUs supports multi-year utilization and orders across packaging/memory/foundry.
  • Select Growth at Reasonable Price: RMBS (IP model with cash-rich balance sheet and HBM leverage) on pullbacks.
  • Underweight/Tactically trade: Crowded GPU leaders into earnings given expectation risk and shifting capital (SoftBank rotation). Maintain a watch with tight risk controls rather than add size pre-print.
  • Risk/Reward:
    • Upside: Multi-year AI capex remains “insatiable” (per AMD), with scarcity premia accruing to back-end and memory. Margin expansion and FCF compounding are achievable at reasonable multiples.
    • Downside: Macro slowdown and capex deferrals could hit cyclical names; supply normalization may compress pricing power; geopolitics can impact foundry/memory.
  • What matters most now:
    • Capacity tightness and pricing in CoWoS, substrates, and HBM.
    • GPU depreciation/resale dynamics extending ROI windows.
    • Hyperscaler spend levels and mix (training vs inference) driving bill of materials for memory and packaging.
    • Balance-sheet strength and FCF conversion to weather volatility.

Signals and Analysis (Include Sources)

  • Wells Fargo urges profit-taking in tech

    • What happened: The bank warned that stretched expectations, heavy AI-related capex, and debt financing concerns raise near-term disappointment risk for AI leaders. CNBC
    • Why it matters: Multiple compression risk for crowded AI bellwethers supports rotating into less expensive, cash-generative enablers (packaging, substrates, memory) with more favorable valuation and free cash flow visibility.

  • D1 Capital buys Nvidia and Meta

    • What happened: Dan Sundheim’s hedge fund opened positions in key AI beneficiaries during Q3. CNBC
    • Why it matters: Despite volatility, institutions are adding AI risk on pullbacks, implying confidence in downstream monetization and continued revenue growth.

  • SoftBank divests Nvidia to fund OpenAI

    • What happened: SoftBank sold its entire Nvidia stake, reallocating proceeds to a stake in OpenAI. CNBC Daily Open, Startup Ecosystem Canada
    • Why it matters: Signals a strategic shift from hardware leaders to model layer equity optionality. For value investors, it underscores crowding risk in GPUs and the merit of under-owned infrastructure enablers.

  • GPU depreciation lengthens to six years (CoreWeave)

    • What happened: CoreWeave’s CEO reiterated six-year depreciation for GPU infrastructure, indicating longer useful life and residual value. CNBC
    • Why it matters: Extends cash-on-cash returns for GPU lessors, supports secondhand markets, and stabilizes order books for server OEMs and component suppliers (HBM, substrates, packaging).

  • AMD flags multi-year AI growth

    • What happened: AMD guided for ~35% annual growth over 3–5 years on “insatiable” AI chip demand, reinforcing MI-series accelerator ramp. CNBC Daily Open
    • Why it matters: A second source of high-performance accelerators broadens demand for HBM and advanced packaging—benefiting MU, AMKR, IBIDEN, and related toolmakers.

  • TSMC highlighted as top AI infrastructure play

    • What happened: TSMC’s strong revenue and net income growth, plus raised 2025 guidance, spotlight its centrality to AI chip supply. Forbes
    • Why it matters: Foundry economics and CoWoS capacity remain bottlenecks; TSMC’s scale and pricing power support robust free cash flow and durable moat.

  • Market positioning into Nvidia earnings

    • What happened: Equities turned choppy as investors debated AI momentum; Nvidia report is the near-term catalyst. CBS News
    • Why it matters: Elevated expectations can trigger volatility. Favor accumulating enablers with lower P/E and higher FCF conversion on dips.

  • Berkshire adds Alphabet

    • What happened: Berkshire Hathaway’s largest new addition was Alphabet, maintaining Amazon—both hyperscaler AI leaders. Fortune
    • Why it matters: Hyperscaler cash flow moats drive sustained accelerator and HBM demand. Validates the secular runway for AI infra vendors.

  • Quantum computing initiative (early-stage)

    • What happened: HPE and chip industry partners, with a Nobel laureate, aim to build a practical quantum supercomputer. iTnews
    • Why it matters: Not an immediate revenue driver for AI accelerators, but a longer-dated disruptive vector; minimal near-term impact on AI chip cash flows.

  • Intel Arrow Lake refresh (client CPUs)

    • What happened: Leaked specs show higher clocks/cores for consumer CPUs. VideoCardz
    • Why it matters: Peripheral to data center accelerators; limited read-through for AI chip cycle near term.

Investment Thesis

  • Action: Buy selectively. Accumulate “picks-and-shovels” (AMKR, KLIC, TSM, MU, IBIDEN) on weakness. Keep RMBS on a buy-the-dip list. Maintain a watch on crowded GPU leaders into earnings; consider trims if multiples stretch versus guide.
  • Risk/Reward: Favorable. Structural demand (training and inference) plus evidence of longer GPU life supports multi-year capex. Back-end capacity scarcity (packaging/substrates/HBM) provides pricing power and utilization. Valuations in selected enablers remain reasonable vs. growth trajectory.
  • Themes to monitor:
    • HBM supply/ASP trajectory and mix shift.
    • CoWoS/advanced packaging lead times and capex announcements.
    • Hyperscaler capex cadence and ROI commentary.
    • GPU secondary market pricing and utilization rates (depreciation evidence).
    • Balance-sheet discipline and FCF conversion vs. growth claims.

In short, let the market argue about top-tick GPU prints; we prefer the durable cash flows in the bottlenecks that make AI accelerators possible.

Back to News