This piece asks whether Intel is due for a cyclical recovery or facing a long-term decline. The guest, Todd Ahlsten, thinks the market is too pessimistic, which gives Intel a low bar to clear. If CEO Gelsinger's aggressive technology roadmap (five nodes in four years) works, Intel could become a tech giant again, but execution is a huge risk. Key names: NVIDIA (AI chips are booming, strong moat), AMD (taking Intel's market share in CPUs and GPUs), and TSMC (the benchmark for chipmaking, but faces geopolitical risk).
This episode features Todd Ahlsten, Chief Investment Officer of Parnassus Investments, who has nearly 30 years of experience in semiconductor industry analysis and has lived through eight industry cycles. The main thread of this episode is a deep dive into whether Intel represents a "cyclical recovery" or a "structural decline." The most weighty judgment of the entire program is: Todd Ahlsten believes that due to extremely pessimistic market consensus, Pat Gelsinger's Intel turnaround plan currently faces a "two-foot-high hurdle" (i.e., extremely low market expectations). If the plan succeeds, Intel could join the "Magnificent Seven" within 3–5 years, but this rests entirely on the daunting technological catch-up of "five nodes in four years."
Todd Ahlsten argues that the core of Intel CEO Pat Gelsinger's turnaround strategy is the "five nodes in four years" technology roadmap and a corporate structure split.
1. Technology Roadmap: The plan is to jump across five process nodes (Intel 7 → Intel 4 → Intel 3 → Intel 20A → Intel 18A) in four years. This is not just a catch-up in EUV (extreme ultraviolet lithography) but also encompasses a comprehensive overhaul of next-generation transistor architectures and packaging technologies, including Gate-All-Around and PowerVia (backside power delivery). Ahlsten describes this as "building a plane while flying."
2. Corporate Split: The company is being split into two businesses: design and foundry. The design division focuses on CPUs, while the foundry division (Intel Foundry Solutions) aims to become a global wafer foundry, serving other chip companies.
3. Ecosystem: The key to the foundry business's success is building a complete EDA (electronic design automation) ecosystem to attract customers to port their designs onto Intel's process. The appointment of former Cadence CEO Lip-Bu Tan to the board is seen as a positive signal.
Todd Ahlsten points out that Intel's decline from its peak stems from over-reliance on the PC cash cow, missing the mobile market and EUV technology, and inefficient capital allocation.
1. Missed Mobile: In the mid-2000s, Intel had the opportunity to make chips for devices like Apple's iPhone but declined due to reluctance to dilute its high-margin PC business. This caused it to miss the mobile computing wave.
2. Technology Misjudgment: During the critical transition from 10nm to 7nm, TSMC aggressively embraced EUV, while Intel tried to extend the life of older technologies like "double patterning," causing it to fall behind TSMC in process technology. This triggered a chain reaction: AMD moved its designs to TSMC, putting Intel at a disadvantage in both cost and performance.
3. Capital Waste: The $15 billion acquisition of Altera, $15 billion acquisition of Mobileye, and roughly $9 billion acquisition of McAfee failed to generate expected returns and instead consumed capital. As a result, its PC and server market share fell from nearly 90% to about 80% and 70%, respectively; gross margins dropped 17 percentage points from their peak; and operating margins fell from 30% to 7%.
Todd Ahlsten believes Intel's opportunities lie in the AI-driven GPU market and the "foundry world" demand spurred by geopolitics, but the challenges are execution and customer adoption.
1. GPU Market: Intel's Gaudi series GPUs, acquired through the purchase of Habana Labs, are positioned as "good enough" AI training and inference chips. Ahlsten argues that the key is whether they can get onto the procurement lists of major cloud providers and carve out a share. His milestone: Whether the Gaudi GPU business can reach $3 billion in annual revenue by the end of 2025.
2. Geopolitics: The fragility of the chip supply chain makes Intel a "safety net" for the U.S. The roughly $8–12 billion in subsidies from the CHIPS Act (part of its planned $125 billion in capital expenditure over five years) will significantly reduce its cost disadvantage, giving it a more competitive edge in the "foundry world." Ahlsten notes that giants like Apple, Google, and Nvidia have incentives to find alternative suppliers to TSMC.
3. Risks and Uncertainties: The biggest risk is technology execution. If the five-nodes-in-four-years plan fails, the entire turnaround story collapses. In addition, the shift to customer-centricity is a major challenge; Intel needs to move from "designing for itself" to "serving customers" and build a robust foundry ecosystem. Ahlsten reminds investors to watch the progress of this transformation.
| Instrument | Analyst's View | Key Data |
|---|---|---|
| NVIDIA | Bullish, also holds | The market for its AI chips is enormous; the AI accelerator market could reach $400 billion by 2027; the CUDA ecosystem is a formidable moat. |
| AMD | Neutral, also holds | Continues to gain share from Intel in the PC and server CPU markets; its GPU business is also growing (2024 revenue guidance significantly raised). |
| TSMC | Risk flag, but acknowledges its position | Considered a benchmark for Intel in the foundry space, but faces geopolitical risk; Intel is competing with a 30% cost disadvantage. |
| Mobileye | Call option, no explicit buy/sell | Intel holds over 80% of the shares; market cap is roughly $25–30 billion; a potential asset in the autonomous driving space. |
| Altera (Programmable Solutions Group) | Neutral, potential value unlock | Intel acquired it for $15 billion; currently seeking external capital, may be spun off or IPO'd. |
| Mask Writing Business | Bullish, potential value unlock | Sold a 20% stake at a $4 billion valuation; internal technology viewed as highly valuable; may be independently spun off. |
1. "Two-Foot-High Hurdle": Todd Ahlsten believes market expectations for Intel are extremely pessimistic, providing a low-base advantage for the turnaround plan. Supporting evidence: If Intel can achieve a 25% operating margin in 2026 (corresponding to about $4 in earnings per share), the stock's fair value should be in the $70s, far above the current price of around $47.
2. "Moore's Law Squared": Ahlsten cites Jensen Huang's view that the value of a GPU lies in its combined hardware and software (e.g., CUDA) effect, enabling performance improvements far beyond traditional Moore's Law. Supporting evidence: This explains why Nvidia can command such high valuations and margins, and also highlights the importance of Intel building a software stack for Gaudi.
3. "Cycle Within a Cycle": Ahlsten notes that the current semiconductor cycle is unusual: sub-markets like PC, server, GPU, and memory may be in different phases of the cycle, and the overall duration may be longer. Supporting evidence: AI investment is driven by companies like Meta and Google, which have "infinite" balance sheets, unlike past bubbles fueled by debt financing. This makes the cycle more resilient, with fundamentals expected to improve through 2024–2025.
4. "Control in Semiconductors Belongs to the Shovels, Not the Bread": Ahlsten argues that investing in "shovel" companies like EDA (electronic design automation) and manufacturing equipment (e.g., Cadence, Synopsys) is less cyclical and offers more stable returns than investing in direct GPU or CPU manufacturers. Supporting evidence: Regardless of which chip company ultimately wins, designing and manufacturing chips requires these fundamental tools; their revenue growth is almost independent of downstream chip cycle fluctuations.
5. "AI Investment's Gold Rush and Gold Miners": Ahlsten warns that although the current AI investment frenzy is booming, the market will face a commercialization validation test by mid-to-late 2025, which could lead to a significant correction. Supporting evidence: He cites New Street Research data: if the AI accelerator market reaches $400 billion in 2027, corresponding data center spending would amount to $1.1 trillion. This requires downstream applications (e.g., subscription revenue from AI assistants) to generate meaningful commercial returns, or else an investment bubble will emerge.
6. "Turnarounds Take Time, Patience Is the Biggest Advantage": Ahlsten emphasizes that turnarounds in semiconductor manufacturing typically take years. Pat Gelsinger proposed a "five-year turnaround plan" when he took office in 2021, which requires significant patience from investors. Supporting evidence: From process qualification to customer onboarding to mass production, every step is fraught with technical challenges and lengthy certification cycles; it cannot be achieved overnight.