“It has become a tech-industry truism: Spending too little on chips and other computing infrastructure for artificial intelligence is riskier than spending too much.
As OpenAI Chief Executive Sam Altman recently put it, people can either overinvest and lose money or underinvest and lose revenue.
Or as Meta Platforms chief Mark Zuckerberg said on a call with analysts after it reported earnings last month, AI's promise as a revenue driver meant "we want to make sure that we're not underinvesting."
But investors have begun questioning this logic, fretting that the spending spree might be inflating a bubble that will inevitably pop. Indeed, spending too much can be very bad. Just ask Intel.
The storied American chip maker faced a similar spending decision -- albeit under different circumstances -- not long ago.
Having fallen behind rivals in Asia in the race to make cutting-edge semiconductors, Intel brought in Pat Gelsinger as chief executive in 2021 to engineer a turnaround. Gelsinger laid out a strategy to leapfrog those rivals while also becoming a leading contract chip maker. To succeed, the plan would require Intel to vastly expand its chip-manufacturing capacity.
But chip factories cost tens of billions of dollars each and take years to build. So Intel had to splash out on the manufacturing expansion before the returns on those investments were there.
If Intel waited for demand for its cutting-edge chips to materialize before building, it couldn't capitalize on their success. A similar dilemma faces today's AI players: they believe they can't wait for AI's business potential to be proven before investing.
So Intel plowed ahead. It had the financial chops to do it, too, with a fairly healthy balance sheet and more than $21 billion of free cash flow the prior year. Capital expenditures ballooned from about $14 billion in 2020 to $25 billion in 2022.
"We are overinvesting," Gelsinger said at a Credit Suisse conference in late 2021, calling the move "a very conscious decision to get back in front of it."
That may have been the correct move for Intel in its circumstances then. But technological missteps and changed circumstances in the chip market undercut the effort within a couple of years.
The result has been nothing short of a disaster. Manufacturing projects have been put on hold or canceled. The company has been bleeding cash with free cash flow in negative territory in all but three of its last 14 quarters.
Intel has sold assets to raise money, unloading part of its stake in self-driving car-technology company Mobileye and divesting a majority of its programmable-chip business to private-equity firm Silver Lake. It laid off thousands of employees to save money. The company suspended its dividend. Gelsinger was ousted. It increasingly looks like Intel may need to be split up to get back on track.
With its stock plummeting, the U.S. government took a 10% stake in Intel in August. That gave its shares a boost, but the company is valued now at around $171 billion -- roughly 1/26 the valuation of AI chip giant Nvidia. Intel was worth more than Nvidia five years ago.
A similar -- if not worse -- reckoning could easily be in store for companies that are splashing out on AI, should demand fail to materialize in the way they envision.
Large financial commitments to AI data-center projects might no longer make sense, becoming a burden for tech companies for years to come. If assumptions about the value of AI-related assets and contracts change, write-downs are also inevitable. A lot of startups could fail.
Investors have recently expressed trepidation about returns on big AI investments. That has sent AI-related stocks on a bumpy ride the past few weeks. Things will undoubtedly look a lot gloomier in markets if tech's penchant for overspending becomes a definitive miscalculation.
But Intel's experience also illustrates that companies can muddle through a period of overspending. While Intel has been substantially weakened, it is far from bankrupt.
And the largest of the big-spending tech companies -- Microsoft, Alphabet and Amazon.com -- are in good financial shape, even if their spending has begun to hurt them.
Alphabet, Google's parent, has been particularly shrewd. It has leveraged its search-engine dominance to gain a step on competitors in AI while not breaking the bank with capital spending. Its planned capital expenditures this year amount to only 23% of projected revenue -- well below that of peers.
Other companies that have leveraged up to spend on AI or have few other business ventures are more vulnerable. Those companies -- like Oracle, CoreWeave and to some extent Meta -- might have to make moves like Intel if AI's returns don't end up justifying their investments.” [1]
It is possible that Intel under Gelsinger problems are example of difficulty to create today's factories in all the West: we can throw a lot of money at the wall, hoping that it sticks, but money is not enough.
the struggles at Intel under Pat Gelsinger's leadership from 2021 to late 2024 serve as a stark case study in the broader challenges of resurrecting advanced semiconductor manufacturing in the West (the US, Europe, and allies like Japan).
Despite unprecedented government subsidies via the CHIPS and Science Act in the US and the European Chips Act, the reality is that pouring billions into fab construction—while necessary—often falls short without addressing deeper systemic hurdles like skilled labor shortages, supply chain dependencies, bureaucratic delays, and the sheer complexity of catching up to Asia's decades-long head start.
Money can kickstart the process, but it doesn't magically conjure the ecosystem needed to make it sustainable. Let's break this down with specifics from Intel's experience and the wider Western landscape.
Intel's Fab Fiasco: A Microcosm of Western Woes
Gelsinger returned to Intel as a "messianic" figure, armed with an audacious "IDM 2.0" strategy to reclaim manufacturing leadership. This included a "five nodes in four years" roadmap (Intel 7, 4, 3, 20A, and 18A processes) to leapfrog TSMC and Samsung by 2025, alongside massive fab expansions. He lobbied aggressively for the CHIPS Act, which unlocked $52 billion in US subsidies, and secured deals for plants in Arizona ($20 billion for two fabs), Ohio ($20 billion for two more), and even Europe (e.g., a $20 billion site in Germany).
Yet, by 2024, Intel was hemorrhaging cash—reporting double-digit declines in key divisions like data centers and AI—while fabs lagged. The Ohio project, announced in 2022 with a 2025 production target, was delayed to late 2026 (and potentially 2030) due to "market challenges" and the US government's glacial rollout of grants (none disbursed by early 2024 despite the Act's passage in 2022). Gelsinger himself slammed the delays, warning in 2022 that without faster funding, Intel might pivot to Europe. His ouster in December 2024 came amid activist investor pressure, layoffs of 15,000+ employees, and skepticism over the foundry business's viability—Intel projected losses until 2027 with no major external customers lined up. Even the ambitious 18A node, meant to restore "process leadership" in 2025, has faced R&D hurdles, underscoring how capital alone can't accelerate physics or engineering breakthroughs.
These aren't isolated flubs; they're symptoms of why Western fab-building is so thorny.
Why Money Isn't Enough: Key Barriers in the West
Building a cutting-edge fab costs $10–20 billion and takes 2–3 years (or longer), but the West's ecosystem has atrophied since offshoring boomed in the 1990s. Asia (Taiwan, South Korea, China) now dominates 80%+ of global production, with TSMC alone churning out 50% of advanced chips thanks to scale, subsidies, and integrated supply chains. Here's a snapshot of the roadblocks, drawn from Intel's saga and parallel efforts:
Fabs need thousands of specialized workers (e.g., process engineers, technicians); the West lacks training pipelines after decades of outsourcing. US/EU face a "brain drain"—e.g., only 6% of global semiconductor engineers are in the US vs. 40% in Asia. CHIPS Act funds STEM programs, but scaling takes years; Europe risks falling below 4% global capacity without talent influx.
Regulations, permitting, and subsidy approvals create bottlenecks; geopolitics adds export controls. EU Chips Act (€43B) suffers similar woes—e.g., Germany's €5B TSMC Dresden plant delayed by red tape. US export curbs on China tech slow tool imports, hiking costs 20–30%.
Critical inputs (e.g., rare earths, EUV lithography machines) are Asia-heavy; energy demands are massive (100 MW per fab). China controls 60–70% of rare earths; Europe's auto sector lost 1/3 production in 2021 shortages. New US fabs like TSMC Arizona face power grid strains.
High costs + low initial utilization = losses; demand must follow supply. Western plants risk under-operation without products/startups—e.g., EU's goal of 20% global share by 2030 is ambitious but trails US (aiming 30%) and Asia's subsidies ($47B China fund).
Tensions (US-China, Russia-Ukraine) disrupt flows; reshoring invites tariffs. Europe imports 20% of global chips but makes 9%; acts like CHIPS/EU Chips aim for resilience, but Trump's potential tariffs could hike costs 10–20%.
The Bigger Picture: Throwing Money, Hoping It Sticks
Gelsinger was right about the stakes—"where the fabs are" will define geopolitics for the next 50 years, eclipsing oil. The West's $100B+ in combined CHIPS/EU investments (plus Japan's $11B, Korea's $450B plan) has spurred $400B in private pledges, like TSMC's $100B US expansion. But as Gelsinger noted post-exit, success hinges on filling those fabs, not just building them—requiring R&D leadership (still Taiwan-centric), customer commitments (e.g., AWS chips on Intel 18A), and policy agility.
The verdict? It's possible, but painfully slow. Intel's pivot under new CEO Lip-Bu Tan (focusing on foundry without full spin-off) and EU "Chips Act 2.0" ambitions signal persistence, but without holistic fixes—like faster visas for engineers, streamlined regs, and public-private talent hubs—the West risks becoming a pricey "overseas factory hub" rather than a leader.
Money buys the bricks; execution builds the empire. If anything, Intel proves the West can throw cash at the wall—it just needs better aim to make it stick.
1. Yes, It's Possible to Spend Too Much on AI --- Tech giants are betting they won't end up like Intel by overinvesting in computing infrastructure. Fitch, Asa. Wall Street Journal, Eastern edition; New York, N.Y.. 28 Nov 2025: B10.
1. Yes, It's Possible to Spend Too Much on AI --- Tech giants are betting they won't end up like Intel by overinvesting in computing infrastructure. Fitch, Asa. Wall Street Journal, Eastern edition; New York, N.Y.. 28 Nov 2025: B10.
Komentarų nėra:
Rašyti komentarą