Is the current wave of AI investment a speculative bubble that will correct sharply?

A rough timeline of how the AI investment cycle built up and how the bubble question emerged:

• November 2022: A widely used AI chatbot is released to the public, triggering a surge in commercial interest in large language models.
• 2023–2024: The largest U.S. cloud and platform companies repeatedly raise their capital-spending guidance, citing AI demand. Orders for AI accelerator chips outpace supply.
• June 2024: The leading AI chipmaker briefly becomes the most valuable company on U.S. markets. In the same period, a major investment bank publishes a research report questioning whether projected AI capital spending will generate adequate returns.
• January 2025: One large cloud provider states it plans to spend roughly 80 billion dollars on AI-capable data centers in its 2025 fiscal year.
• 2025: Michigan lawmakers advance legislation exempting qualifying data-center equipment from state sales and use tax; data-center proposals move forward in several Michigan communities.
• August 2025: A university research group reports that the large majority of enterprise generative-AI pilots it studied showed no measurable effect on profit and loss.
• 2026: Dozens of AI startups carry billion-dollar-plus valuations without disclosed profits. Analysts debate whether investment among chipmakers, model developers, and cloud providers is partly "circular" — each booking the others' commitments as demand.

Source: Microsoft on the Issues, "The Golden Opportunity for American AI," January 3, 2025, blogs.microsoft.com

Source: Goldman Sachs Global Investment Research, "Gen AI: Too Much Spend, Too Little Benefit?", June 2024 (research report).

Source: Michigan Legislature, 2025 data-center tax legislation, bill text and history at legislature.mi.gov

The debate is often framed as a single yes-or-no question, but analysts hold at least four distinct positions. The two poles are below; two middle positions follow.

Middle position 1 — partial correction, not full collapse: AI capabilities are real, but near-term revenue projections are overstated. A large drop in AI company valuations is likely, while the underlying technology and infrastructure stay useful. The comparison is the 2001 telecom correction rather than the full dotcom collapse.

Middle position 2 — different layers, different risk: The application layer of AI startups may be in bubble territory while the infrastructure layer of chipmakers, data centers, and cloud providers is more defensible. Treating "AI" as one asset class hides these distinctions.

Source: Goldman Sachs Global Investment Research, "Gen AI: Too Much Spend, Too Little Benefit?", June 2024 (research report).

The strongest version of the argument that current AI investment is a speculative bubble:

• Spending exceeds revenue by a wide margin. Capital outlays for accelerator chips and data centers are running well ahead of the incremental AI revenue the same companies disclose. A bubble forms when investment is justified by expected future demand rather than current results.
• Enterprise returns are hard to find. Independent research on corporate generative-AI pilots reports that the large majority produced no measurable effect on profit and loss, suggesting the technology is not yet paying for itself in typical business use.
• Financing looks circular. Chipmakers invest in model developers that buy their chips; cloud providers count one another's multi-year commitments as demand. This can inflate apparent order books without new end-customer money entering the system.
• Unit economics are unproven. Price competition among model providers and the ongoing cost of running models for each query may keep subscription and usage revenue from covering training and hardware costs.
• Concentration raises the stakes. A small number of firms, and a single dominant chip supplier, account for much of the market value tied to AI. If expectations reset, the re-pricing could be sharp and fast.
• The historical pattern is consistent. The dotcom and telecom cycles both saw infrastructure overbuilt years ahead of usable demand. The technology endured, but many investors and operators did not.

What this would mean for Michigan: if demand projections reset, some announced data-center projects could be paused, scaled back, or built out more slowly, leaving communities with approvals and infrastructure commitments ahead of the jobs and tax revenue that were projected.

Source: university research group report on enterprise generative-AI adoption, August 2025 (study finding roughly 95 percent of enterprise pilots showed no measurable return).

Source: Goldman Sachs Global Investment Research, "Gen AI: Too Much Spend, Too Little Benefit?", June 2024 (research report).

The strongest version of the argument that this is durable infrastructure investment rather than a bubble:

• Revenue is growing fast from a small base. AI-related cloud revenue and model-subscription revenue are rising at high percentage rates. Early-stage markets often show large spending ahead of large revenue before the two converge.
• The assets are durable and reusable. Data centers, electrical interconnections, fiber, and networking equipment hold value across many computing workloads. The dotcom build left fiber and server capacity that produced enormous value in the following decade.
• Capacity is constrained, not demand. Cloud providers report that AI compute is sold out and rationed. That points to real, paid demand rather than speculative accumulation.
• Measured productivity gains exist. Controlled studies show time savings in software development, customer support, and document-heavy work. Even if near-term forecasts are too high, a real long-run return may still be present.
• The spending is balance-sheet funded. Much of the current investment comes from the cash flow of highly profitable incumbents rather than debt or retail speculation, which limits the risk of a broader financial shock if valuations fall.
• Strategic necessity sustains investment. Firms and governments treat AI capacity as competitively essential, which tends to keep investment going through a downturn rather than collapsing it.

What this would mean for Michigan: if the buildout reflects real long-term demand, data centers in the Grand Rapids area and elsewhere in the state would represent long-lived infrastructure, and the 2025 tax incentives would be competing for genuinely durable investment.

Source: Michigan Public Service Commission, utility integrated resource plans and load-forecast filings, michigan.gov/mpsc

Source: quarterly capital-expenditure and segment-revenue disclosures in the public filings of the largest U.S. cloud and platform companies (company 10-K and 10-Q filings).

• ~80 billion dollars: AI-capable data-center spending stated by one large cloud provider for its 2025 fiscal year.
• Above 3 trillion dollars: peak market value reached by the leading AI chipmaker in 2024, among the highest ever recorded for a U.S. company.
• Roughly 95 percent: share of enterprise generative-AI pilots in one 2025 study that showed no measurable profit-and-loss impact.
• ~1 trillion dollars: scale of projected AI capital spending that a 2024 investment-bank report said would need to generate a large, and so far unproven, payoff.
• Dozens: AI startups valued at 1 billion dollars or more as of 2026 without a disclosed path to profit.
• Michigan, 2025: legislation advanced to exempt qualifying data-center equipment from state sales and use tax.
• West Michigan: a large data-center campus has operated in the Grand Rapids area since the late 2010s, and additional projects have been proposed in nearby communities.

Figures are approximate and drawn from company statements, regulatory filings, and published research. Where a company or agency has not released an exact number, the range is described rather than stated precisely.

Source: Microsoft on the Issues, "The Golden Opportunity for American AI," January 3, 2025, blogs.microsoft.com

Source: leading AI chipmaker quarterly results and historical share data, investor relations site (see Source Documents).

Source: Michigan Legislature bill text and analyses, legislature.mi.gov

• Capital-spending guidance: quarterly forecasts from the largest cloud providers. Flat or falling guidance would be the clearest early signal of a turn; continued increases would support the infrastructure case.
• Disclosed AI revenue versus spend: whether company filings begin to break out AI revenue and show it closing the gap with capital outlays.
• Enterprise return studies: whether later research shows corporate AI use moving from pilots to measurable profit.
• Financing structure: a shift toward debt, vendor financing, or off-balance-sheet vehicles to fund data centers would raise the risk profile.
• Michigan grid filings: data-center load in utility integrated resource plans and interconnection queues at the Michigan Public Service Commission, and whether announced projects reach construction or are paused.
• Local decisions: Cascade Charter Township Planning Commission and Township Board action on any data-center site plans, rezonings, or a moratorium.

Source: Michigan Public Service Commission dockets and case filings, michigan.gov/mpsc

Source: Cascade Charter Township meeting agendas, packets, and minutes, cascadetwp.com

This is a national economic question, but the decisions that touch West Michigan happen in rooms residents can attend:

• Attend local meetings. Cascade Charter Township Board and Planning Commission meetings are open to the public. Schedules, agendas, and packets are posted at cascadetwp.com.
• Submit written comment. The Township accepts written public comment on development proposals, including data-center site plans, ahead of scheduled meetings.
• Follow state utility dockets. The Michigan Public Service Commission holds comment periods on utility resource plans that account for data-center electricity demand. Case information is at michigan.gov/mpsc.
• Contact your state legislators. Data-center tax incentives are set in state law. Find your representative and senator and read the relevant bills at legislature.mi.gov.
• Weigh in below. Use the weigh-in on this page to record where you land on the bubble question, and add your reasoning to the discussion.

• Microsoft on the Issues — "The Golden Opportunity for American AI," January 3, 2025 (statement of roughly 80 billion dollars in fiscal-2025 AI data-center spending).
• Nvidia Investor Relations — quarterly results and historical share data for the leading AI chipmaker.
• Goldman Sachs Global Investment Research, "Gen AI: Too Much Spend, Too Little Benefit?", June 2024 — research report questioning whether projected AI capital spending will earn an adequate return.
• University research group report on enterprise generative-AI adoption, August 2025 — study reporting that most enterprise AI pilots showed no measurable profit impact.
• Michigan Legislature — 2025 data-center sales and use tax exemption legislation, bill text, analyses, and voting history.
• Michigan Public Service Commission — utility integrated resource plans and interconnection dockets that reflect data-center electricity demand.
• U.S. Bureau of Economic Analysis — national accounts data on private fixed investment in information-processing equipment and software.
• Cascade Charter Township — Board and Planning Commission agendas, packets, and minutes.

Items without a direct link are named in full so the primary document can be located; the description gives the publisher, title, and date.