Executive summary
BluSky AI, Inc. (BSAI)—formerly Inception Mining—has posted a parabolic year-to-date advance following a rapid pivot to “AI.” That price action is a classic late-cycle tell: narrative first, fundamentals later. At the same time, industry-wide AI-infrastructure capex has entered a super-cycle, pulling forward years of spend into today’s budgets. While AI is genuinely useful, the scale, timing, and financing of this build-out increasingly resemble prior bubbles. The closest rhyme is the late-1990s/early-2000s fiber overbuild transformative technology paired with a boom-bust investment cycle once supply outran monetizable demand (a well-worn script throughout historical innovation cycles). Our stance is unchanged: acknowledge AI’s utility, but allocate only where priced usage, durable margins, and realistic hurdle-rate paybacks convert capital into cash.
Case study: BluSky AI as a speculative signal
In 2025, Inception Mining rebranded to BluSky AI and pivoted from precious metals mining toward modular data-center infrastructure for AI. Soon after, the stock rocketed 1,345% year-to-date in 2025 at the time of writing despite a limited history of contracted, bankable cash flows. Even the company’s web domain, bluskyaidatacenters.com, literally pairs the cycle’s two hottest nouns, “AI” and “data centers,” in the URL. That branding choice is not determinative, but it is emblematic of momentum-chasing in late cycles.
Consider the ticker itself: BSAI—an unfortunate coincidence that reads as “B.S. AI.” Prior to March 10, 2025, shares traded near $0.0007 (about seven hundredths of a cent). The 1-for-1,000 reverse split and rebrand improved surface optics while leaving fundamentals untouched, a typical late-cycle behavior that often precedes dilution.
Reality check: as of June 30, 2025, public data shows that BSAI had only $889 in cash, $32,802 in total assets, and no revenue for the past four years. Yet its current market cap is $144 million, a significant gap between its fundamentals and the hype.
The AI capex super-cycle: scale, timing, reflexivity
AI infrastructure spending is setting records. Hyperscalers and platforms are committing unprecedented sums to data centers, GPUs, networking, and power. Spend is front-loaded: management teams are racing to secure land, interconnects, and equipment before competitors do. The procurement impulse is reflexive—visibility of rivals’ orders forces even cautious players to accelerate, regardless of near-term monetization. Equity enthusiasm lowers the cost of capital and encourages marginal projects that would otherwise fail to meet hurdle rates. Vendors, lessors, and utilities are all incentivized to build. That is how gluts form in capital-intensive systems: minor forecasting errors compounded across many balance sheets. If AI revenue ramps more slowly than capex commitments, returns compress, and capacity becomes a tax on future free cash flow rather than a tailwind. On some estimates, providers are investing roughly $4–$6 of capex for each $1 of AI revenue in the near term, a ratio that must fall for value creation.
Call-Out: Oracle–OpenAI and the Scale of Commitments
Recent reporting indicates that OpenAI has agreed to procure approximately $300 billion of Oracle compute over five years, starting in 2027, tied to a multi-gigawatt (≈4.5 GW) power build-out. If executed, this would be among the most significant cloud commitments on record—a vivid example of front-loaded infrastructure bets. The investment case hinges on durable, priced usage that recoups depreciation, energy, and networking costs. This is precisely the kind of mega-commitment that can amplify the capital cycle: if monetization lags, negative operating leverage first appears in customers—and then in suppliers. For more of our market takes, see past HCM blogs.
A Parallel Tell: CoreWeave’s Fragile Economics
CoreWeave—an AI “neocloud” that rents GPUs under multi-year contracts—illustrates how froth can extend to supposedly “infrastructure” names. Recent analysis highlights growth driven by asset-backed debt and vendor financing, heavy customer concentration (notably Microsoft/OpenAI), discount pricing versus hyperscalers, and limited proprietary IP. The model resembles a levered GPU-rental scheme whose unit economics depend on optimistic utilization, residual values for rapidly obsolescing chips, and a low cost of capital. As hyperscalers internalize compute and workloads shift toward cheaper inference hardware, that model faces compression—classic bubble mechanics as debt-funded capacity chases narrative demand. For our philosophy on balancing narrative and numbers, read Exploring the Reasons Why Value Investing Works.
A history rhyme: the fiber overbuild
Technology can be transformative, whereas the investment cycle can be destructive. In the late 1990s, carriers laid staggering miles of fiber, anticipating exponential traffic. Utilization proved to be a fraction of expectations, bandwidth prices collapsed, and highly leveraged players failed—Global Crossing’s 2002 bankruptcy is emblematic. Years later, that “excess” capacity became the backbone of broadband—but equity had been wiped out. The mechanics were straightforward: optimistic demand curves drove simultaneous capacity additions, unit prices fell, leverage turned from friend to foe, and equity absorbed the hit. The lesson today is not “AI is a fad”—it isn’t. The lesson is that usefulness doesn’t immunize investors from the capital cycle when supply outruns bankable demand.
“This time is different”—and why that may not save equity
There are fundamental differences from 2000. AI already generates enterprise value in select workflows; inference workloads are growing; and software vendors are shipping AI-native features, demonstrating a visible willingness to pay. Power constraints and permitting may slow supply in specific geographies. Yet the hard constraints remain: the cash math of depreciation, energy, networking, and labor must be recouped via priced usage with proper margins after customer-acquisition costs and model-update churn. Enterprise adoption typically follows budgeting cycles; CFOs require proof of durable productivity gains before committing to multi-year spending. Vendor concentration introduces correlation risk if a cohort delays capital expenditures in 2026–2027.
Bubble diagnostics (2025)
- Narrative outruns numbers: outsized price reactions to name/ticker changes and vague “AI adjacency.”
- Mega-contracts whose economics rely on future revenue streams rather than current cash flow.
- Sector investment growth is far outpacing sector operating cash growth.
- Vendor-financed or lease-heavy expansion that pushes cash costs forward.
- Monetization gap: user excitement without priced usage at scale.
- Uniform positioning: the same long-AI trade across managers and balance sheets.
Bottom line: bubbles don’t die of skepticism; they die of cash-flow math. When monetization lags investment, operating leverage turns against the builders and equity becomes the shock absorber. In prior cycles, stories that looked like BluSky AI were far more likely to end in serial dilution or effective zeros for common shareholders than in durable cash generation.
As contrarian value investors, we fund cash, not fashion: priced usage, durable margins, and payback at sober hurdle rates. Everything else is narrative we are content to leave to others. It is no surprise that episodes like BluSky AI revive the old quip that “the stock market is a casino”; we disagree—over time, prices follow cash flows, which is precisely where disciplined value investing finds its edge.
Thank you for reading. As always, we welcome your questions.
Discover more about Hoover Capital Management and explore our Global Opportunities strategy. For inquiries, please contact us. View our disclosures.