네비우스가 망하는 시나리오

Most likely failure scenario

If application-layer software gets commoditized by tools like Claude Code, Cursor, and agent workflows, then AI-native companies and leading enterprises may choose to own more of their software infrastructure themselves. In that world, a neocloud like $NBIS may have less bargaining power than investors expect. Bare-metal providers like IREN could end up better positioned.

At the same time, GPUs depreciate over time, while AI compute prices and token prices likely fall over the long run. Demand can still grow while unit economics deteriorate. If pricing falls faster than demand rises, NBIS may grow revenue without generating much real profit.

Full bear case against investing in NBIS

1. The moat may be structurally weak

• NBIS's core asset is still a GPU cluster plus data center capacity.

• That is closer to an infrastructure business than a software platform with deep switching costs.

• If customers can move based on performance, availability, and price, long-term excess returns are hard to protect.

• The key question is whether customers use NBIS because it is uniquely valuable, or simply because capacity is scarce right now.

2. AI compute may become commoditized

• GPU instances should get cheaper over time as more supply comes online.

• Token pricing is also under structural pressure from competition and efficiency gains.

• Even if AI demand grows massively, providers still lose if supply growth and price compression outpace demand growth.

• That creates the classic infrastructure trap: strong top-line growth without durable value creation.

3. GPU depreciation may be more brutal than people think

• GPUs depreciate much faster than real estate or traditional server infrastructure.

• One generation behind can already be enough to reduce pricing power.

• That forces operators to reinvest sooner than expected.

• Faster depreciation pressures both accounting profits and real cash returns.

4. Long-term competition with hyperscalers is unfavorable

• AWS, Azure, and Google have lower capital costs, stronger customer lock-in, and broader product bundles.

• They do not just sell compute. They bundle storage, security, databases, networking, deployment, and observability.

• That makes it hard for a standalone AI infrastructure provider to defend pricing in the long run.

• If custom silicon improves, the relative value of GPU-focused neoclouds could fall even further.

5. Customer concentration could break the story quickly

• Much of the NBIS thesis relies on large contracts and a few major counterparties.

• If a customer like Meta or Microsoft delays, renegotiates, or reduces demand, the growth story can crack fast.

• Businesses like this look amazing while anchor customers stay committed, but markets get ruthless once concentration risk becomes visible.

• Large customers also tend to keep most of the negotiating power, which means they may capture the economics instead of NBIS.

6. It is a capital-intensive business where mistakes are expensive

• Data centers, power, land, cooling, networking, and GPUs all require major upfront spending.

• If demand forecasts are even slightly wrong, fixed costs can crush profitability.

• A software company misses growth and gets repriced. A capital-heavy infra company misses growth and can damage the balance sheet.

• In other words, the upside is driven by contracts, but the downside is driven by capital structure.

7. Dilution and external financing may continue

• High-growth infrastructure companies almost always need more capital.

• If internal cash flow is not enough, they turn to equity issuance, convertible debt, or more leverage.

• That means the business can get bigger while per-share value does not improve much.

• If the market turns before financing is complete, dilution can become especially painful.

8. This may just be a scarcity window, not a permanent advantage

• Infrastructure cycles often follow the same pattern: shortage, aggressive buildout, oversupply, margin collapse.

• If NBIS looks great mainly because GPU supply is tight today, then its premium may disappear once supply catches up.

• Markets also tend to price that shift before earnings visibly deteriorate.

• That creates the dangerous setup where fundamentals still look strong while the stock starts breaking first.

9. Without software/platform revenue, the multiple may compress

• Investors want to treat NBIS as more than a hosting or rental business.

• But that only works if NBIS proves real platform, workflow, tooling, or software-layer stickiness.

• If it remains mostly a GPU landlord, the market may steadily lower the valuation multiple.

• A meaningful part of the current thesis depends on NBIS becoming a better business than it is today.

10. NVIDIA dependence is also a hidden risk

• NVIDIA alignment looks like a strength, but it is also dependency risk.

• GPU allocation, product roadmap changes, pricing, or partner policy shifts are outside NBIS control.

• Over time, customers may also work harder to reduce NVIDIA dependence through custom silicon or alternative architectures.

• If that happens, part of the NBIS thesis weakens at the source.

11. Execution risk is substantial

• Power access, permits, construction delays, network bottlenecks, GPU delivery timing, and cluster ramp issues all matter.

• Winning contracts and delivering profitable capacity on time are not the same thing.

• Small operational problems can quickly become revenue delays, cost overruns, and trust issues.

• When a stock is priced for strong execution, even minor misses can be punished hard.

12. The sovereign AI angle may be overstated

• European data sovereignty is an attractive narrative, but narratives do not automatically create high-margin businesses.

• Regulation can help demand, but it can also slow procurement cycles and deployment.

• There is still a big gap between a compelling geopolitical story and a durable economic moat.

• A good story is not always a good business.

13. Subsidiary value may distract from core business risk

• Assets like ClickHouse, Avride, Toloka, and TripleTen can make the total story look more attractive.

• But if the core AI infrastructure business turns out to be structurally weak, those assets do not solve the core problem.

• In some cases, sum-of-the-parts framing can mask how fragile the main business really is.

• When the market loses confidence in the core, it often discounts everything together.

14. This may be a trade, not a long-term compounder

• Even the best version of the thesis may simply be a multi-year scarcity trade.

• If so, owning it like a forever compounder is a category mistake.

• Investors may be tempted to frame NBIS as the "Amazon of AI infrastructure," but it may end up being repriced as a cyclical capital-heavy provider instead.

• That kind of rerating can destroy returns even if the business keeps growing.

Bottom line

The core bear argument is simple: if NBIS looks attractive mainly because of a temporary GPU shortage rather than a durable moat, then the company may end up trapped by high capex, fast depreciation, weak switching costs, customer concentration, and long-term pressure from hyperscalers.

What would weaken this bear case

• A much broader customer base

• Proof that long-term contracts translate into strong gross margins

• Meaningful software/platform revenue that increases stickiness

• High utilization that holds even after major capacity expansion

• Evidence of pricing power even during a looser supply environment