Most Investors Are Chasing Flashy IPOs. CoreWeave Is the One I Predict Will Outperform by 2028.

CoreWeave reported a $104 billion revenue backlog at the end of Q2, plus $25 billion of additional net customer commitments early in Q3, supported by expanding demand from Meta, Anthropic, Jane Street, Caterpillar, and other enterprise customers. Contracted power capacity increased to 4.2GW by August from 3.7GW at Q2-end, while its Nvidia partnership—backed by a $2 billion Nvidia investment earlier in 2026—targets more than 5GW of AI infrastructure by 2030. The company is broadening beyond GPU cloud rentals into inference, agentic AI and enterprise implementation services, strengthening its longer-term AI-platform proposition.

Analysis

CRWV's strategic upside is not simply incremental GPU leasing; owning the software/workflow layer can raise switching costs and improve revenue durability versus commodity-like capacity brokers. The near-term issue is that this broadened stack also raises execution risk: field engineering and enterprise deployments are labor-intensive, lower-margin initially, and may not scale at the same pace as infrastructure commitments. Investors should require evidence that non-training workloads improve utilization and gross-profit dollars per deployed megawatt rather than merely expand the sales narrative.

The central equity risk is a duration mismatch: customer commitments support demand visibility, but the company must finance and energize capacity before associated revenue and cash conversion arrive. Over the next 1-3 months, CRWV will trade on financing terms, power-delivery milestones, and customer concentration disclosures more than backlog headlines; higher rates, delayed interconnects, or GPU delivery slippage would pressure both FCF and the valuation multiple. NVDA benefits from accelerated systems pull-through, but its economic exposure is materially cleaner because it sells equipment upfront while CRWV retains multi-year utilization, power-price, and credit risk.

Contrarian view: the market may be over-crediting the backlog as equivalent to contracted cash flow. Hyperscalers and frontier labs can be sophisticated counterparties, but their demand can migrate to self-build or competing capacity when supply loosens; enterprise customers may also consume inference more unevenly than training. Conversely, if capacity remains structurally scarce through 2027, CRWV's contracted power portfolio becomes an option on AI demand, and the stock could re-rate sharply once it demonstrates utilization, customer diversification, and debt-funded buildout economics together.

CAT and BSY are secondary beneficiaries only if physical-AI deployments translate into funded customer projects, not pilots. CAT's opportunity is higher attachment of autonomy, fleet-management, and service revenue; BSY gains if digital-twin workflows become standard enterprise inputs. Neither should be treated as a clean proxy for CRWV's infrastructure economics.

Key Decisions for Investors

• Maintain a watch, not a core long, in CRWV until the next results confirm active-capacity commissioning, utilization, gross margin, capex per MW, and incremental financing cost. Initiate only after those metrics show capacity growth without a worsening FCF/debt trajectory; a customer cancellation, delayed power delivery, or weaker utilization would falsify the thesis.
• Prefer long NVDA versus CRWV over the next 6-12 months: NVDA captures system demand with superior margins and materially less balance-sheet duration risk. Use the pair as a hedge against AI infrastructure demand remaining strong while neocloud financing or execution disappoints.
• For a higher-risk tactical CRWV position, use defined-risk call structures 6-9 months out only after an earnings release that quantifies backlog conversion and customer concentration. Avoid unhedged common-stock exposure ahead of funding announcements; dilution or expensive secured debt is the most likely adverse catalyst.
• Monitor CAT and BSY for 6-18 month confirmation through disclosed AI-enabled product bookings, recurring software attach rates, and margin contribution. Treat enterprise physical-AI announcements without measurable deployment or ARR metrics as non-investable marketing signals.