Anthropic cozies up to Elon Musk, IBM makes its case to win in AI, and Cerebras has big IPO plans

Anthropic signed a major deal to use all of SpaceX’s Colossus data center capacity for AI inference, underscoring intense demand for compute amid rapid enterprise AI expansion. The article also highlights IBM’s push to move AI beyond experimentation, a wave of enterprise AI partnerships, and Cerebras planning a $3.5 billion IPO at a $27 billion valuation. Mixed earnings read through for AI-linked names and infrastructure providers suggests continued volatility rather than a clear sector trend.

Analysis

The clearest near-term winner is the infrastructure stack that monetizes AI demand without taking model risk. A large inference commitment to a third-party data center reinforces that compute scarcity is still the binding constraint, which should keep pricing power elevated for network, optics, power, and data-center interconnect vendors over the next 2-4 quarters. That favors names with exposure to scale-out networking and AI facilities buildouts more than the headline model providers, because the incremental dollar of spend still leaks into picks-and-shovels rather than software gross margin expansion.

Enterprise AI is shifting from experimentation to procurement, and that’s where the second-order alpha is. Platform vendors with workflow control points can become the toll collectors if they own identity, orchestration, governance, and observability; the risk is that the market overpays for “AI native” narratives while underestimating how much of the budget stays in incumbent systems of record. In that framing, the strongest beneficiaries are the companies that can sell AI as a replacement for fragmented point solutions, not just as another chatbot layer.

The setup is more mixed for AI infrastructure operators and neo-clouds. When a single customer can absorb a meaningful chunk of capacity, utilization looks strong until it doesn’t; contract concentration and capital intensity mean any renegotiation or demand pause can compress equity quickly. For hardware-linked names, the key watchpoint is whether demand is broadening enough to support multi-quarter capex, or whether this is still a lumpy, deal-driven market that can reverse if inference economics improve or model providers slow their expansion.

The contrarian view is that the market may be too focused on who wins the AI spending race and not enough on margin transfer. If enterprises successfully standardize agent tooling, the value capture could shift away from pure-play application vendors toward incumbent platforms and infrastructure layers. That makes the current leadership in high-multiple AI software vulnerable if next-quarter guidance stops improving faster than spend, while the more boring control-plane names may quietly compound for longer.

Key Decisions for Investors

• Long IBM / short high-multiple AI app basket over 3-6 months: own the control-plane beneficiary versus names where AI monetization is still aspirational; favor a 1.5-2.0x upside on IBM relative to software-beta compression if enterprise budgets consolidate.
• Add on weakness in DDOG and PLTR into the next 2 earnings cycles: both benefit from AI observability and deployment expansion, but only if guidance confirms spend is broadening; risk/reward is attractive on pullbacks because AI attach rates can re-rate revenue growth again.
• Short CRWV on any post-news strength for a 1-3 month horizon: the market is likely to underprice concentration and capex risk until utilization or customer mix deteriorates; use a tight stop if backlog converts faster than expected.
• Overweight AMZN, CSCO, and GLW as a diversified AI infrastructure basket into next quarter: these have more durable second-order exposure to logistics, networking, and optical buildout than single-name AI compute exposure; better downside capture if inference spend remains sticky.
• Neutral-to-short NET on rallies over the next 1-2 quarters: AI narrative is supportive, but the market is already discounting a higher-level platform role; unless enterprise AI control becomes meaningfully differentiated, multiple expansion looks capped versus peers with harder operational leverage.