Colony: AI Investment Will Continue Despite Tech Selloff

Forrester founder, chairman and CEO George Colony says the AI boom may be entering a new phase, with custom chips potentially challenging Nvidia’s dominance and value shifting toward software. He cautions that the transition will not be smooth and that companies must adapt to AI computing or risk being left behind.

Analysis

Economics, not a near-term displacement call

The key risk to NVIDIA is erosion of pricing power, not necessarily a collapse in accelerator demand. Hyperscaler-designed chips can divert predictable, high-volume workloads from merchant GPUs while leaving NVIDIA better positioned in workloads where flexibility, software maturity, and rapid model iteration matter. That substitution could pressure future unit economics before it materially reduces total AI infrastructure spending.

The second-order effect is that custom silicon still draws on external design, foundry, and advanced-packaging capacity; it may redistribute value across the supply chain rather than remove it. Software is a less certain beneficiary: lower inference costs could expand usage, but customers may capture the savings, and software vendors still need demonstrable paid adoption to convert AI features into durable revenue.

Time horizon: an interview alone is unlikely to change near-term earnings expectations. Over 1–3 months, watch hyperscaler capex commentary and evidence of production deployment—not chip announcements. Over 6–18 months, track workload share, NVIDIA data-center growth and guidance, and whether software firms disclose incremental AI monetization. The contrarian risk is treating custom chips as a clean NVIDIA substitute: fragmented tooling and workload-specific economics could slow migration. The thesis weakens if NVIDIA sustains guidance and pricing while custom-chip announcements fail to translate into production workloads.

Key Decisions for Investors

• No trade on this interview alone; its information content is low and the structural substitution thesis needs deployment and financial evidence.
• Watch NVIDIA’s next guidance and data-center commentary for evidence of pricing or workload-share deterioration. A guidance cut or sustained deceleration alongside verified customer migration would strengthen a relative short thesis; resilient guidance and continued broad demand would falsify it.
• For a 6–18 month relative-value setup, consider a small, defined-risk underweight in NVIDIA against diversified software exposure only after paid AI adoption appears in reported results. Do not assume software captures inference-cost savings without revenue or retention evidence.
• Monitor hyperscaler capex, custom-chip production deployments, and foundry/advanced-packaging constraints: rising custom-chip use with stable aggregate capex points to share redistribution, while falling capex would signal a broader demand risk.