Jordan Nanos

Jordan Nanos × Dylan Patel × Jordan Schneider

NVIDIA’s leadership shock has turned alleged chip diversion into an export enforcement crisis, though evidence that Jensen Huang knew remains circumstantial.Patel says U.S.-controlled GPU rentals could preserve a remote cutoff over Chinese compute while earning providers millions per megawatt; Schneider questions whether U.S. frontier labs would gain more value from scarce capacity.The panel supports MATCH restrictions to constrain China’s chipmaking scale, while NVIDIA’s outlook depends on executive retention and Rubin roadmap execution.

Jordan Nanos × Maya Barkin × Wayne Ma × Jeremie Eliahou Ontiveros

Roughly three-quarters of 2027 capacity in the top 20 US states is under construction, supporting the 50-plus GW global forecast; most of 2028’s 62 GW remains unstarted.Texas, nearly 30% of 2028 additions, now includes on-site power in its permit freeze; an Abbott win could allow a December restart, while his Democratic opponent’s conditions could add five to nine months.

Jordan Nanos × Howie × Sherry Zhang

2025 patents point to ASML's unannounced die-to-wafer bonder, several years out, leveraging its metrology and wafer-stage strengths.Logic’s shorter, denser connections offer the clearer efficiency case; HBM’s higher costs and customer requests for fewer layers leave use unlikely before 2029, if at all.Yield still hinges on six-axis alignment, particle control and supplier integration across the conveyor design.

Jordan Nanos × Jeremie Ontiveros × Reyk Knuhtsen

SemiAnalysis raised its 2027 data-center forecast from 43 GW to 50 GW, citing building-level tracking against expectations near 25 GW.More than $1 trillion in signed leases and modular builds, with DayOne timelines around nine months versus 18 months traditionally, support delivery; delays and gradual ramps are already modeled.Power remains constrained, while potential data-center oversupply entering 2028 is Jeremie’s main risk, not a formal forecast.

Jordan Nanos × Jeremie Ontiveros × Reyk Knuhtsen

SemiAnalysis is raising its 2027 data-center forecast from 43GW to roughly 50GW, citing building-by-building tracking, over $1 trillion in signed leases, and self-build starts rising quarter after quarter.Cited modular builds can compress traditional 18-month timelines to 6–9 months, but the team warns the constraint could flip to data-center oversupply in 2028 as chip supply expands.

Kimbo Chen × Wenyao Gao × Jordan Nanos

GLM-5.3’s sparse attention eases bandwidth pressure, not KV-cache capacity needs, while cheaper serving could lift HBM and bandwidth demand.AMD’s MI355 reportedly matched B200 cost at 150 tokens per second per user within weeks, underscoring software-driven cost volatility.Post-training reportedly lifted agentic benchmarks by more than 10%, though GLM-5.3 is not Mythos-class and technical disclosure remains limited.

Alex Kantrowitz × Dylan Patel × Jordan Nanos

SemiAnalysis estimates next-year US AI infrastructure spending near $2 trillion, or 5–6% of a $30-trillion-plus economy, including capacity for overseas inference demand.Hyperscaler capex grew 116% year over year in Q3 2026, but demand exceeds supply; returns depend on six-year GPU and 15-year data-center lives, while neo-cloud security and internal model use remain risks.

Dan Nishball × Jordan Nanos

SemiAnalysis estimates $11.3 trillion in cumulative AI IT and data-center CapEx for 2024-29, implying roughly $7 trillion of debt by 2029 and $1 trillion of net funding this year.Hyperscaler and NVIDIA backstops, five-year offtakes, and frontier API gross margins near 90% could support financing, but lenders still face residual-value, churn, and execution risks.

Jordan Nanos × Abhilash Jain

SemiAnalysis Consulting is turning bespoke AI-infrastructure diligence into scalable products, including a token-supply simulator linking chip deployment, workload allocation, throughput, and token pricing.NeoCloud underwriting centers on contractual survivability: 99.9% availability expectations, sub-90% performance, and mismatched 15–20-year leases versus five-year GPU contracts can threaten financing and trigger remedies, while capacity timelines determine whether clients build or lease.

Jordan Nanos × Cam Quilici × Bryan Shan × Alec Ibarra

Engrams co-design model architecture and memory hierarchy, retrieving 2-gram and 3-gram embeddings from DRAM or SSD.Hybrid Engram-MoE layers outperform either endpoint, while AgentX traces show 100,000 input tokens at P50 and cache-hit rates above 98%.DRAM offload matters for hundreds of agents at 40–60 tokens/s; TPU V7's early Pareto position still leaves software, utilization, pricing, and demand unresolved.