RockstarMarkets
All news
Markets · Narrative··Updated 2d ago
Part of: Semiconductor Cycle

Semiconductor Capex Cycle Showing No Peak; Data Center Demand Remains Red-Hot

Despite fears of AI capex exhaustion, hyperscaler spending on chips and infrastructure continues unabated. Earnings from NVIDIA, Broadcom, and AI infrastructure startups show no slowdown, keeping semiconductor, equipment, and data center names in sustained rally mode.

R
Rocky AI · RockstarMarkets desk
Synthesised from 8 wires · 28 mentions in the last 24h
Sentiment
+80
Momentum
80
Mentions · 24h
28
Articles · 24h
34
Affected sectors
Related markets

Key facts

  • Cerebras raising IPO price to $150-$160 on AI infrastructure demand surge
  • NVIDIA, AVGO, Broadcom showing zero signs of AI capex slowdown in guidance
  • CoreWeave funding rounds at rising valuations; competing with legacy data centers
  • Memory chip margin projections extended through 2027; no peak consensus yet
  • Semiconductor equipment makers seeing sustained order backlogs

What's happening

Concerns about AI capex 'peaking out' have been thoroughly dismissed by recent earnings and forward guidance. NVIDIA, Broadcom (AVGO), and memory chipmakers are reporting sustained demand from cloud hyperscalers (AWS, Azure, GCP) and emerging AI infrastructure players building private data centers. Cerebras Systems is raising its IPO price range to $150-$160 per share (from $115-$125) on surging investor demand, signaling that AI compute infrastructure is still in a growth phase rather than maturity.

CoreWeave, a GPU-focused data center operator backed by Nvidia, has raised funding at rising valuations and is competing aggressively with legacy colocation providers. Multiple traders and analysts flagged CoreWeave's deal with NVIDIA and other hyperscalers as evidence that the infrastructure cycle has years of runway left. Wells Fargo and other sell-side shops continue to raise price targets on semiconductor equipment makers and chipmakers on the back of strong forward orders.

The memory chip supercycle narrative is fueled by this belief that DRAM and NAND demand will remain elevated through 2027 as AI model training and inference scale globally. However, if hyperscalers suddenly slow capex (due to recession, profitability concerns, or a shift to in-house chips like Google's TPU), memory and foundry utilization could collapse quickly, similar to prior semiconductor cycles. For now, the narrative is that there is no peak; the only debate is how steep the eventual decline will be.

The implication is that semiconductor equipment (ASML, LRCX, KLAC) and high-margin chip manufacturers will sustain elevated earnings through at least mid-2027, supporting continued strength in semiconductor indices (SOXX, SMH, SOXL). If this thesis breaks, mean reversion will be severe.

What to watch next

  • 01NVIDIA Q2 guidance Wed: any commentary on capex cycle duration
  • 02Hyperscaler earnings late May: AWS, Azure, GCP capex guidance trends
  • 03ASML, LRCX earnings: equipment demand signals from foundry utilization
Mention velocity · last 24 hours
Coverage from these sources
Previously on this story

Related coverage

More about $NVDA

Topic hub
Semiconductor Cycle: AI Capex, Memory and the SOX Trade

Live coverage of the AI semiconductor cycle — NVDA, AVGO, AMD, ASML, memory demand, capex run rates and overbought signals.