Signal Over Noise
A live map of the AI build-out — where the capital is actually flowing, and where the story is getting ahead of the fundamentals.
Global Indices
Supply Chain Heatmap
Today's Focus
Overnight US → Next Session
NVIDIA and Broadcom set the pricing anchor for the whole accelerator complex. A strong session in this pair is one of the more reliable next-session tells for the chip-equipment and test names that supply into the same build-out.
Microsoft, Amazon and Alphabet's capex commentary is the single most-watched signal for the power and cooling infrastructure names that build out the data centres behind that spending.
Supermicro and Dell set the tone for AI-server build volume. Networking and interconnect suppliers into the same racks tend to trade on the same signal a session later.
Supply Chain Map
The AI Hype Trends
The Memory Supercycle: How HBM Reallocation Is Squeezing Conventional DRAM
HBM's share of DRAM wafer output has roughly doubled in eighteen months. That reallocation, not new demand alone, is what's driving conventional DRAM pricing power.
The mechanism, in one sentence
Every wafer redirected to HBM production is a wafer that isn't making conventional DDR5, and HBM's die size and yield profile mean it consumes disproportionately more wafer capacity per bit than the memory it's displacing — so a modest shift in AI-driven HBM demand produces an outsized squeeze on general-purpose memory supply.
Why this cycle looks different from past memory cycles
Historically, memory cycles were driven by demand swings against a relatively fixed supply curve. This one is different: supply itself is being actively reallocated by producers chasing HBM's higher margins, which makes the conventional-DRAM shortage a byproduct of AI investment rather than a standalone demand story — and harder to resolve with a simple capacity add, since new capacity is being built HBM-first.
What could break the cycle
The clearest risk to the thesis is a faster-than-expected ramp in HBM4 yields freeing up capacity for conventional memory again, or a slowdown in accelerator unit growth reducing HBM pull. Neither shows up in supplier guidance yet, but both are the things worth tracking each earnings season rather than taking current pricing power as structural and permanent.