Starting from Micron's financial report: The next bottleneck of AI (wealth code) Micron's financial report exceeded expectations, indicating a stronger than expected storage cycle. It may also mean that the system level expansion of AI data centers is still ongoing, potentially breaking the previous storage supply and demand cycle. In this case, storage should not be discussed separately. AI data centers are strongly bound systems, rather than independent component combinations In AI architecture: Computing power (GPU) Storage (HBM/DRAM) Connection (Copper/Light/SerDes) The three have a strong binding relationship. The expansion of computing power will inevitably drive storage; Storage expansion inevitably drives connections. Therefore, Micron exceeded expectations and essentially conveyed: System level capital expenditures are still in the upward range. When storage becomes a bottleneck, the system naturally moves towards pooling The current so-called 'insufficient storage' does not mean absolute capacity shortage, but rather: HBM is expensive and scarce Storage is bound to a single GPU card The utilization rate is highly uneven within the cluster Under these conditions, the natural selection at the system level is: Improve storage utilization through pooling/sharing. This is not a conceptual upgrade, but a passive response at the engineering level. Pooling does not eliminate problems, but rather transfers them Once storage changes from 'local' to 'shared', the system architecture undergoes fundamental changes: Original: GPU → Local HBM Now: GPU → Connection → Storage Pool The core constraints that come with it become: Is the delay low enough Is the bandwidth high enough Is the shaking controllable Is the power consumption acceptable The conclusion is clear: Pooling shifts the bottleneck from storage chips to connected systems. Key clarification: Pooling ≠ more connections, but higher difficulty connections Pooling does not mean: The number of cables has significantly increased And it means: The performance requirements for a single connection have significantly increased Specifically manifested as: Higher bandwidth Lower latency Stronger certainty Stricter constraints on signal integrity and power consumption In a pooling architecture, connections go from being a component to becoming a system level critical path. This explains the significant differentiation of connecting companies Therefore, Micron's financial report cannot be simply extrapolated as' all connected companies will benefit '. The strength of the benefit depends on the location of the connection. More directly beneficial connection types Short distance high-speed connection within GPU/cabinet High end SerDes Low latency solutions such as DAC/AEC These connections are between computing power and storage, and cannot be bypassed. Connection types with a later rhythm Cabinet to cabinet, data center level optical interconnect 800G/1.6T optical module It's not that they don't benefit, but rather: The pace of increasing volume is slower More dependent on overall architecture maturity Compress logic into a causal chain Storage scarcity → System attempts pooling → Storage access across nodes Delay and bandwidth become core constraints → Connection upgrade to system bottleneck The value of connections with high technological barriers is increasing summary What Micron's financial report truly strengthens is not just the 'bull market in storage', And it will also drive upstream, downstream, and related sectors: The next bottleneck for AI systems is shifting from computing power to collaborative storage and connectivity capabilities. At this stage, it is often not the storage quantity that determines the upper limit, But rather whether the system can stably and efficiently connect these storage devices together. Finally, have you obtained the wealth code? If not, welcome to join the group discussion. Please find the group QR code at the top Disclaimer: I hold assets mentioned in the tweet, and my views are biased and not investment advice, dyor