The hottest Datacenters Substack posts right now

Advanced 3nm (TSMC N3) wafer capacity is deeply constrained because most leading AI accelerators are moving to N3, so compute deployments are bottlenecked and TSMC is prioritizing AI customers which may push others to diversify to Samsung or Intel.
Memory is the next big bottleneck: HBM demand is surging, it consumes far more wafer capacity per bit than commodity DRAM, and higher HBM pin-speed requirements plus rising DRAM prices mean suppliers will struggle to meet accelerator needs without charging premiums.
A small release valve exists if smartphone demand falls (freeing some N3 wafers) and CoWoS packaging constraints are easing, but memory, datacenter power, and packaging limits mean hyperscalers’ higher capex won’t immediately solve the compute shortage.

Water use by datacenters is often overstated when reported without context; cooling architecture, power source, location, and whether you count direct vs. embedded water all hugely change the footprint.
A concrete comparison shows a 400 MW datacenter can use ~346 million gallons/year while an average In-N-Out store uses ~147 million gallons/year, so that datacenter is roughly equivalent to 2.5 burger joints and can produce billions of tokens per burger of water footprint.
Mitigations and accounting matter: hybrid dry/adiabatic cooling, power choices, chip-manufacturing impacts, and onsite water recycling can greatly reduce net blue-water use, and standardized water accounting is needed for fair comparisons.

When governments label tech firms as national security risks for refusing certain military uses, it creates political loyalty tests that scare off investors and can slow innovation.
Multiple breakthrough technologies—AI/AGI, nuclear, quantum, genomics, and space—are accelerating at once and driving a global race for economic and strategic leadership.
That rapid progress brings real risks: geopolitical shocks can disrupt chip and supply chains, data centers raise energy and inflation concerns, and job losses and public backlash are growing policy challenges.

Datacenters are crucial for AI and require significant power. As demand for AI grows, datacenters must adapt to handle higher power loads efficiently.
New designs and standards are emerging in the datacenter industry. For example, Nvidia's new hardware needs liquid cooling and high power densities, which older designs can't support.
Companies like Meta are making big changes to remain competitive. They scrapped older datacenters to build new ones that can handle greater energy demands and performance requirements.

Nvidia's GB200 NVL72 poses an existential threat to competitors in the datacenter CPU market, offering a unique 2:1 ratio and improved integration perspective.
The reintroduction of mainframes signifies a strategic move by Nvidia, with the new AI mainframe/appliance providing massive TCO advantages and performance gains.
Jensen's benevolent trade offer complements technical excellence in the face of political challenges, aiming to secure Nvidia's position in the market and potentially disrupt the status quo.

Get a weekly roundup of the best Substack posts, by hacker news affinity:

Altman is seeking an unprecedented $7 trillion to invest in AI infrastructure, which includes developing GPUs, energy supply improvement, and expanding data center capacity.
The $7 trillion investment is meant to propel technological advancements to a level comparable to the impact of the Industrial Revolution, focusing on long-term projects over decades rather than immediate outcomes.
Despite the astronomical sum, the $7 trillion investment may not seem as excessive considering the potential growth of the global economy and the transformative nature of the projects Altman aims to support.

Debate continues on the need for a standard protocol for request/response paradigm beyond TCP and UDP.
The Internet has extensively used RPC for distinct purposes like email, network management, and name resolution.
Comparisons between TCP and RPC in datacenters highlight the need for a specialized transport protocol optimized for request/reply workloads.

IC-packaging is a crucial element in the semiconductor industry where chips are protected in enclosures for specific applications.
Advanced packaging is crucial for complex chips like server chips or AI processing units to save space and reduce latency between components.
Generative AI, datacenters, and edge computing benefit from advanced packaging to handle large amounts of data efficiently, despite challenges such as high power consumption.