- Future of Computing Newsletter
- Posts
- š„µš„ Data Centers Are Melting the Grid. Here's What Comes Next.
š„µš„ Data Centers Are Melting the Grid. Here's What Comes Next.
A Newsletter for Entrepreneurs, Investors, and Computing Geeks
Happy Monday! This weekās deep dive looks at data center cooling. As power densities increase and chips get hotter, managing heat is becoming a key constraint for AI infrastructure. We explore the technologies aiming to solve it.
In our spotlights, we cover SpiNNcloudās neuromorphic supercomputer deployment in Leipzig and Kleiner Perkins-backed Ambiqās small cap IPO debut.
We also highlight key headlines across AI, semiconductors, quantum, neuromorphic, photonics, and data centers, along with curated readings on smarter chips, long-range qubit coupling, and global data center capacity. Funding news was dominated by OpenAIās $8.3 billion round, alongside several notable Series A to C deals
In our bonus sections (yes, two this time), we take a closer look at Metaās superintelligence ambitions and Germanyās updated Hightech Agenda, which may be worth a glance for some founders.
Deep Dive: Data Centers Are Melting the Grid. Here's What Comes Next.
As power densities rise and system architectures grow more complex, chips run hotter, making cooling a critical constraint. It already accounts for up to 45% of a data centerās energy use and could reach 5% of global electricity consumption by 2030.
In this deep dive, we break down three dimensions shaping the future of thermal management: 1) Cooling architectures, 2) cooling mechanisms, and 3) an emerging strategy beyond cooling.
Cooling Architectures
Direct-to-Chip Cooling
This is one of the most widely adopted solutions in data centers today. It involves placing a cold plate directly on the chip package. Liquid coolant circulates through the plate, absorbing heat and carrying it away.
Where itās used: e.g. AI servers, HPC clusters, and systems where air cooling no longer suffices
Why it works: Efficient heat transfer with relatively low infrastructure change
Limitations: Canāt always target specific die hotspots; performance depends on coolant flow rate and material conductivity
On-Chip Cooling
This method tackles heat at the source, directly on or within the silicon. Microchannels, vapor chambers, or embedded cooling structures extract heat from hotspots at the transistor or die level.
Where itās used: e.g. ultra high power density chips, 3-D stacked chips, or AI accelerators to cool at the heat source
Why it works: Targets the most critical thermal zones with high precision
Limitations: Requires custom chip packaging and is not yet standard in data center hardware
Immersion Cooling
In immersion systems, entire servers are submerged in a non-conductive liquid. Heat dissipates through contact with the fluid, which may remain liquid (single-phase) or evaporate and condense (two-phase).
Where itās used: Densely packed compute clusters, edge data centers, and environments where airflow is constrained
Why it works: Uniform cooling across all components; can reduce energy use and noise
Limitations: Requires new workflows, server designs, and liquid handling infrastructure
Cooling Mechanisms
Single-Phase Cooling
In single-phase systems, the coolant remains in the same physical state (liquid) throughout the entire heat removal process. Heat is absorbed as the fluid passes over or through hot components, but no evaporation occurs.
Where itās used: Most current data centers, especially with direct-to-chip systems
Why it works: Simpler to control and integrate, avoids managing vapor behavior
Limitations: Lower heat transfer efficiency than two-phase; larger flow rates may be needed to compensate
Two-Phase Cooling
In two-phase systems, the coolant evaporates as it absorbs heat, turning into vapor. That vapor is then condensed back into liquid elsewhere in the system. The phase change dramatically improves heat transfer.
Where itās used: Advanced cooling setups, including some immersion systems and advanced cold plates
Why it works: Phase change allows more efficient heat removal, especially at high power densities
Limitations: Requires careful control of bubble dynamics and pressure; harder to standardize and integrate
Beyond Cooling: Reusing Waste Heat
Most cooling systems focus on removing heat, but increasingly, thereās interest in reusing it and thereby turning thermal loss into value. Two-phase and immersion systems are especially promising here, as they output higher temperatures suitable for district heating or industrial use. Adoption, however, remains limited by infrastructure complexity.
If youāre curious how new materials can unlock more efficient heat transfer and make advanced cooling systems viable at scale, read our interview with Apheros co-founder and CEO Julia Carpenter.
Spotlights
š§ SpiNNcloud to deploy neuromorphic supercomputer at Leipzig University (Data Center Dynamics)
āGerman neuromorphic supercomputing company SpiNNcloud will deliver a brain-inspired supercomputer to Leipzig University. The system, which will be used to support research into drug discovery, will comprise 656,640 cores and simulate a minimum of 650 million neurons for AI, HPC, and other applications, making it the largest SpiNNcloud Server System to be deployed to-date.ā
Keep an eye out for our upcoming interview with SpiNNcloud on the Future of Computing blog!
𦾠Kleiner Perkinsābacked Ambiq pops on IPO debut (TechCrunch)
āAmbiq Micro, a 15-year-old manufacturer of energy-efficient chips for wearable and medical devices, closed its first day of trading on Wednesday at $38.53 a share, a 61% increase from the $24 IPO price the company set the previous day.ā
Relevant to investors assessing exit opportunities: āThe success of the IPO signals strong investor demand in the public market for new small-cap companies benefiting from AI innovation. [ā¦] Ambiq closed its first day as a public company with a valuation of $656 million (excluding employee options).ā
Headlines
Last weekās headlines span billion-dollar chip deals, progress in quantum control systems, and emerging applications in neuromorphic and photonic tech.
š¤ AI
𦾠Semiconductors
Infineon Technologies, NXP and STMicroelectronics face rising competition in $132āÆB automotive semiconductor race (Yole Group)
āļø Quantum Computing
A new openāsource program for quantum physics helps researchers obtain results in record time (Phys.org)
CERN Researchers Demonstrate Antimatter Qubit, but Maybe Donāt Expect That Antimatter Quantum Computer Just Yet (The Quantum Insider)
Newly Patented Cooling Tech Promises Cheaper, Simpler Access to SubāKelvin Temperatures (The Quantum Insider)
Next defense departmentās Xā37B mission to carry quantum sensor, laser link experiments (Breaking Defense)
š§ Neuromorphic Computing
āTheyāre Watching Us Thinkā: Panic Erupts as BraināInspired Vision Tech Gives Robots and EVs HumanāLike Perception in Real Time (Caernarfon Herald)
Braināinspired AI breakthrough: Machines learn to see smarter (The Brighter Side)
ā”ļø Photonic / Optical Computing
Xanadu and HyperLight Unveil Advances in Photonic Chips, Setting New Quantum Computing Benchmarks (The Quantum Insider)
š„ Data Centers
Meta to spend up toāÆ$72āÆB on AI infrastructure ināÆ2025 as compute arms race escalates (TechCrunch)
AI Data Centers Market Size and Forecast 2025 to 2034 (Precedence Research)
Readings
This weekās reading list explores AI-driven semiconductor innovation, new models of quantum connectivity, brain-inspired machine learning, and global data center capacity.
𦾠Semiconductors
The New Chips Designed to Solve AIās Energy Problem (Wall Street Journal) (6 mins)
How AI Will Impact Chip Design And Designers (Semiconductor Engineering) (11 mins)
New Wave of Research Expands View of What Semiconductors Can Do (All About Circuits) (6āÆmins)
āļø Quantum Computing
Topological Edge States Enable Long-Range Coupling Between Distant Qubits (Quantum Zeitgeist) (7 mins)
š§ Neuromorphic Computing
From AI to Organoids: How Growing BraināLike Structures Are Advancing Machine Learning (Unite.AI) (11āÆmins)
š„ Data Centers
Data Center Capacity Around the World (Visual Capitalist) (Interactive Map)
Funding News
Last weekās funding activity ranges from early-stage AI and secure computing to major rounds in semiconductors, photonics, and cloud. OpenAIās $8.3B round leads the list, with several Series B and C deals showing sustained later-stage interest.
Amount | Name | Round | Category |
---|---|---|---|
ā¬4.8M | AI | ||
$8.5M | Semiconductors | ||
$9.7M | AI | ||
$21M | Cloud | ||
$31M | Semiconductors | ||
$50M | Photonics / Optical | ||
$51.6M | Semiconductors | ||
$100M | Cloud | ||
$125M | Semiconductors | ||
Undisclosed | Quantum | ||
$8.3B | AI |
PS: While Tzafon is an āagentic AIā startup, we think it still belongs in this list given its focus on building scalable compute infrastructure alongside its core software work.
Bonus 1: Zuckerbergās Superintelligence Vision
Mark Zuckerberg published a rather abstract and visionary letter on Superintelligence, focusing more on long-term vision than near-term technology or business impact.
The timing coincided with a blockbuster quarterly report, featuring earnings per share up 38% compared to the same period last year. The press responded, reflecting growing investor interest and mixed views on Metaās AI trajectory.
Zuckerberg claims āsuperintelligence is now in sightā as Meta lavishes billions on AI (The Guardian)
Bonus 2: Germanyās āHightech Agendaā ā Boring to Some, Relevant to Others
Germany has updated its āHightech Agendaā, a strategy to boost its role in key technologies. While not too exciting for many, it could be relevant to founders in Germany. We selected the computing-related resources (German only):
Hightech Agenda Deutschland: Quantentechnologie (Bundesministerium für Forschung, Technologie und Raumfahrt)
Hightech Agenda Deutschland: Mikroelektronik (Bundesministerium für Forschung, Technologie und Raumfahrt)