Cool Vector

The growing implementation of liquid cooling technologies is forcing data center MEP engineers to rethink their fundamental design assumptions, says Chris Sutton, founder of Sutton Engineering which specializes in data center design.
Sutton tells Cool Vector that, while the electrical side of data center infrastructure reached a level of maturity some time ago, the mechanical side is now in the middle of a significant transformation, and keeping up with that transformation comes down to having a team of skilled engineers. Sutton Engineering's biggest focus right now is building a training program capable of bringing engineers from adjacent fields up to speed in the rapidly evolving world of data center MEP design.
Access the full transcript and a searchable content library at the Cool Vector Substack: https://coolvector.substack.com/p/the-big-game-changers-for-data-center?r=4tjd55&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
#coolvector #datacenter #AI #energy #infrastructure #MEP

Emerging and developing markets remain significantly underinvested in digital infrastructure, says Obinna Isiadinso, the global sector lead for data centers and cloud services investments at the International Finance Corporation (IFC).
"India today has something like 1.7 gigawatts of capacity, compared to Northern Virginia that has 3 or 4 gigawatts," Isiadinso tells Cool Vector, citing the country as an example of an emerging market facing a compute constraint. 
Isiadinso, whose team has invested in data centers since 2005 and now holds roughly $400 million in the sector out of a broader $3 billion digital infrastructure portfolio, walks through how the IFC identifies opportunity across roughly 20 priority markets, why its decades-long presence gives it credibility with regulators and governments other capital providers haven't yet reached, and which regions are pulling ahead in the race for capacity.
Key Takeaways:
• Asia leads IFC's regional exposure, followed by Latin America and Africa. India and Malaysia are the standout Asian markets, with recent IFC commitments including debt financing to expand NTT's platform in India and Yondr Group's 300-megawatt hyperscale facility in Johor Bahru, Malaysia.
• Brazil, South Africa, and Poland anchor their regions. Brazil is IFC's top Latin American market, where it holds exposure through Scala Data Centers; South Africa holds roughly 60% of Africa's data center capacity; and Poland remains IFC's primary Eastern European focus, with growing interest in Romania and Bulgaria.
• AI demand is emerging alongside cloud. While collocation and cloud demand still dominate, IFC is starting to see meaningful AI-driven demand in its largest markets — particularly Brazil, India, and Malaysia.
• Projects above roughly 300 megawatts can outgrow the pool of institutional buyers able to acquire them once stabilized, though Isiadinso says those cases remain rare — most projects under 100 megawatts see strong acquisition interest.
Access the full transcript and a searchable archive at the Cool Vector Substack.
#digitalinfrastructure #datacenter #AI

Building AI infrastructure at gigawatt scale means "developing city-sized machines,'" says Craig Deering, senior construction manager at Oracle. 
Drawing on decades of experience — from dismantling AT&T and building the Baby Bells to leading cloud and now AI data center projects — Deering tells Cool Vector: "The challenges of developing a city-sized machine have been known for at least a century. We just need to realize that we're building at that scale."
Deering shares his outlook on the economics of AI compute, the human toll of building at unprecedented speed, and why he remains bullish on AI's impact on jobs despite decades spent watching technology reshape his industry.
Key Takeaways:
• Job sites that once had 1,200 workers now have 6,000 to 8,000, split across multiple companies — a human resource challenge on a scale Deering compares to building the Panama Canal or Gilded Age company towns like Hershey, Pennsylvania• Driving down the incremental cost of AI token production is the industry's central equation, tying directly into Jevon's paradox: as compute gets cheaper, demand explodes• Structural job displacement is real, but Deering argues history — from hand-drafting to CAD to BIM — shows technology ultimately expands opportunity rather than eliminating it
Access the full transcript and a searchable archive at the Cool Vector Substack.
#digitalinfrastructure #datacenter #AI

Data centers have access to plenty of power, but it's often stranded rather than put to use, says Ken Sullivan, co-founder and CEO of Bay Compute.
"Power is the bottleneck to AI growth," Sullivan says. "A lot of the 'colos' and data centers are operating at 30, 40, 50% average power utilization."
Sullivan lays out how the shift toward AI inference is intensifying the power problem, why colocation operators are now being forced to optimize infrastructure they've never had to think about before, and why he believes the industry's component-by-component fixes are hitting diminishing returns.
Key Takeaways:
• Inference workloads create a much larger amplitude in power spikes than traditional cloud or training data centers, driving down average utilization and worsening the stranded-capacity problem — especially at edge and colocation facilities.
• Bay Compute's customers are unlocking 10-20% more capacity. By reducing peak power events, the company says its customers are able to free up additional capacity and sell it on to their own customers.
• Unlike hyperscalers, which have spent 20 years building tools to optimize dollar-per-token and watt-per-token, colocation facility operators historically passed power costs straight to tenants and had no incentive to optimize until new power became scarce.
• Most data center projects aren't approved until a power contract is secured, a shift from the old land-first model, pushing developers into secondary and tertiary markets where power is available.
Access the full transcript and a searchable content library at the Cool Vector Substack: https://open.substack.com/pub/coolvector/p/solutions-for-the-data-center-power?r=4tjd55&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
#coolvector #datacenter #AI #inference #energy #power #colocation

The biggest constraint facing data centers is finding the people capable of operating those data centers, says Brent Burnett, Managing Director and Head of Infrastructure and Real Assets at Hamilton Lane.
Burnett tells Cool Vector that as capital floods into digital infrastructure, the pool of management teams equipped to run complex, cap-ex-heavy platforms remains small, and the price of accessing them keeps growing.
In an in-depth conversation with Cool Vector's David Snow and United Integrity Advisors' Eli Scher, Burnett explains how Hamilton Lane screens operators, why current valuations concern him, and how the firm structured its recent continuation vehicle investment in Flexential.
Key takeaways:
• Digital infrastructure now dominates infrastructure portfolios. What was 8-10% of a typical fund's allocation a decade ago has grown to 30-40% today, with most fresh capital targeting data centers specifically.
• Valuation risk is "very acute" right now. Multiples that historically ran 16-18x have climbed into the mid-to-high 20s, and Burnett says he's seen irrational excitement among sellers with unrealistic expectations.
• Hamilton Lane avoids speculative build-and-hope bets. The firm favors platforms where growth is backed by existing, already-contracted tenants and a powered, permitted land bank.
• While many believe the biggest risk is technology disruption, it's actually execution. Burnett and Scher agree over-leveraged capital structures paired with growth plans that don't materialize pose a greater threat to the sector than any black-swan tech shift.
• Community and political pushback are rising risks too. As data centers scale, "stroke of pen risk" is growing, and Burnett says operators need to get smarter about community engagement or risk political backlash to their build-out plans.
Access the full transcript and a searchable content library on the Cool Vector Substack: https://open.substack.com/pub/coolvector/p/for-hamilton-lane-good-data-center?r=4tjd55&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
#coolvector #datacenter #digitalinfrastructure #privateequity
Note: Hamilton Lane has $1 trillion in assets under management and supervision, composed of $141.8 billion in discretionary assets and $905.3 billion in non-discretionary assets, as of March 31, 2026.

The digital infrastructure industry is incredibly diverse and needs talent of all types, says Harris Hamid, a computer science student at Stevens Institute of Technology in Hoboken, New Jersey, and a student advocate for non-profit Nomad Futurist Foundation. 
"No matter what you're doing, you can find a spot here," Hamid tells Cool Vector. 
Hamid emphasizes how the industry welcomes people from all backgrounds, recalling how he has already met graphic designers, engineers, and career-changers with different areas of expertise to offer. He encourages students to see Nomad Futurist as a resource for learning and networking, regardless of their major, emphasizing that there's a place in the industry for everyone.
Access the transcript and a searchable library of content at the Cool Vector Substack.
#datacenter #digitalinfrastructure #computerscience

Walter Cannon helped create and implement LinkNYC, the project that brought free public Wi-Fi to all five boroughs of New York City. Cannon tells Cool Vector about how the project came to be under the Bloomberg Administration, as well as the logistical and political hurdles of rolling out thousands of Wi-Fi kiosks.
Looking ahead, Cannon - a self-described "all-around technology guy" — discusses the growing need for low-latency infrastructure closer to end users, especially as AI adoption becomes more widespread.
Access the transcript and a searchable library of content at the Cool Vector Substack: 
#datacenter #digitalinfrastructure #newyorkcity #wifi

Artificial intelligence tools are being adopted across construction, including digital infrastructure construction, says Indigo Pinto, Co-Founder IBB Solutions, a management firm for digital infrastructure construction projects.
Pinto is also an ambassador for Nomad Futurist, and she reflects on discovering Nomad Futurist and how the phrase "the future is now" resonated with her view of the industry's trajectory, particularly when it comes to the rising generation of talent entering the field.
Access the full transcript and a searchable library of content at the Cool Vector Substack.
#datacenter #digitalinfrastructure #constructionmanagement #coolvector

The data center industry is leaving enormous value on the table by failing to apply chip-level simulation techniques to the full infrastructure stack, argues Sherman Ikemoto of Cadence. Closing the “chip-to-chiller” design gap is now an existential requirement, given the pace of AI compute density growth.
In an in-depth Cool Vector interview, Cadence’s Group Director of Business Development describes how “the timescales to build chips, and to build the facilities that house the chips, are totally different. So there's this natural massive gap in the design chain.”
Key takeaways from Ikemoto’s interview:
• If data center developers are able to compress a 24-month AI factory standup to 18 or 16 months, that translates directly into billions of dollars of accelerated return.
• For three decades, rack power density grew ~10% annually. In the AI era, it has jumped to 60–100% per year—an order-of-magnitude acceleration that fundamentally breaks traditional data center cooling technologies and design methodologies.
• Backing off GPU power utilization by 10–20% frees up enough headroom to add more GPUs to the same power envelope.
Access the full transcript and a searchable content library at the Cool Vector Substack: https://coolvector.substack.com/p/from-chip-to-chiller?r=4tjd55
#datacenter #digitalinfrastructure #ai

Stan Hanks invented a commodities market for broadband while at Enron, and has well informed views on the need for a similar market for compute. The main challenge: units of compute are proving difficult to standardize. 
In this episode of Cool Vector, Stan Hanks of Kreneon, Wayne Nelms of Ornn, Hadassa Lutz of Cloud2Ground and host David Snow join Cool Vector to examine whether compute can be turned into a tradable commodity — drawing a direct line from Hanks' experience creating a broadband futures market at Enron to the emerging effort to do the same thing for GPU capacity. 
Compute has the economic conditions for a futures market — surging demand, volatile prices, massive capital at risk — but lacks the defining characteristic of a true commodity, which is fungibility. Workload built for one chip architecture can't simply be swapped to another. 
This fascinating conversation takes place as the one-year lease price of the H100 GPU has jumped nearly 40 percent between late 2025 and early 2026, and market players are now are racing to launch the first regulated compute futures products.
Key takeaways from this episode:
• Token usage is opaque and hard to predict - Hanks compares tokens to a foreign currency where you don't know how much work you're actually getting — likening the experience to a parking meter that speeds up mid-session, demanding another quarter well before the hour you thought you'd paid for.
• Anthropic and ChatGPT are token 'price setters' - Nelms observes that the closed-source model providers — OpenAI and Anthropic — produce their own tokens and set their own prices, making the token market far less liquid and competitive than the GPU rental market where multiple neo clouds are offering access to relatively comparable hardware.
• GPU price fluctuations call for a futures market - The one-year lease price for the H100 GPU jumped 38.2% between October 2025 and March 2026, precisely the kind of volatility that Wayne Nelms argues a liquid futures market — with prices written into debt covenants and used as hedging benchmarks — would give infrastructure investors the tools to manage. 
• How Hanks created a broadband futures market while at Enron - By building his own fiber network to create a naturally long position, then standardizing both the buy-side and sell-side contracts around a common definition of bandwidth capacity, Stan used Enron's $20 billion treasury to move the market from bespoke, one-off bilateral deals toward something that looked and behaved like a commodity.
• Compute has a 'perfect opportunity' for commodity market mechanics - Han ks draws a direct parallel between the broadband boom of the late 1990s and today's compute buildout — massive capital being deployed into infrastructure without visibility into future demand and price — and argues that this information vacuum is precisely the condition under which commodity market mechanics have historically proven most valuable.
• Stan Hanks barely escaped the wreckage of Enron - At a famous analyst meeting in January 2000, Stan was presented with a transcript of promises made to the market about products that were pure science fiction, realized he would be asked to attest to their veracity, said he couldn't do it, and left the following Monday — narrowly avoiding a collapse that would eventually make him a DOJ witness for eleven years.
Access the full transcript and a searchable content library on the Cool Vector Substack.
#coolvector #datacenter #GPU #tech #commodities #enron #digitalinfrastructure