Your data center is either closer than you think or much farther away

houston voices

A map of U.S. data centers. Courtesy of Rice Businesses Wisdom

A new study shows why some facilities cluster in cities for speed and access, while others move to rural regions in search of scale and lower costs. Based on research by Tommy Pan Fang (Rice Business) and Shane Greenstein (Harvard).

Key findings:

  • Third-party colocation centers are physical facilities in close proximity to firms that use them, while cloud providers operate large data centers from a distance and sell access to virtualized computing resources as on‑demand services over the internet.
  • Hospitals and financial firms often require urban third-party centers for low latency and regulatory compliance, while batch processing and many AI workloads can operate more efficiently from lower-cost cloud hubs.
  • For policymakers trying to attract data centers, access to reliable power, water and high-capacity internet matter more than tax incentives.

Recent outages and the surge in AI-driven computing have made data center siting decisions more consequential than ever, especially as energy and water constraints tighten. Communities invest public dollars on the promise of jobs and growth, while firms weigh long-term commitments to land, power and connectivity.

Against that backdrop, a critical question comes into focus: Where do data centers get built — and what actually drives those decisions?

A new study by Tommy Pan Fang (Rice Business) and Shane Greenstein (Harvard Business School) provides the first large-scale statistical analysis of data center location strategies across the United States. It offers policymakers and firms a clearer starting point for understanding how different types of data centers respond to economic and strategic incentives.

Forthcoming in the journal Strategy Science, the study examines two major types of infrastructure: third-party colocation centers that lease server space to multiple firms, and hyperscale cloud centers owned by providers like Amazon, Google and Microsoft.

Two Models, Two Location Strategies

The study draws on pre-pandemic data from 2018 and 2019, a period of relative geographic stability in supply and demand. This window gives researchers a clean baseline before remote work, AI demand and new infrastructure pressures began reshaping internet traffic patterns.

The findings show that data centers follow a bifurcated geography. Third-party centers cluster in dense urban markets, where buyers prioritize proximity to customers despite higher land and operating costs. Cloud providers, by contrast, concentrate massive sites in a small number of lower-density regions, where electricity, land and construction are cheaper and economies of scale are easier to achieve.

Third-party data centers, in other words, follow demand. They locate in urban markets where firms in finance, healthcare and IT value low latency, secure storage, and compliance with regulatory standards.

Using county-level data, the researchers modeled how population density, industry mix and operating costs predict where new centers enter. Every U.S. metro with more than 700,000 residents had at least one third-party provider, while many mid-sized cities had none.

ImageThis pattern challenges common assumptions. Third-party facilities are more distributed across urban America than prevailing narratives suggest.

Customer proximity matters because some sectors cannot absorb delay. In critical operations, even slight pauses can have real consequences. For hospital systems, lag can affect performance and risk exposure. And in high-frequency trading, milliseconds can determine whether value is captured or lost in a transaction.

“For industries where speed is everything, being too far from the physical infrastructure can meaningfully affect performance and risk,” Pan Fang says. “Proximity isn’t optional for sectors that can’t absorb delay.”

The Economics of Distance

For cloud providers, the picture looks very different. Their decisions follow a logic shaped primarily by cost and scale. Because cloud services can be delivered from afar, firms tend to build enormous sites in low-density regions where power is cheap and land is abundant.

These facilities can draw hundreds of megawatts of electricity and operate with far fewer employees than urban centers. “The cloud can serve almost anywhere,” Pan Fang says, “so location is a question of cost before geography.”

The study finds that cloud infrastructure clusters around network backbones and energy economics, not talent pools. Well-known hubs like Ashburn, Virginia — often called “Data Center Alley” — reflect this logic, having benefited from early network infrastructure that made them natural convergence points for digital traffic.

Local governments often try to lure data centers with tax incentives, betting they will create high-tech jobs. But the study suggests other factors matter more to cloud providers, including construction costs, network connectivity and access to reliable, affordable electricity.

When cloud centers need a local presence, distance can sometimes become a constraint. Providers often address this by working alongside third-party operators. “Third-party centers can complement cloud firms when they need a foothold closer to customers,” Pan Fang says.

That hybrid pattern — massive regional hubs complementing strategic colocation — may define the next phase of data center growth.

Looking ahead, shifts in remote work, climate resilience, energy prices and AI-driven computing may reshape where new facilities go. Some workloads may move closer to users, while others may consolidate into large rural hubs. Emerging data-sovereignty rules could also redirect investment beyond the United States.

“The cloud feels weightless,” Pan Fang says, “but it rests on real choices about land, power and proximity.”

---

This article originally appeared on Rice Business Wisdom. Written by Scott Pett.

Pan Fang and Greenstein (2025). “Where the Cloud Rests: The Economic Geography of Data Centers,” forthcoming in Strategy Science.

Search trends predict future retail revenue up to three quarters in advance. Photo via Pexels

Forecasting retail sales just got smarter, thanks to Google searches

houston voices

Imagine being able to predict the future of retail sales — not with months-old economic reports, but with real-time digital breadcrumbs left behind by consumers. Thanks to new research from K. Ramesh (Rice Business) and Gary Lind (University of Pittsburgh), that future is already here.

Key findings

  • Search trends predict future retail revenue — up to three quarters ahead.
  • Markets don’t fully price in search-based insights, especially before earnings announcements.
  • Trading strategies using search data outperform traditional asset-pricing models by 2–3%.
  • Analysts often underuse this data, resulting in predictable forecast errors.
  • At the macro level, search trends closely track — and often lead — official retail sales and GDP.

Their study shows that Google search trends are powerful predictors of consumer demand, capable of forecasting U.S. retail sales up to three quarters in advance. The implications span industries and audiences — from Wall Street traders to Main Street retailers, and from data-driven CFOs to curious business students.

A smarter, faster crystal ball for retail demand

Retail sales drive nearly 70% of U.S. gross domestic product (GDP), making them one of the most important indicators of economic health. Yet traditional forecasting relies heavily on backward-looking data and broad economic indicators. Ramesh and Lind’s work offers a leap forward: search data from Google Trends — publicly available and easy to use — can offer a faster, more precise signal of where sales are headed.

“If you want to understand consumer demand,” says Ramesh, the Herbert S. Autrey Professor of Accounting at Rice Business, “look at what people are searching for — not just what companies report after the fact. These digital footprints can tell us where revenue is going, often before the market catches up.”

This low-cost approach stands in stark contrast to expensive alternatives, such as satellite images of parking lots, in-store foot traffic from mobile data, or retail scanner feeds — tools often reserved for hedge funds and large institutional investors. Google Trends offers a scalable, accessible option for anyone seeking insight into consumer behavior.

Why this matters for investors and analysts

The researchers analyzed search patterns for nearly 200 publicly traded U.S. retail companies between 2004 and 2019. They found that firms with rising search volumes tend to report higher revenues — often well before analysts or investors react.

In fact, investment strategies built around these trends earned 2–3% higher returns than traditional models would predict. And most of these gains were concentrated around earnings announcements — a sign that markets were slow to recognize what search data had already revealed.

The advantage is even more striking for companies that fly under Wall Street’s radar. For firms with limited analyst coverage, the predictive power of search data on stock performance was more than 80% stronger.

A new tool for analysts — if they’ll use it

Despite the clear signal in the data, the study finds that analysts routinely fail to incorporate search-based insights into their forecasts. That gap translates into missed revenue predictions and market inefficiencies — which sophisticated investors can exploit.

“The search data are out there,” says Lind, a Rice Business Ph.D. ’19 graduate now on faculty at Pitt Business. “They’re public, powerful, and free. Analysts who ignore them are missing an opportunity.”

The impact of this research extends well beyond equity markets. By aggregating Google search data across the retail sector, the authors found strong alignment with U.S. Census retail sales figures and GDP trends. In other words, these digital signals reflect not just firm-level interest, but macroeconomic momentum.

A data wake-up call

This makes search trends a potentially valuable early-warning system for policymakers, central banks, and anyone responsible for interpreting the economic landscape. During volatile or uncertain periods, such tools could offer precious foresight into shifting consumer behavior.

In a digital era where consumers research before they shop — whether online or in-store — it’s no surprise that their Google queries leave valuable clues. What’s surprising is how often businesses, investors, and even economists still overlook them, Ramesh says. “Search data is a goldmine. And it's just sitting there, waiting to be used.”

---

This article originally appeared on Rice Business Wisdom.

Lind and Ramesh, “Using Internet Search Data to Predict Aggregate Retail Sales and Enhance Firm-Level Revenue Expectations,” Contemporary Accounting Research (April 2025): 1-32. https://doi.org/10.1111/1911-3846.13043.

There's no crystal ball, but this researcher from Rice University is trying to see if some metrics work for economic forecasting. Photo via Getty Images

Houston researcher tries to crack the code on the Fed's data to determine economic outlook

houston voices

Research by Rice Business Professor K. Ramesh shows that the Fed appears to harvest qualitative information from the accounting disclosures that all public companies must file with the Securities and Exchange Commission.

These SEC filings are typically used by creditors, investors and others to make firm-level investing and financing decisions; and while they include business leaders’ sense of economic trends, they are never intended to guide macro-level policy decisions. But in a recent paper (“Externalities of Accounting Disclosures: Evidence from the Federal Reserve”), Ramesh and his colleagues provide persuasive evidence that the Fed nonetheless uses the qualitative information in SEC filings to help forecast the growth of macroeconomic variables like GDP and unemployment.

According to Ramesh, the study was made possible thanks to a decision the SEC made several years ago. The commission stores the reports submitted by public companies in an online database called EDGAR and records the IP address of any party that accesses them. More than a decade ago, the SEC began making partially anonymized forms of those IP addresses available to the public. But researchers eventually figured out how to deanonymize the addresses, which is precisely what Ramesh and his colleagues did in this study.

"We were able to reverse engineer and identify those IP addresses that belonged to Federal Reserve staff," Ramesh says.

The team ultimately assembled a data set containing more than 169,000 filings accessed by Fed staff between 2005 and 2015. They quickly realized that the Fed was interested only in filings submitted by a select group of industry leaders and financial institutions.

But if Ramesh and his colleagues now had a better idea of precisely which bellwether firms the Fed focused on, they still had no way of knowing exactly what Fed staffers had gleaned from the material they accessed. So the team decided to employ a measure called "tone" that captures the overall sentiment of a piece of text – whether positive, negative, or neutral.

Building on previous research that had identified a set of words with negatively toned financial reports, Ramesh and his colleagues examined the tone of all the SEC filings accessed by Fed staff between one meeting of the Federal Open Markets Committee (FOMC) and the next. The FOMC sets interest rates and guides monetary policy, and its meetings provide an opportunity for Fed officials to discuss growth forecasts and announce policy decisions.

The researchers then examined the Fed's growth forecasts to see if there was a relationship between the tone of the documents that Fed staff examined in the period between FOMC meetings and the forecasts they produced in advance of those meetings.

The team found close correlations between the tone of the reports accessed by the Fed and the agency’s forecasts of GDP, unemployment, housing starts and industrial production. The more negative the filings accessed prior to an FOMC meeting, for example, the gloomier the GDP forecast; the more positive the filings, the brighter the unemployment forecast.

Ramesh and his colleagues also compared the Fed's forecasts with those of the Society of Professional Forecasters (SPF), whose members span academia and industry. Intriguingly, the researchers found that while the errors in the SPF's forecasts could be attributed to the absence of the tonal information culled from the SEC filings, the errors in the Fed’s forecasts could not. This suggests both that the Fed was collecting qualitative information that the SPF was not—and that the agency was making remarkably efficient use of it.

"They weren’t leaving anything on the table," Ramesh says.

Having solved one mystery, Ramesh would like to focus on another; namely, how does the Fed identify bellwether firms in the first place?

Unfortunately, the SEC no longer makes IP address data publicly available, which means that Ramesh and his colleagues can no longer study which companies the Fed is most interested in. Nonetheless, Ramesh hopes to use the data they have already collected to build a model that can accurately predict which firms the Fed is most likely to follow. That would allow the team to continue studying the same companies that the Fed does, and, he says, “maybe come up with a way to track those firms in order to understand how the economy is going to move.”

------

This article originally ran on Rice Business Wisdom and was based on research from K. Ramesh is Herbert S. Autrey Professor of Accounting at Jones Graduate School of Business at Rice University.

Ad Placement 300x100
Ad Placement 300x600

CultureMap Emails are Awesome

Property Showcase

Houston unicorn closes $421M to fuel first phase of flagship energy project

Heating Up

Houston geothermal unicorn Fervo Energy has closed $421 million in non-recourse debt financing for the first phase of its flagship Cape Station project in Beaver County, Utah.

Fervo believes Cape Station can meet the needs of surging power demand from data centers, domestic manufacturing and an energy market aiming to use clean and reliable power. According to the company, Cape Station will begin delivering its first power to the grid this year and is expected to reach approximately 100 megwatts of operating capacity by early 2027. Fervo added that it plans to scale to 500 megawatts.

The $421 million financing package includes a $309 million construction-to-term loan, a $61 million tax credit bridge loan, and a $51 million letter of credit facility. The facilities will fund the remaining construction costs for the first phase of Cape Station, and will also support the project’s counterparty credit support requirements.

Coordinating lead arrangers include Barclays, BBVA, HSBC, MUFG, RBC and Société Générale, with additional participation from Bank of America, J.P. Morgan and Sumitomo Mitsui Trust Bank, Limited, New York Branch.

“As demand for firm, clean, affordable power accelerates, EGS (Enhanced Geothermal Systems) is set to become a core energy asset class for infrastructure lenders,” Sean Pollock, managing director, project Finance at RBC Capital Markets, said in a news release. “Fervo is pioneering this step change with Cape Station, a vital contribution to American energy security that RBC is proud to support.”

The oversubscribed financing marks Cape Station’s shift from early-stage and bridge funding to a long-term, non-recourse capital structure, according to the news release.

“Non-recourse financing has historically been considered out of reach for first-of-a-kind projects,” David Ulrey, CFO of Fervo Energy, said in a news release. “Cape Station disrupts that narrative. With proven oil and gas technology paired with AI-enabled drilling and exploration, robust commercial offtake, operational consistency, and an unrelenting focus on health and safety, we have shown that EGS is a highly bankable asset class.”

Fervo continues to be one of the top-funded startups in the Houston area. The company has raised about $1.5 billion prior to the latest $421 million. It also closed a $462 million Series E in December.

According to Axios Pro, Fervo filed for an IPO that would value the company between $2 billion and $3 billion in January.

---

This article first appeared on EnergyCapitalHTX.com.

Houston food giant Sysco to acquire competitor in $29 billion deal

Mergers & Acquisitions

Sysco, the nation's largest food distributor, will acquire supplier Restaurant Depot in a deal worth more than $29 billion.

The acquisition would create a closer link between Sysco and its customers that right now turn to Restaurant Depot for supplies needed quickly in an industry segment known as “cash-and-carry wholesale.”

Sysco, based in Houston, serves more than 700,000 restaurants, hospitals, schools, and hotels, supplying them with everything from butter and eggs to napkins. Those goods are typically acquired ahead of time based on how much traffic that restaurants typically see.

Restaurant Depot offers memberships to mom-and-pop restaurants and other businesses, giving them access to warehouses stocked with supplies for when they run short of what they've purchased from suppliers like Sysco.

It is a fast growing and high-margin segment that will likely mean thousands of restaurants will rely increasingly on Sysco for day-to-day needs.

Restaurant Depot shareholders will receive $21.6 billion in cash and 91.5 million Sysco shares. Based on Sysco’s closing share price of $81.80 as of March 27, 2026, the deal has an enterprise value of about $29.1 billion.

Restaurant Depot was founded in Brooklyn in 1976. The family-run business then known as Jetro Restaurant Depot, has become the nation's largest cash-and-carry wholesaler.

The boards of both companies have approved the acquisition, but it would still need regulatory approval.

Shares of Sysco Corp. tumbled 13% Monday to $71.26, an initial decline some industry analysts expected given the cost of the deal.

Houston researcher builds radar to make self-driving cars safer

eyes on the road

A Rice University researcher is giving autonomous vehicles an “extra set of eyes.”

Current autonomous vehicles (AVs) can have an incomplete view of their surroundings, and challenges like pedestrian movement, low-light conditions and adverse weather only compound these visibility limitations.

Kun Woo Cho, a postdoctoral researcher in the lab of Rice professor of electrical and computer engineering Ashutosh Sabharwal, has developed EyeDAR to help address such issues and enhance the vehicles’ sensing accuracy. Her research was supported in part by the National Science Foundation.

The EyeDAR is an orange-sized, low-power, millimeter-wave radar that could be placed at streetlights and intersections. Its design was inspired by that of the human eye. Researchers envision that the low-cost sensors could help ensure that AVs always pick up on emergent obstacles, even when the vehicles are not within proper range for their onboard sensors and when visibility is limited.

“Current automotive sensor systems like cameras and lidar struggle with poor visibility such as you would encounter due to rain or fog or in low-lighting conditions,” Cho said in a news release. “Radar, on the other hand, operates reliably in all weather and lighting conditions and can even see through obstacles.”

Signals from a typical radar system scatter when they encounter an obstacle. Some of the signal is reflected back to the source, but most of it is often lost. In the case of AVs, this means that "pedestrians emerging from behind large vehicles, cars creeping forward at intersections or cyclists approaching at odd angles can easily go unnoticed," according to Rice.

EyeDAR, however, works to capture lost radar reflections, determine their direction and report them back to the AV in a sequence of 0s and 1s.

“Like blinking Morse code,” Cho added. “EyeDAR is a talking sensor⎯it is a first instance of integrating radar sensing and communication functionality in a single design.”

After testing, EyeDAR was able to resolve target directions 200 times faster than conventional radar designs.

While EyeDAR currently targets risks associated with AVs, particularly in high-traffic urban areas, researchers also believe the technology behind it could complement artificial intelligence efforts and be integrated into robots, drones and wearable platforms.

“EyeDAR is an example of what I like to call ‘analog computing,’” Cho added in the release. “Over the past two decades, people have been focusing on the digital and software side of computation, and the analog, hardware side has been lagging behind. I want to explore this overlooked analog design space.”

View All Listings