Orbital Data Centers: Why the Hype Outpaces Reality
Opens in a new window Opens an external website Opens an external website in a new window
This website utilizes technologies such as cookies to enable essential site functionality, as well as for analytics, personalization, and targeted advertising. To learn more, view the following link: Privacy Policy
Manage Preferences
Storage Preferences
IEEE.orgIEEE XploreIEEE StandardsIEEE Job SiteMore Sites
The July issue of _IEEE Spectrum_ is here!
Close bar;)
[](https://spectrum.ieee.org/ "Return to homepage")
The Orbital Data Center Hype Machine Is Already in Orbit
Share
FOR THE TECHNOLOGY INSIDER
Search:
Explore by topic
AerospaceAIBiomedicalClimate TechComputingConsumer ElectronicsEnergyHistory of TechnologyRoboticsSemiconductorsTelecommunicationsTransportation
FOR THE TECHNOLOGY INSIDER
Topics
AerospaceAIBiomedicalClimate TechComputingConsumer ElectronicsEnergyHistory of TechnologyRoboticsSemiconductorsTelecommunicationsTransportation
Sections
FeaturesNewsOpinionCareersDIYEngineering Resources
More
NewslettersSpecial ReportsCollectionsExplainersTop Programming LanguagesRobots Guide ↗IEEE Job Site ↗
For IEEE Members
Current IssueMagazine ArchiveThe InstituteThe Institute Archive
For IEEE Members
Current IssueMagazine ArchiveThe InstituteThe Institute Archive
IEEE Spectrum
About UsContact UsReprints & Permissions ↗Advertising ↗
Follow IEEE Spectrum
[](https://twitter.com/ieeespectrum)[](https://www.facebook.com/IEEE.Spectrum)[](https://www.instagram.com/ieeespectrum/)[](https://www.threads.net/@ieeespectrum)[](https://www.linkedin.com/company/ieee-spectrum/)[](https://www.youtube.com/c/ieeespectrum)[](https://www.tiktok.com/@ieeespectrum)[](https://spectrum.ieee.org/customfeeds/feed/all-topics/rss)
Support IEEE Spectrum
_IEEE Spectrum_ is the flagship publication of the IEEE — the world’s largest professional organization devoted to engineering and applied sciences. Our articles, videos, and infographics inform our readers about developments in technology, engineering, and science.
About IEEEContact & SupportAccessibilityNondiscrimination PolicyTermsIEEE Privacy PolicyCookie PreferencesAd Privacy Options
© Copyright 2026 IEEE — All rights reserved. A public charity, IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity.
Enjoy more free content and benefits by creating an account
Saving articles to read later requires an IEEE Spectrum account
The Institute content is only available for members
Downloading full PDF issues is exclusive for IEEE Members
Downloading this e-book is exclusive for IEEE Members
Access to _Spectrum_ 's Digital Edition is exclusive for IEEE Members
Following topics is a feature exclusive for IEEE Members
Adding your response to an article requires an IEEE Spectrum account
Create an account to access more content and features on _IEEE Spectrum_ , including the ability to save articles to read later, download Spectrum Collections, and participate in conversations with readers and editors. For more exclusive content and features, consider Joining IEEE .
Join the world’s largest professional organization devoted to engineering and applied sciences and get access to all of Spectrum’s articles, archives, PDF downloads, and other benefits. Learn more about IEEE →
Join the world’s largest professional organization devoted to engineering and applied sciences and get access to this e-book plus all of _IEEE Spectrum’s_ articles, archives, PDF downloads, and other benefits. Learn more about IEEE →
Close
Access Thousands of Articles — Completely Free
Create an account and get exclusive content and features: **Save articles, download collections,** and **post comments** — all free! For full access and benefits, subscribe to _Spectrum_.
AerospaceAIMagazineJuly 2026Opinion
The Orbital Data Center Hype Machine Is Already in Orbit
Why the stars—and the math—won’t align for space compute anytime soon
9 hours ago
4 min read
[](javascript:;) 3
Harry Goldstein is Editor in Chief of IEEE Spectrum.
!Image 1: Globe wrapped in multicolored pins and connecting lines, symbolizing global networks
Edmon de Haro
[](mailto:?subject=The%20Orbital%20Data%20Center%20Hype%20Machine%20Is%20Already%20in%20Orbit&body=https://spectrum.ieee.org/orbital-data-center-hype)
orbital data centerssatellitesspacexelon muskStarcloudAIgpus
“The lowest-cost place to put AI will be in space, and that will be true within two years, maybe three at the latest,” SpaceX founder Elon Musk told the World Economic Forum in Davos this past January, as his company was preparing to go public.
Later that month, SpaceX filed an application with the Federal Communications Commission for an orbital data center constellation of up to 1 million satellites in low Earth orbit, 500 to 2,000 kilometers above Earth. And just three days before the IPO, he discussed some initial design specifications for a new AI-1 satellite data center in a video interview.
Musk is prone to hyperbole when it comes to timelines. Full self-driving cars by 2017. First human mission to Mars in 2024. Ten thousand Optimus humanoid robots by the end of 2025. Et cetera. For orbital data centers, which he says will be a cost-effective alternative to terrestrial data centers within three years, the math won’t make sense for several years, if ever.
Consider this: There are roughly 14,500 active satellites in orbit. Musk’s Starlink constellation accounts for about two thirds of those. Both the launch cadences and satellite-manufacturing capacity would have to scale up astronomically to deploy a million orbital data center satellites.
For context, there have been roughly 7,000 orbital launches in all of human history. To loft 1 million satellites into low Earth orbit on SpaceX’s Starship, which is designed to carry up to 60 satellites per vehicle, would require 16,666 launches exclusively devoted to satellite deployments. Considering that SpaceX launched a record 165 orbital missions in 2025, even at 10 times that cadence, it would take a decade. And how long would it take to build 1 million satellites, given Starlink’s current pace of around 4,000 per year and a generous tenfold increase in capacity? Short of a manufacturing revolution, try 25 years.
The reality is that the vision of massive constellations of orbital data centers is nowhere close to being realized.
As this month’s cover story, “Why Orbital Data Centers Are So Hard” by Andrew Cavalier of ABI Research, makes clear, the reality is that the vision of massive constellations of orbital data centers is nowhere close to being realized.
Dina Genkina, _IEEE Spectrum_’s computing and hardware editor, put the idea into perspective: “Starcloud (a startup that has applied to the FCC for an 88,000 orbital data center satellite constellation) sent one Nvidia H100 GPU in space so far. Their radiator was too weak to let the chip run at full power.”
As Cavalier shows, cooling even a single Nvidia H100 GPU in space is difficult: It draws 700 watts, which will require 1.4 square meters of radiator at 60 °C. A 40-kilowatt rack of servers will need an 80-m² radiator; a 100-megawatt data center will require 2,500 of those radiators. Some astronomers are understandably concerned that a million satellites with giant radiative wings would blot out the stars.
So if the economics doesn’t make sense, if the chips are at the mercy of the radiative ravages of space, and if humanity will lose its view of the stars, not to mention increasing the risk of triggering the Kessler syndrome, why are the hyperscalers hyping orbital data centers?
Genkina offered the obvious answer: sweet, sweet moolah. “The Elon Musk part of it is honestly genius because he’s got xAI building the data centers, SpaceX sending them to space, and Tesla building solar panels,” Genkina says. “It’s almost like he’s paying himself.”
Two Analyst’s Views of SpaceX’s Proposed AI1 Data Center Satellite
**Michael Pierce, Principal at Technology Strategy Partners**
Musk’s timelines are notoriously overly ambitious, but I think SpaceX’s orbital data centers might reach cost parity with terrestrial data centers in 5 to 10 years. The Starlink laser-link network already exists as the communication backbone for any SpaceX compute constellation, and that infrastructure is what no new entrant can replicate quickly. The chip-agnostic payload design probably reflects their disclosed difficulty securing AI silicon as much as any modularity philosophy. My view is that the only realistic near-term application is a SpaceX mega-constellation for inference. Training workloads likely cannot tolerate the synchronization and latency constraints of a distributed orbital system.
Our report analyzed the market from the integrator’s vantage point, but AI1 is what it looks like when one player has assembled all the necessary advantages simultaneously. The question is whether the terrestrial data center industrial base will degrade or improve on economics. I don’t have insight into SpaceX’s internal costs, as opposed to public pricing, on all their components, so it’s hard to say if they’ll completely dominate or not. Even if they are not cost competitive with terrestrial data centers for another 5 to 10 years, it may simply be faster to get new compute that just happens to be in space.
**Matt Hasan, AI strategist and independent consultant**
My initial view is that AI1 does not fundamentally change the rationale for space-based data centers as much as it changes the timeline and scale. The underlying drivers remain the same: escalating AI compute demand, growing power constraints on terrestrial grids, and the desire to colocate energy generation with computation.
What AI1 does signal is that the concept is beginning to move from theoretical discussion toward engineering and capital allocation decisions. The announcement adds credibility to the idea that hyperscale computing infrastructure may eventually expand beyond terrestrial constraints rather than simply competing for increasingly scarce grid capacity on Earth.
That said, significant economic and technical questions remain. Launch costs, maintenance, hardware replacement cycles, thermal management, latency-sensitive workloads, and overall system economics will ultimately determine whether space-based data centers become a mainstream extension of AI infrastructure or remain a niche capability for specialized applications. The key development is not that these questions have been resolved, but that major industry players now appear willing to invest resources toward answering them.
From Your Site Articles
- Nvidia Sends a Powerful GPU to Space ›
- Why Orbital Data Centers Are Harder Than Silicon Valley Thinks ›
- How Stupid Would It Be to Put Data Centers in Space? ›
Related Articles Around the Web
- Orbital data centers, part 1: There’s no way this is economically viable, right? - Ars Technica ›
orbital data centerssatellitesspacexelon muskStarcloudAIgpus
is Editor in Chief of _IEEE Spectrum_.See full bio→
The Conversation (0)
Publish
Sort by
Emily Bender Sets the Record Straight on “Stochastic Parrots”
30 Jun 2026
7 min read
[](javascript:;) 1
AIEnergySponsored ArticleComputing
As AI Reshapes Global Energy Systems, Melbourne Leads Through Engineering Collaboration
5 hours ago
6 min read
[](javascript:;) 1
SemiconductorsAIMagazineFeatureJuly 2026Telecommunications
AI Is Designing Radio Chips That Humans Couldn’t Even Imagine
24 Jun 2026
14 min read
[](javascript:;) 21
Related Stories
ComputingMagazineFeatureJuly 2026
Why Orbital Data Centers Are Harder Than Silicon Valley Thinks
Vibecoded Malware Is Flooding the Internet
Satellites Overwhelm Space Telescopes With “Railroad” Tracks
SemiconductorsAIMagazineFeatureJuly 2026
The Lab Mistake That Might Revolutionize Computing
The result is a simple and efficient neuromorphic device that mimics a brain cell
29 Jun 2026
10 min read
[](javascript:;) 6
[](mailto:?subject=The%20Lab%20Mistake%20That%20Might%20Revolutionize%20Computing&body=https://spectrum.ieee.org/artificial-neurons-on-silicon-chips)
Vertical
Dan Page
DarkGray
**Today, you****probably asked** a question of a large language model, or accepted a connection suggestion on LinkedIn, or watched a recommended video on YouTube, or took a different route to work based on a traffic prediction from Google Maps. In other words, you probably used artificial intelligence. But what you might not know is how much energy that interaction consumed or why.
!Image 6
AI requires processing massive amounts of data, which is usually done in large data centers populated by thousands of GPUs capable of executing up to trillions of operations per second. But each of those GPUs achieves that by consuming as much as 1,000 watts apiece. For comparison, if you’ve got a newer smartphone, it probably uses less than 1 W. That kilowatt figure puts GPUs on the same level as vacuum cleaners, dishwashers, and stoves, but with the big difference that data-center processors are operating uninterrupted around the clock.
Fundamentally, a lot of this inefficiency is because GPUs are trying to simulate the workings of artificial neural networks using software and billions of transistors, which requires using energy to move massive amounts of data. What’s more, the simulated artificial neurons that make up these networks lack even a fraction of the complex computing behavior of the biological neurons that comprise the most energy-efficient computing system that we know, the human brain.
!Image 7: Gloved hand with tweezers holding a tiny swab over colorful striped backgroundDan Page
The brain is roughlyone million times as energy efficient at many of the comparable tasks we set for AI. To try to approach these efficiencies, a radically different way of computing called neuromorphic engineering is seeking to build electronic components and circuits that act more like the brain’s neurons and the synapses that connect them.
Huge amounts of work have gone into making electronics operate more like biological neurons and synapses. Some research has focused on developing new, experimental devices, but they aren’t yet reliable enough to be used in large systems. Other efforts aim to implement neurons and synapses by interconnecting many complementary metal-oxide-semiconductor (CMOS) transistors—the workhorses of digital logic—to simulate a single neuron and synapse. But this approach requires so many transistors (and a few bulky capacitors) that it greatly limits the size of the system that can be constructed, making it unclear how such brain-inspired hardware could ever scale up and compete with state-of-the-art GPUs.
But all along there was an artificial neuron and a synapse—each a single device—hiding in plain sight. We found them last year. They were each made possible by an ordinary CMOS transistor—and not even a very good one at that. This is the story of their accidental discovery and their great promise for lowering the environmental footprint of AI.
Biological and artificial neurons
Modern digital electronics is based on producing and manipulating the ones and zeros of the binary code through the operation of metal-oxide-semiconductor field-effect transistors. MOSFETs have evolved in recent years, but their classic form consists of a piece of silicon that has been doped to contain an excess of either positive (_p_-type) or negative (_n_-type) charge carriers. (CMOS logic contains transistors of both types.) The device has two terminals connected to the silicon through regions highly doped with the opposite polarity of the rest of the silicon—the source and the drain. Another terminal, the gate, sits atop the silicon that separates the source from the drain. The gate itself doesn’t connect directly to this silicon, instead resting above a thin layer of insulating dielectric.
Notably, there is a fourth terminal that attaches to the bulk of the silicon; think of this bulk terminal as connecting to the underside of the chip. It doesn’t typically get much attention, but it’s very important to our story.
When voltage is applied at the gate and the bulk terminal is grounded, charge carriers of the same polarity as the source and drain are attracted to the channel region. In the case of an _n_-type source and drain, that will be electrons; for _p_-type it will be holes. The presence of these charges forms a conductive channel that reduces the resistance between the source and the drain by several orders of magnitude, and the device switches on. As the voltage at the gate increases, this physical phenomenon produces a current signal that, when plotted against the gate voltage, rises steadily. This response is ideal for logic gates, converters, multiplexers, memories, and other digital circuits. But it is not a good fit for mimicking the behavior of a neuron.
In real neural tissue, brain cells, called neurons, consist of a cell body, a long projection called an axon, and short branching projections called dendrites. The suite of behaviors and computing this collection of components is capable of is rich and broad, but the portion that artificial neural networks hope to copy is this: When the cell body’s voltage is perturbed enough to reach a particular threshold, a self-propagating pulse of voltage, called an action potential, shoots down the axon. The axon terminates in a synapse, an electrochemical connection between the axon and another neuron’s dendrites. The action potential will then temporarily boost the voltage of this next neuron, by an amount that depends on the strength of the synaptic connection. If enough action potentials reach these dendrites in a given time—from this neuron or from others that might also form synapses there—the cell body’s voltage will surpass the threshold and trigger its own action potential.
The MOSFET Neuron
The unusual action the authors discovered is understandable if you consider that a MOSFET contains a hidden bipolar-junction transistor.
!Image 8: MOSFET diagrams with carrier flow and plot of drain current versus drain voltage #### **TRANSISTOR BEHAVIOR**
Under normal operation, with the bulk terminal grounded, increasing voltage at the drain leads to current that increases steadily. When the voltage decreases, current follows the same sloped path. Although some pairs of electrons and holes are created by current crashing into silicon atoms, these are swept away before they can accumulate.
!Image 9: NSRAM transistor diagrams with bias circuits and I\u2013V curve highlighting C and D states #### **NSRAM BEHAVIOR**
Adding resistance to the bulk terminal means these extra holes pile up, increasing the bulk voltage relative to the source. Once that voltage reaches a certain value, the hidden transistor activates, causing current to spike. Current remains high until the drain voltage drops past a certain point.
To get closer to the behavior of real neurons, artificial neurons should produce a current spike when a critical voltage threshold is crossed and then quickly relax back to a resting state on their own. This spike needs to be sudden—nonlinear. It should also exhibit some hysteresis; that is, the activation and relaxation voltages should be different from each other to ensure that current flows only for a certain amount of time.
What’s wanted from an artificial synapse, the thing that connects two artificial neurons, is less complicated, but equally important. The main thing is that its conductance can be electronically adjustable. The device’s conductive states should increase and decrease in a linear pattern and remain stable over time.
No single MOSFET working under the standard operation mechanism can reproduce either of these neural properties. Instead, it’s been done by combining them into complex circuits. Until now, each neuron and each synapse has been implemented by interconnecting dozens and sometimes even hundreds of MOSFETs, which is highly inefficient in terms of area, performance, and cost. To limit the amount of space needed, chips can multiplex their signals, sending them to neurons and synapses serially, but such sequential processing introduces additional delays.
Despite these area-and-time penalties on tasks such as audio processing, computer vision, or health monitoring, state-of-the-art brain-inspired microchips have achieved power reductions up to a thousandfold compared with those of GPUs or CPUs on the same task. If we could create neurons and synapses from individual devices that are readily manufacturable instead, we might target more massive implementations while maintaining energy efficiency.
Reinventing the MOSFET for AI
Working in our laboratory in 2024, one of my students was measuring a memory circuit that consisted of one transistor and one memristor—a type of nonvolatile memory device first fabricated in 2008. The student’s memristor circuit was built from two-dimensional material atop a silicon microchip containing MOSFETs. The MOSFETs were created in a commercial foundry using fabrication technology called the 180-nanometer node, which was cutting-edge in the year 2000.
One day the student forgot to connect the bulk terminal of the transistor. What he observed was a sudden increase in current with high nonlinearity that self-relaxed when the voltage was ramped down (a phenomenon called a hysteresis loop). This was a very promising neuronlike behavior!
After a fruitless week of trying to think of an explanation for this behavior, I (Lanza) asked Pazos, then my postdoctoral fellow, to try to observe and control this phenomenon in chips without memristors. This time, we applied pulses of voltage—like the spikes a neuron would produce—instead of the ramped voltage that my student used when he first saw the peculiar behavior.
Pazos’s new data helped us understand what was going on. The key was that oft-ignored fourth, or bulk, terminal of a MOSFET. Under ordinary operation, many mobile charge carriers flitting through the channel collide with the silicon atoms, producing free pairs of electrons and holes—a process known as impact ionization. The electric field created by the potential difference between the source and the drain causes these new free electrons to drift toward the positively biased drain and the holes to move toward the bulk terminal, which is usually grounded, removing the charge without any drama.
However, when the bulk terminal of the transistor is floating—unconnected as it was in my student’s experiment—the holes produced by impact ionization cannot be driven to the ground. Instead, they accumulate in the bulk of the silicon, increasing its voltage. Then things start to get interesting.
It helps here to imagine a MOSFET as two different kinds of transistors occupying the same physical space—the intentionally constructed MOSFET and a hidden, bipolar junction transistor. A bipolar device transmits a current signal across two _p_-_n_ junctions, in this case the interfaces between the source and the channel region and the channel and the drain. This signal is in proportion to a smaller current at a third terminal in between, called the base. In our experiment, that third terminal is the bulk.
!Image 10: Diagram of a leaky integrate-and-fire neuron converting input spikes to output spikesArtificial neurons integrate input from other neurons— spikes of voltage called action potentials—according to the strength of their connections, or synapse weight. If these spikes are strong enough and overlap in time often enough, the neuron will fire off its own action potential and return to a low-voltage state. MARIO LANZA & SEBASTIAN PAZOS
To get current flowing through a bipolar transistor, you need a big enough potential difference between the base and one of the other terminals, so that current can get across the _p_-_n_ junction. Let’s say this “threshold voltage” is 0.7 volts, although the real number depends on device geometry and silicon doping. In our device, that potential difference comes from those holes that were accumulating in the bulk, because it was not connected to ground. Once it reaches the threshold voltage, the device becomes sharply conductive, producing an abrupt increase of current. This sharp current increase eventually falls off once the drain voltage is lowered, because that lowering reduces the rate at which holes are generated in the bulk. The remaining excess holes recombine with stray electrons or leak away, and finally the bulk voltage falls. This cycle of hole accumulation, current spike, and hole removal gives rise to a hysteresis loop, very much like the electrical behavior of a biological neuron as it integrates ionic currents, fires a spike, and relaxes back to its resting voltage.
Initially, we observed this behavior only in a few transistors, and the relaxation time was very different for each of them. So, to try to control it better, we adjusted the resistance of the bulk terminal using a second MOSFET. Simply setting that resistance suddenly caused all the transistors to fire at the same voltage with hardly any variability. In other words, we found we could create perfect electronic neuron behavior in a single silicon transistor by controlling the bulk contact resistance. Setting the resistance can be done by doping the silicon during fabrication, but we think the two-transistor cell—where one acts as the bulk resistance—offers much greater versatility because it allows for electronic control.
We had to make sure the phenomenon would last, otherwise such a device would be useless. To our delight, every single one of the devices we tested worked over 10 million cycles. Not even one of them failed during our tests.
The MOSFET Synapse
!Image 11: Diagram of MOSFET showing biasing to increase or decrease channel conductanceUnder ordinary operation, the relationship between gate voltage and drain current is straightforward, but the presence of electrons trapped in the gate dielectric reduces current, and the presence of trapped holes increases it. Creating these two states is done by manipulating the bulk voltage relative to the drain and source. MARIO LANZA & SEBASTIAN PAZOS
To be honest, we were amazed. Dozens of research groups and companies all around the world have spent many millions of U.S. dollars over the past 20 years trying to emulate these neural behaviors using experimental memristor-like devices and other things, with limited success, mainly due to reliability and cost issues. We managed it in the cheapest and most industry-standard device: the MOSFET. This result was so shocking that we decided to confirm it using microchips from a different foundry. It was successful: All the behaviors could be reproduced, and perfect yield was achieved once again.
We were happy with the results and had started the process of filing for a patent and writing up our findings for the journal _\_Nature\__, when our lab made another astonishing discovery: The same kind of MOSFET could act as a synapse, too!
Recall that in ordinary operation some electrons crash into silicon atoms to create pairs of electrons and holes. We noticed that at specific values of bulk resistance a significant amount of the charge from this impact ionization would get trapped in the gate dielectric. This trapped charge interferes with the flow of current through the MOSFET, effectively changing the device’s conductance. Importantly, this new conductance is stable and adjustable at will. It was then that we realized the MOSFET could also be used as an electronic synapse.
As it was in the neuron transistor, the bulk terminal was the key. A negative bulk-source voltage drives electrons into the dielectric, decreasing conductance. A positive one pushes holes in, increasing it.
From neuromorphic device to circuit to system
Here’s how the MOSFET synapse and the MOSFET neuron, together called a neurosynaptic random-access memory, or NSRAM, could work together to achieve a simple neural circuit: Say you had a circuit consisting of three synapse MOSFETs and a neuron MOSFET. The synapses have already been programmed as we’ve described, so that each has a different conductance. Spikes of voltage with different patterns and frequencies are applied to the gate of each of these transistors. What emerges from their drains are spikes of current with amplitudes modulated by the synapses conductance values.
The spikes converge at the drain of the neuron MOSFET. With each spike, impact ionization causes charge to build in the bulk of the silicon. Some of it will drain away, but if enough spikes arrive in a short enough period of time, the bulk voltage will reach a value at which the “hidden” transistor triggers a spike of current through the MOSFET. This current would then go on to become the input to other MOSFET synapses, and so on. The behavior is exactly the kind of integrate-and-fire action real neural circuits deliver.
The competitive advantage of our single-MOSFET electronic neurons and synapses is straightforward: We can produce with only one or two transistors the electronic signals that today require, at an industrial level, dozens and sometimes even hundreds of components. And moreover, unlike other emerging technologies, our solution is fully compatible with today’s silicon manufacturing lines and exhibits a yield of 100 percent in key figures of merit with near-zero variability.
Building functional circuits for brain-inspired computing and AI based on this technology is as exciting as it is laborious. It will require us to improve our computer models to resemble the behavior of both devices more accurately and to do so with computational efficiency. We must also perform accurate circuit- and system-level simulations to validate computing architectures, design peripheral circuitry to drive and convert signals, and undergo multiple fabrication rounds to optimize performance.
But all that will be worthwhile, because it could result in brain-inspired microchips for AI with better energy efficiencies than what we have now. These chips will first be a fit for smaller-scale, “edge-AI” tasks, such as bringing greater intelligence to battery-powered systems. But if we can scale up such chips, maybe in the long run they can compete with state-of-the-art GPUs.
From Your Site Articles
- Neuromorphic Chips Are Destined for Deep Learning—or Obscurity ›
- New Hardware Mimics Spiking Behavior of Neurons With High Efficiency ›
Related Articles Around the Web
- A CMOS Spiking Neuron for Brain-Inspired Neural ... ›
Keep Reading ↓
{"imageShortcodeIds":[]}
AICareersSponsored ArticleComputing
Why Does a Bank Need a Chief Scientist?
Inside Capital One’s quest to transform finance for millions of Americans through AI-driven innovation
25 Jun 2026
6 min read
[](javascript:;) 5
!Image 12: Silhouetted team working on laptops in a glass-walled office at sunset.
Capital One is building a scientific community and research organization to advance the frontier of AI and scientific discovery in finance.
Adobe Stock
[](mailto:?subject=Why%20Does%20a%20Bank%20Need%20a%20Chief%20Scientist%3F&body=https://spectrum.ieee.org/capital-one-science-ai-finance-innovation)
_This article is brought to you by Capital One._
After five years leading natural language understanding and eventually the entire Alexa AI organization at Amazon, Prem Natarajan made a nontraditional move: He became Chief Scientist at a bank. Not just any bank: Capital One, a financial institution serving over 100 million customers, helping everyday Americans manage their financial lives.
For Natarajan, a veteran of DARPA-funded research and academia who had watched machine learning evolve from task-specific applications to foundation models, the logic was clear. Some of the most interesting advances in AI research and deployment were shifting from big tech’s horizontal platforms to industry verticals like finance, where the most complex problems aren’t just building models but making AI work under the constraints of real-world customer problems, contextual business knowledge, continuous learning, with an incredibly high bar for accuracy and privacy.
That’s also what made Capital One the right place to do it. For decades, the company has been recognized as one of the most data- and analytics-driven financial institutions in the industry. Its business model from the very beginning was built around using data and technology to personalize financial products for customers. A decade ago, Capital One went all in on the cloud and rebuilt its data ecosystem, creating a unified environment for data, compute, and AI and machine learning experimentation. Today, its modern infrastructure, disciplined approach to governance, and deep bench of talent form the foundation that allows it to lead in enterprise AI.
Advances in AI research and deployment are shifting from big tech’s horizontal platforms to industry verticals like finance.
So, why does a bank need a Chief Scientist? The answer lies in a fundamental misconception about AI in financial services. Most financial institutions still view AI as a technology to deploy – leveraging the latest large language model, deploying it through APIs, and integrating it into existing workflows – rather than a scientific discipline. Capital One is doing something different: building a scientific community and research organization to solve real-world customer problems and invent impactful AI solutions that don’t yet exist.
While widely available foundation models can handle general tasks, they can’t yet solve many domain-specific challenges, such as detecting fraud in real-time across billions of transactions, or providing state-of-the-art conversational tools so customers can engage when, how, and where they want to.
These challenges of making AI reliable, scalable, and well governed require original research and scientific innovation that is funneled back into the business to create real-world applications to address customer needs.
The Constraints That Demand Innovation
!Image 13: Headshot of a suited man against a blue gradient background.Prem Natarajan, an IEEE Fellow, is Chief Scientist at Capital One. “If you want to solve really important problems in AI and see your work come to life, this is one of the few places you can do that,” he says.Capital One
Because banks are dealing with people’s finances, there is an incredibly high bar for getting it right when it comes to AI. Take fraud, for example. Even a minor fraud event can have a devastating impact on certain customers. The best fraud models and platforms can detect and help mitigate fraud in the time it takes someone to tap their card, which is table stakes for protecting customers and their financial information with accuracy and speed. Looking at these types of challenges, Capital One and Natarajan saw that serving millions of customers meant solving AI problems at a scale and complexity that many enterprises don’t encounter. These same constraints create a unique research environment.
At Capital One, the approach to building AI is to provide value to customers in ways never possible before, improving their financial lives and meeting them where they are with services they actually need. That focus, combined with massive scale and world-class risk management requirements, makes the scientific problems both harder and just as consequential as those found in most big tech labs.
Advancing AI Through “Destination-Back Thinking”
Capital One’s approach to AI research and innovation starts with what Natarajan calls “destination-back thinking.” Rather than asking what’s possible with current technology, the team envisions the customer experience they want to deliver – perhaps a car buyer who works long days and can only research the options at 10 p.m., or a customer facing an unexpected expense who needs immediate, personalized guidance – and then works backward to identify the scientific breakthroughs required to get there.
“You’re thinking back from where you’re providing incredibly valuable services,” Natarajan explains. “Once you have that vision clearly, you work back and say, what are the gaps? What are the things we need to invent?” This ensures that when problems are solved, the impact is essentially guaranteed, because the team has already identified what will make a tangible difference in customers’ lives.
But methodology alone isn’t enough. Capital One’s nearly 15-year bet on cloud-first architecture created something rare in financial services: a unified data and compute ecosystem that can support the kind of scientific experimentation typically seen in big tech research labs. As the only major U.S. bank to go all-in on public cloud infrastructure, Capital One eliminated the legacy systems that can constrain AI research at most financial institutions. This modern tech stack enables rapid iteration, large-scale model training, and what Natarajan calls “continuous learning,” systems that improve after deployment rather than degrading over time. This unique approach to infrastructure is a critical component in making new categories of research possible.
Agentic AI: From Research to Production
The research agenda manifests in systems already serving customers. Early last year, Capital One launched what may be the first fully agentic AI customer service experience built entirely in-house by a bank: a car buying tool that takes actions on behalf of customers based on their requests, not just answers questions. Behind it lies extensive research into multi-agentic AI reasoning systems that can navigate real-time data, business knowledge, constraints, and guardrails, with various agents that can work together to accomplish complex tasks.
Capital One has launched a fully agentic AI customer service experience powered by extensive research into multi-agentic reasoning systems that can navigate real-time data.
The team is also working on solving things like tokenization challenges, protecting sensitive data while enabling model training. To accelerate this cutting-edge work, Capital One has established partnerships with Columbia University, the University of Southern California, and the University of Illinois, and became the only bank funding NSF’s national AI research centers in 2025, investing millions in initiatives that span mental health, materials discovery, science, technology, engineering, and mathematics education, human-AI collaboration, and drug development.
In the spring of 2026, the company hosted its inaugural AI Symposium to deepen connections and foster insight-sharing between the scientific AI community, leading AI labs, startups, and its own technology, science, and AI leaders and partners.
Building a World-Class AI Organization
!Image 14: Blue \u201cCapital One\u201d wordmark with a red swoosh above the text. Capital One is building the next generation of AI talent. Join the team inventing impactful AI solutions to shape the future of finance. Learn more at https://capitalone.science/
External validation suggests the strategy is working. Evident AI ranked Capital One as the leading bank in AI talent and a global leader in AI innovation for three consecutive years, noting the bank accounted for 38 percent of all AI patents filed by the top 50 financial institutions. Capital One was also recognized by IFI Insights as the only financial institution among the top U.S. patent leaders in agentic and generative AI in 2025, alongside the likes of Google, NVIDIA, DeepMind, IBM, Microsoft, Intel, Adobe and Samsung. Capital One’s AI team – which has experience from leading AI labs and top universit