MIT researchers developed a photon-shuttling “interconnect” that can facilitate remote entanglement, a key step toward a practical quantum computer.
Adam Zewe | MIT News
Publication Date: March 21, 2025
Quantum computers have the potential to solve complex problems that would be impossible for the most powerful classical supercomputer to crack.
Just like a classical computer has separate, yet interconnected, components that must work together, such as a memory chip and a CPU on a motherboard, a quantum computer will need to communicate quantum information between multiple processors.
Current architectures used to interconnect superconducting quantum processors are “point-to-point” in connectivity, meaning they require a series of transfers between network nodes, with compounding error rates.
On the way to overcoming these challenges, MIT researchers developed a new interconnect device that can support scalable, “all-to-all” communication, such that all superconducting quantum processors in a network can communication directly with each other.
They created a network of two quantum processors and used their interconnect to send microwave photons back and forth on demand in a user-defined direction. Photons are particles of light that can carry quantum information.
The device includes a superconducting wire, or waveguide, that shuttles photons between processors and can be routed as far as needed. The researchers can couple any number of modules to it, efficiently transmitting information between a scalable network of processors.
They used this interconnect to demonstrate remote entanglement, a type of correlation between quantum processors that are not physically connected. Remote entanglement is a key step toward developing a powerful, distributed network of many quantum processors.
“In the future, a quantum computer will probably need both local and nonlocal interconnects. Local interconnects are natural in arrays of superconducting qubits. Ours allows for more nonlocal connections. We can send photons at different frequencies, times, and in two propagation directions, which gives our network more flexibility and throughput,” says Aziza Almanakly, an electrical engineering and computer science graduate student in the Engineering Quantum Systems group of the Research Laboratory of Electronics (RLE) and lead author of a paper on the interconnect.
Her co-authors include Beatriz Yankelevich, a graduate student in the EQuS Group; senior author William D. Oliver, the Henry Ellis Warren (1894) Professor of Electrical Engineering and Computer Science (EECS) and professor of Physics, director of the Center for Quantum Engineering, and associate director of RLE; and others at MIT and Lincoln Laboratory. The research appears today in Nature Physics.
In the new work, they took that architecture a step further by connecting two modules to a waveguide in order to emit photons in a desired direction and then absorb them at the other end.
Each module is composed of four qubits, which serve as an interface between the waveguide carrying the photons and the larger quantum processors.
The qubits coupled to the waveguide emit and absorb photons, which are then transferred to nearby data qubits.
The researchers use a series of microwave pulses to add energy to a qubit, which then emits a photon. Carefully controlling the phase of those pulses enables a quantum interference effect that allows them to emit the photon in either direction along the waveguide. Reversing the pulses in time enables a qubit in another module any arbitrary distance away to absorb the photon.
“Pitching and catching photons enables us to create a ‘quantum interconnect’ between nonlocal quantum processors, and with quantum interconnects comes remote entanglement,” explains Oliver.
“Generating remote entanglement is a crucial step toward building a large-scale quantum processor from smaller-scale modules. Even after that photon is gone, we have a correlation between two distant, or ‘nonlocal,’ qubits. Remote entanglement allows us to take advantage of these correlations and perform parallel operations between two qubits, even though they are no longer connected and may be far apart,” Yankelevich explains.
However, transferring a photon between two modules is not enough to generate remote entanglement. The researchers need to prepare the qubits and the photon so the modules “share” the photon at the end of the protocol.
Generating entanglement
The team did this by halting the photon emission pulses halfway through their duration. In quantum mechanical terms, the photon is both retained and emitted. Classically, one can think that half-a-photon is retained and half is emitted.
Once the receiver module absorbs that “half-photon,” the two modules become entangled.
But as the photon travels, joints, wire bonds, and connections in the waveguide distort the photon and limit the absorption efficiency of the receiving module.
To generate remote entanglement with high enough fidelity, or accuracy, the researchers needed to maximize how often the photon is absorbed at the other end.
“The challenge in this work was shaping the photon appropriately so we could maximize the absorption efficiency,” Almanakly says.
They used a reinforcement learning algorithm to “predistort” the photon. The algorithm optimized the protocol pulses in order to shape the photon for maximal absorption efficiency.
When they implemented this optimized absorption protocol, they were able to show photon absorption efficiency greater than 60 percent.
This absorption efficiency is high enough to prove that the resulting state at the end of the protocol is entangled, a major milestone in this demonstration.
“We can use this architecture to create a network with all-to-all connectivity. This means we can have multiple modules, all along the same bus, and we can create remote entanglement among any pair of our choosing,” Yankelevich says.
In the future, they could improve the absorption efficiency by optimizing the path over which the photons propagate, perhaps by integrating modules in 3D instead of having a superconducting wire connecting separate microwave packages. They could also make the protocol faster so there are fewer chances for errors to accumulate.
“In principle, our remote entanglement generation protocol can also be expanded to other kinds of quantum computers and bigger quantum internet systems,” Almanakly says.
This work was funded, in part, by the U.S. Army Research Office, the AWS Center for Quantum Computing, and the U.S. Air Force Office of Scientific Research.
Last year, Nvidia’s annual GTC conference—hailed as the “Woodstock of AI”—drew a crowd of 18,000 to a packed arena befitting rock legends like the Rolling Stones. On stage, CEO Jensen Huang, clad in a shiny black leather jacket, delivered his keynote for the AI chip behemoth’s annual developer’s conference with the flair of a headlining act.
Today, a year later, Huang was onstage once again, shooting off a series of T-shirt cannons and clad this time in an edgy motorcycle black leather jacket worthy of a halftime show. This time, Nvidia-watchers tossed around the metaphor of the “Super Bowl of AI” like a football. Nvidia did not shy away from the pigskin comparison, offering a keynote “pre-game” event and a live broadcast that had guest commentators like Dell CEO Michael Dell calling plays on how Nvidia would continue to rule the AI world.
As Huang took the stage in front of a stadium-sized image of the Nvidia headquarters—making sure to highlight the “gaussian splatting” 3D rendering tech behind it to his high-tech audience—his message was clear, even if unspoken: Nvidia’s best defense is a strong offense. With recent reasoning models from Chinese startup DeepSeek shaking up AI, followed by others from companies including OpenAI, Baidu and Google, Nvidia wants its business customers to know they need its GPUs and software more than ever.
That’s because DeepSeek’s R1 model, which debuted in January, created some doubts about Nvidia’s momentum. The new model, its maker claimed, had been trained for a fraction of the cost and computing power of U.S. models. As a result, Nvidia’s stock took a beating from investors worried that companies would no longer need to buy as many of Nvidia’s chips.
Reasoning models require more computing power
But Huang thinks those selling off made a big mistake. Reasoning models, he said, require morecomputing power, not less. A lot more, in fact, thanks to their more detailed answers, or in the parlance of AI folks, “inference.” The ChatGPT revolution was about a chatbot spitting out answers to queries—but today’s models must “think” harder, which requires more “tokens,” or the fundamental units text models use—whether it’s a word in a phrase or just part of a word.
The more tokens used, the more efficiency customers demand, and the more computing power AI reasoning models will require. So making sure Nvidia customers can process more tokens, faster, is the not-so-secret Nvidia play—and Huang did not need to mention DeepSeek until one hour into the keynote to get that point across.
All of the Nvidia GTC announcements that followed were positioned with that in mind. Stock-watchers might well have wanted to see an accelerated timeline for Nvidia’s new AI chip, the Vera Rubin, to be released at the end of 2026, or more details about the company’s short-term roadmap. But Huang focused on the fact that while AI pundits had insisted over the past year that the pace of AI once rapid improvements were slowing down, Nvidia believes getting AI improvements to “scale” is increasing faster than ever. Of course, that would be to Nvidia’s benefit in terms of revenue. “The amount of computation we need as a result of agentic AI, as a result of reasoning, is easily 100 times more than we thought we needed this time last year,” Huang said.
Will Nvidia’s efforts to drive growth be enough to win?
Nvidia’s announcements that followed were all about making sure customers understand they will have everything they need to keep up in a world where extreme speed at providing detailed answers and better reasoning will be the difference between a company’s AI success and failure. Blackwell GPUs, Nvidia’s latest, top of the line AI chips, are in full production—with 3.6 million of them already used. An upgraded version, the Blackwell Ultra, boasts 3x performance. The new Vera Rubin chip and infrastructure is on the way. Nvidia’s “world’s smallest AI supercomputer” is at the ready. Software for AI agents is quickly being used in the physical world, including self-driving cars, robotics, and manufacturing.
But will Nvidia’s efforts to drive growth be enough to keep enterprise companies investing in Nvidia products? Will buying Nvidia’s costly AI chips—which can cost between $30,000 to $40,000 each, prove too expensive, given the still-unclear-ROI of AI investments? Ultimately, Nvidia’s premium picks and shovels require enough customers willing to keep digging.
Huang is confident that there are enough—and that Nvidia’s Super Bowl win is not just a victory for the 31-year-old company. “Everyone wins,” he insisted.
Perhaps, but there is no doubt that as Nvidia seeks to establish a dynasty in the AI era, expectations remain higher than ever. Huang, for his part, appears undaunted even as the AI continues to evolve at high speed. He’s always reaching for the brass ring, it seems—or in this case, the Super Bowl ring.
Separating AI reality from hyped-up fiction isn’t always easy. That’s why we’ve created the AI Hype Index—a simple, at-a-glance summary of everything you need to know about the state of the industry.
The past few months have demonstrated how AI can bring us together. Meta released a model that can translate speech from more than 100 languages, and people across the world are finding solace, assistance, and even romance with chatbots. However, it’s also abundantly clear how the technology is dividing us—for example, the Pentagon is using AI to detect humans on its “kill list.” Elsewhere, the changes Mark Zuckerberg has made to his social media company’s guidelines mean that hate speech is likely to become far more prevalent on our timelines.
Estimates for how much it would cost VA to fully deploy its new electronic health record system have ranged from $16.1 billion to almost $50 billion.
The Department of Veterans Affairs needs to figure out the cost of its troubled electronic health record modernization program as it moves to restart deployments of the new software, lawmakers and witnesses warned during a House hearing on Monday.
VA initially signed a $10 billion contract — which was later revised to over $16 billion — with Cerner in May 2018 to modernize its legacy health record system and make it interoperable with the Pentagon’s new health record, which was also provided by Cerner. Oracle later acquired Cerner in 2022.
Almost as soon as the new EHR system went live in 2020 at the Mann-Grandstaff VA Medical Center in Spokane, Washington, however, the modernization project was beset by a host of problems, including cost overruns, patient safety concerns and technical glitches. VA subsequently paused deployments of the EHR system in April 2023 as part of a “reset” to address problems at the facilities where the software had been deployed.
Following the successful joint VA-DOD rollout of the EHR software at the Captain James A. Lovell Federal Health Care Center in North Chicago, Illinois last March — the sixth facility to use the new system — and citing significant improvements at the other sites using the Oracle Cerner EHR, VA announced in December that it was “beginning early-stage planning” to restart deployments in mid-2026.
Even as VA looks to implement the Oracle Cerner EHR system at four of its Michigan-based medical facilities in 2026, members of the House Veterans’ Affairs Technology Modernization Subcommittee pressed the department on Monday to pin down the long-term costs and time commitments needed to complete the modernization project.
Rep. Tom Barrett, R-Mich., the panel’s chairman, said he was not convinced VA has fixed all of the EHR program’s problems during its reset period and noted that — despite the fact VA is 7 years into the original 10-year contract for the project — “Congress has not received a schedule nor an up-to-date cost estimate to evaluate this program’s current state.”
Cost estimates for the entirety of the project have varied, with GAO Information Technology and Cybersecurity Director Carol Harris noting that these have ranged from $16.1 billion to an independent analysis that pegged the project’s total cost at almost $50 billion.
“While the latter is more realistic, neither reflects the many changes and delays to the program,” Harris said.
Seema Verma, Oracle Health’s executive vice president, said that an accelerated deployment of the new system — bolstered by progress made during the reset phase — would help drive down the project’s total cost and said that the company did not agree with the $50 billion figure cited by Harris.
The department and Oracle Cerner previously renegotiated their existing contract in May 2023 to include more accountability provisions in the agreement and to change the terms of the remaining contract from a 5-year term to five 1-year terms.
With the current contract slated to fully expire in May 2028, and VA still needing to deploy the new software to more than 160 other VA medical facilities, Neil Evans — acting program executive director of VA’s Electronic Health Record Modernization Integration Office — said the project would not be completed by that time.
To get a better understanding of the full scope of the modernization project, acting VA Inspector General David Case said the department “must develop and maintain an integrated master schedule to clearly track and project the program’s cost to completion.”
Evans told the lawmakers that the department is committed to developing a detailed integrated master schedule and updated life cycle cost estimate to help guide the modernization project’s path forward.
“We recognize the importance of providing this information to inform decisionmaking and ensure the success of future deployments,” he said.
The department said it will be adding nine sites to its 2026 deployment schedule, although it added that it will announce the specific facilities later this year.
The Department of Veterans Affairs is planning to increase deployments of its new Oracle Cerner electronic health record system to 13 VA medical facilities in 2026, with the ultimate goal of completing the troubled modernization project as early as 2031.
VA said in December that it would be rolling out the new EHR software at a total of four medical sites in mid-2026. In a Thursday release outlining its new plans, however, VA said it would announce the nine additional medical facilities slated to receive the modernized EHR system later this year, after consulting with VA officials, clinicians and Oracle Cerner representatives.
The department said it is “pursuing a market-based approach to site selection for its deployments going forward,” which will enable it “to scale up the number of concurrent deployments, while also enabling staff to work as efficiently as possible.”
The announcement comes as VA moves out of an operational pause on most rollouts of the new software that was instituted in April 2023 following a series of technical glitches, patient safety concerns and training challenges.
VA has deployed the software at just six of its 170 medical centers. One of those rollouts occurred at a joint VA-Defense Department medical facility in North Chicago last March during the agency’s “reset” period, with that deployment being seen, in part, as a crucial test of the efforts to right the modernization project.
During his January confirmation hearing, current VA Secretary Doug Collins echoed bipartisan concerns about the EHR modernization project but told lawmakers that the effort — which is intended to help streamline the delivery of medical records for servicemembers transitioning from active duty to civilian life — was necessary and that “there’s no reason in the world we cannot get this done.”
At the time, Collins also told lawmakers that he believed the department could restart deployments faster than the mid-2026 timeframe that the then-Biden administration set for resuming EHR system rollouts.
“America’s Veterans deserve a medical records system that’s integrated across all VA and DOD components, and that is exactly what we will deliver,” Collins said in a Thursday statement. “We can and will move faster on this important priority. But we’re going to listen to our doctors, nurses and vendor partners along the way in order to ensure patient safety, quality and customer service.”
Although VA said its goal is to complete all deployments of the new EHR system as soon as 2031, that timeframe already exceeds the limits of the current contract it has with Oracle Cerner.
The department initially entered into a 10-year, $10 billion contract with Cerner in May 2018 to modernize its legacy health record system. The cost of that contract was later revised to over $16 billion, and Cerner was subsequently acquired by Oracle in 2022.
VA and Oracle Cerner subsequently renegotiated their existing contract in May 2023 to include more accountability provisions in the agreement, as well as to revise the remaining contract from an additional 5-year term to five 1-year terms.
In addition to pressing VA to develop a master schedule for future EHR system deployments, lawmakers have pushed for the department to figure out the total cost of its modernization project.
During a House hearing last month, Republicans and Democrats both calledfor the VA official overseeing the EHR system’s rollout to develop an updated lifecycle cost estimate. Estimates for the project’s full completion have ranged from $16.1 billion to almost $50 billion.
Last Thursday, OpenAI released GPT-4.5, a new version of its flagship large language model. With each release of its GPT models, OpenAI has shown that bigger means better. But there has been a lot of talk about how that approach is hitting a wall—including remarks from OpenAI’s former chief scientist Ilya Sutskever. In this edition of What’s Next in Tech, find out everything you need to know about OpenAI’s latest model.
TOMORROW: The AI model market is shifting. Are you ready? Join us tomorrow, March 5, for our latest LinkedIn Live, “Disruption in the AI Model Market,” where we’ll break down the biggest changes shaping the landscape, and what they mean for you. Register for free today.
OpenAI says GPT-4.5 is its biggest and best chat model yet—but it could be the last release in the company’s classic LLM lineup.
Since the releases of its so-called reasoning models o1 and o3, OpenAI has been pushing two product lines. GPT-4.5 is part of the non-reasoning lineup—what Nick Ryder, a research scientist at the company, calls “an installment in the classic GPT series.”
All large language models pick up patterns across the billions of documents they are trained on. Smaller models learned syntax and basic facts. Bigger models can find more specific patterns like emotional cues, such as when a speaker’s words signal hostility, says Ryder: “All of these subtle patterns that come through a human conversation—those are the bits that these larger and larger models will pick up on.”
OpenAI won’t say exactly how big its new model is. But it claims the jump in scale from GPT-4o to GPT-4.5 is the same as the jump from GPT-3.5 to GPT-4o. Read the full story to learn more about how GPT-4.5 was trained, the benchmarks it has been tested on, and some of the model’s skills.
While Modular Open Systems Approach (MOSA) is familiar within the DoD, a new guidebook sheds light on statutes and policies that require it. Implementing a Modular Open Systems Approach in Department of Defense Programs guidebook delivers practical advice on planning, executing, and assessing MOSA. Office of the Under Secretary of Defense for Research and Engineering (OUSD(R&E)) Systems Engineering and Architecture (SE&A) prepared this guidebook. The office works to ensure architectures are modular and open to enhance competition, incorporate innovation, support interoperability, and enable rapid insertion of technology in DoD acquisitions, providing cutting-edge capabilities to the warfighter. Inside this new guidebook you will find: ✅ Implementation principles and best practices 📈 Benefits and challenges of MOSA 💡 Real-world insights from DoD and industry experts Learn more about MOSA: https://lnkd.in/dexBVdVN Download the guidebook here: https://lnkd.in/eQnE4m-W#MOSA#DoD#Acquisition#Innovation#OpenSystems
Ever since World War II, the US has been the global leader in science and technology—and benefited immensely from it. Research fuels American innovation and the economy in turn. Scientists around the world want to study in the US and collaborate with American scientists to produce more of that research. These international collaborations play a critical role in American soft power and diplomacy. The products Americans can buy, the drugs they have access to, the diseases they’re at risk of catching—are all directly related to the strength of American research and its connections to the world’s scientists.
That scientific leadership is now being dismantled, according to more than 10 federal workers who spoke to MIT Technology Review, as the Trump administration—spearheaded by Elon Musk’s Department of Government Efficiency (DOGE)—slashes personnel, programs, and agencies. Meanwhile, the president himself has gone after relationships with US allies.
These workers come from several agencies, including the Departments of State, Defense, and Commerce, the US Agency for International Development, and the National Science Foundation. All of them occupy scientific and technical roles, many of which the average American has never heard of but which are nevertheless critical, coordinating research, distributing funding, supporting policymaking, or advising diplomacy.
They warn that dismantling the behind-the-scenes scientific research programs that backstop American life could lead to long-lasting, perhaps irreparable damage to everything from the quality of health care to the public’s access to next-generation consumer technologies. The US took nearly a century to craft its rich scientific ecosystem; if the unraveling that has taken place over the past month continues, Americans will feel the effects for decades to come.
Most of the federal workers spoke on condition of anonymity because they were not authorized to talk or for fear of being targeted. Many are completely stunned and terrified by the scope and totality of the actions. While every administration brings its changes, keeping the US a science and technology leader has never been a partisan issue. No one predicted the wholesale assault on these foundations of American prosperity.
“If you believe that innovation is important to economic development, then throwing a wrench in one of the most sophisticated and productive innovation machines in world history is not a good idea,” says Deborah Seligsohn, an assistant professor of political science at Villanova University who worked for two decades in the State Department on science issues. “They’re setting us up for economic decline.”
The biggest funder of innovation
The US currently has the most top-quality research institutes in the world. This includes world-class universities like MIT (which publishes MIT Technology Review) and the University of California, Berkeley; national labs like Oak Ridge and Los Alamos; and federal research facilities run by agencies like the National Oceanic and Atmospheric Administration and the Department of Defense. Much of this network was developed by the federal government after World War II to bolster the US position as a global superpower.
Before the Trump administration’s wide-ranging actions, which now threaten to slash federal research funding, the government remained by far the largest supporter of scientific progress. Outside of its own labs and facilities, it funded more than 50% of research and development across higher education, according to data from the National Science Foundation. In 2023, that came to nearly $60 billion out of the $109 billion that universities spent on basic science and engineering.
Past government reports on improper spending are having a moment with Musk’s followers. What do they show?
The return on these investments is difficult to measure. It can often take years or decades for this kind of basic science research to have tangible effects on the lives of Americans and people globally, and on the US’s place in the world. But history is littered with examples of the transformative effect that this funding produces over time. The internet and GPS were first developed through research backed by the Department of Defense, as was the quantum dot technology behind high-resolution QLED television screens. Well before they were useful or commercially relevant, the development of neural networks that underpin nearly all modern AI systems was substantially supported by the National Science Foundation. The decades-long drug discovery process that led to Ozempic was incubated by the Department of Veterans Affairs and the National Institutes of Health. Microchips. Self-driving cars. MRIs. The flu shot. The list goes on and on.
In her 2013 book The Entrepreneurial State, Mariana Mazzucato, a leading economist studying innovation at University College London, found that every major technological transformation in the US, from electric cars to Google to the iPhone, can trace its roots back to basic science research once funded by the federal government. If the past offers any lesson, that means every major transformation in the future could be shortchanged with the destruction of that support.
The Trump administration’s distaste for regulation will arguably be a boon in the short term for some parts of the tech industry, including crypto and AI. But the federal workers said the president’s and Musk’s undermining of basic science research will hurt American innovation in the long run. “Rather than investing in the future, you’re burning through scientific capital,” an employee at the State Department said. “You can build off the things you already know, but you’re not learning anything new. Twenty years later, you fall behind because you stopped making new discoveries.”
A global currency
The government doesn’t just give money, either. It supports American science in numerous other ways, and the US reaps the returns. The Department of State helps attract the best students from around the world to American universities. Amid stagnating growth in the number of homegrown STEM PhD graduates, recruiting foreign students remains one of the strongest pathways for the US to expand its pool of technical talent, especially in strategic areas like batteries and semiconductors. Many of those students stay for years, if not the rest of their lives; even if they leave the country, they’ve already spent some of their most productive years in the US and will retain a wealth of professional connections with whom they’ll collaborate, thereby continuing to contribute to US science.
The State Department also establishes agreements between the US and other countries and helps broker partnerships between American and international universities. That helps scientists collaborate across borders on everything from global issues like climate change to research that requires equipment on opposite sides of the world, such as the measurement of gravitational waves.
The international development work of USAID in global health, poverty reduction, and conflict alleviation—now virtually shut down in its entirety—was designed to build up goodwill toward the US globally; it improved regional stability for decades. In addition to its inherent benefits, this allowed American scientists to safely access diverse geographies and populations, as well as plant and animal species not found in the US. Such international interchange played just as critical a role as government funding in many crucial inventions.
Ever since World War II, the US has been the global leader in science and technology—and benefited immensely from it. Research fuels American innovation and the economy in turn. Scientists around the world want to study in the US and collaborate with American scientists to produce more of that research. These international collaborations play a critical role in American soft power and diplomacy. The products Americans can buy, the drugs they have access to, the diseases they’re at risk of catching—are all directly related to the strength of American research and its connections to the world’s scientists.
That scientific leadership is now being dismantled, according to more than 10 federal workers who spoke to MIT Technology Review, as the Trump administration—spearheaded by Elon Musk’s Department of Government Efficiency (DOGE)—slashes personnel, programs, and agencies. Meanwhile, the president himself has gone after relationships with US allies.
Advertisement
null
These workers come from several agencies, including the Departments of State, Defense, and Commerce, the US Agency for International Development, and the National Science Foundation. All of them occupy scientific and technical roles, many of which the average American has never heard of but which are nevertheless critical, coordinating research, distributing funding, supporting policymaking, or advising diplomacy.
They warn that dismantling the behind-the-scenes scientific research programs that backstop American life could lead to long-lasting, perhaps irreparable damage to everything from the quality of health care to the public’s access to next-generation consumer technologies. The US took nearly a century to craft its rich scientific ecosystem; if the unraveling that has taken place over the past month continues, Americans will feel the effects for decades to come.
Most of the federal workers spoke on condition of anonymity because they were not authorized to talk or for fear of being targeted. Many are completely stunned and terrified by the scope and totality of the actions. While every administration brings its changes, keeping the US a science and technology leader has never been a partisan issue. No one predicted the wholesale assault on these foundations of American prosperity.
“If you believe that innovation is important to economic development, then throwing a wrench in one of the most sophisticated and productive innovation machines in world history is not a good idea,” says Deborah Seligsohn, an assistant professor of political science at Villanova University who worked for two decades in the State Department on science issues. “They’re setting us up for economic decline.”
The US currently has the most top-quality research institutes in the world. This includes world-class universities like MIT (which publishes MIT Technology Review) and the University of California, Berkeley; national labs like Oak Ridge and Los Alamos; and federal research facilities run by agencies like the National Oceanic and Atmospheric Administration and the Department of Defense. Much of this network was developed by the federal government after World War II to bolster the US position as a global superpower.
Before the Trump administration’s wide-ranging actions, which now threaten to slash federal research funding, the government remained by far the largest supporter of scientific progress. Outside of its own labs and facilities, it funded more than 50% of research and development across higher education, according to data from the National Science Foundation. In 2023, that came to nearly $60 billion out of the $109 billion that universities spent on basic science and engineering.
Past government reports on improper spending are having a moment with Musk’s followers. What do they show?
The return on these investments is difficult to measure. It can often take years or decades for this kind of basic science research to have tangible effects on the lives of Americans and people globally, and on the US’s place in the world. But history is littered with examples of the transformative effect that this funding produces over time. The internet and GPS were first developed through research backed by the Department of Defense, as was the quantum dot technology behind high-resolution QLED television screens. Well before they were useful or commercially relevant, the development of neural networks that underpin nearly all modern AI systems was substantially supported by the National Science Foundation. The decades-long drug discovery process that led to Ozempic was incubated by the Department of Veterans Affairs and the National Institutes of Health. Microchips. Self-driving cars. MRIs. The flu shot. The list goes on and on.
In her 2013 book The Entrepreneurial State, Mariana Mazzucato, a leading economist studying innovation at University College London, found that every major technological transformation in the US, from electric cars to Google to the iPhone, can trace its roots back to basic science research once funded by the federal government. If the past offers any lesson, that means every major transformation in the future could be shortchanged with the destruction of that support.
The Trump administration’s distaste for regulation will arguably be a boon in the short term for some parts of the tech industry, including crypto and AI. But the federal workers said the president’s and Musk’s undermining of basic science research will hurt American innovation in the long run. “Rather than investing in the future, you’re burning through scientific capital,” an employee at the State Department said. “You can build off the things you already know, but you’re not learning anything new. Twenty years later, you fall behind because you stopped making new discoveries.”
A global currency
The government doesn’t just give money, either. It supports American science in numerous other ways, and the US reaps the returns. The Department of State helps attract the best students from around the world to American universities. Amid stagnating growth in the number of homegrown STEM PhD graduates, recruiting foreign students remains one of the strongest pathways for the US to expand its pool of technical talent, especially in strategic areas like batteries and semiconductors. Many of those students stay for years, if not the rest of their lives; even if they leave the country, they’ve already spent some of their most productive years in the US and will retain a wealth of professional connections with whom they’ll collaborate, thereby continuing to contribute to US science.
The State Department also establishes agreements between the US and other countries and helps broker partnerships between American and international universities. That helps scientists collaborate across borders on everything from global issues like climate change to research that requires equipment on opposite sides of the world, such as the measurement of gravitational waves.
The international development work of USAID in global health, poverty reduction, and conflict alleviation—now virtually shut down in its entirety—was designed to build up goodwill toward the US globally; it improved regional stability for decades. In addition to its inherent benefits, this allowed American scientists to safely access diverse geographies and populations, as well as plant and animal species not found in the US. Such international interchange played just as critical a role as government funding in many crucial inventions.
Several federal agencies, including the Centers for Disease Control and Prevention, the Environmental Protection Agency, and the National Oceanic and Atmospheric Administration, also help collect and aggregate critical data on disease, health trends, air quality, weather, and more from disparate sources that feed into the work of scientists across the country.
The National Institutes of Health, for example, has since 2015 been running the Precision Medicine Initiative, the only effort of its kind to collect extensive and granular health data from over 1 million Americans who volunteer their medical records, genetic history, and even Fitbit data to help researchers understand health disparities and develop personalized and more effective treatments for disorders from heart and lung disease to cancer. The data set, which is too expensive for any one university to assemble and maintain, has already been used in hundreds of papers that will lay the foundation for the next generation of life-saving pharmaceuticals.
Beyond fueling innovation, a well-supported science and technology ecosystem bolsters US national security and global influence. When people want to study at American universities, attend international conferences hosted on American soil, or move to the US to work or to found their own companies, the US stays the center of global innovation activity. This ensures that the country continues to get access to the best people and ideas, and gives it an outsize role in setting global scientific practices and priorities. US research norms, including academic freedom and a robust peer review system, become global research norms that lift the overall quality of science. International agencies like the World Health Organization take significant cues from American guidance.
What’s coming next for technologies like EVs and wind power?
US scientific leadership has long been one of the country’s purest tools of soft power and diplomacy as well. Countries keen to learn from the American innovation ecosystem and to have access to American researchers and universities have been more prone to partner with the US and align with its strategic priorities.
Just one example: Science diplomacy has long played an important role in maintaining the US’s strong relationship with the Netherlands, which is home to ASML, the only company in the world that can produce the extreme ultraviolet lithography machines needed to produce the most advanced semiconductors. These are critical for both AI development and national security.
International science cooperation has also served as a stabilizing force in otherwise difficult relationships. During the Cold War, the US and USSR continued to collaborate on the International Space Station; during the recent heightened economic competition between the US and China, the countries have remained each other’s top scientific partners. “Actively working together to solve problems that we both care about helps maintain the connections and the context but also helps build respect,” Seligsohn says.
The federal government itself is a significant beneficiary of the country’s convening power for technical expertise. Among other things, experts both inside and outside the government support its sound policymaking in science and technology. During the US Senate AI Insight Forums, co-organized by Senator Chuck Schumer through the fall of 2023, for example, the Senate heard from more than 150 experts, many of whom were born abroad and studying at American universities, working at or advising American companies, or living permanently in the US as naturalized American citizens.
Federal scientists and technical experts at government agencies also work on wide-ranging goals critical to the US, including building resilience in the face of an increasingly erratic climate; researching strategic technologies such as next-generation battery technology to reduce the country’s reliance on minerals not found in the US; and monitoring global infectious diseases to prevent the next pandemic.
“Every issue that the US faces, there are people that are trying to do research on it and there are partnerships that have to happen,” the State Department employee said.
A system in jeopardy
Now the breadth and velocity of the Trump administration’s actions has led to an unprecedented assault on every pillar upholding American scientific leadership.
For starters, the purging of tens of thousands—and perhaps soon hundreds of thousands—of federal workers is removing scientists and technologists from the government and paralyzing the ability of critical agencies to function. Across multiple agencies, science and technology fellowship programs, designed to bring in talented early-career staff with advanced STEM degrees, have shuttered. Many other federal scientists were among the thousands who were terminated as probationaryemployees, a status they held because of the way scientific roles are often contractually structured.
“Diplomacy is built on relationships. If we’ve closed all these clinics and gotten rid of technical experts in our knowledge base inside the government, why would any foreign government have respect for the US in our ability to hold our word and in our ability to actually be knowledgeable?” a terminated USAID worker said. “I really hope America can save itself.”
Now the Trump administration has sought to reverse some terminations after discovering that many were key to national security, including nuclear safety employees responsible for designing, building, and maintaining the country’s nuclear weapons arsenal. But many federal workers I spoke to can no longer imagine staying in the public sector. Some are considering going into industry. Others are wondering whether it will be better to move abroad.
“It’s just such a waste of American talent,” said Fiona Coleman, a terminated federal scientist, her voice cracking with emotion as she described the long years of schooling and training she and her colleagues went through to serve the government.
No vaccine is perfect, but these medicines are still saving millions of lives.
Many fear the US has also singlehandedly kneecapped its own ability to attract talent from abroad. Over the last 10 years, even as American universities have continued to lead the world, many universities in other countries have rapidly leveled up. That includes those in Canada, where liberal immigration policies and lower tuition fees have driven a 200% increase in international student enrollment over the last decade, according to Anna Esaki-Smith, cofounder of a higher-education research consultancy called Education Rethink and author of Make College Your Superpower.
Germany has also seen an influx, thanks to a growing number of English-taught programs and strong connections between universities and German industry. Chinese students, who once represented the largest share of foreign students in the US, are increasingly staying at home or opting to study in places like Hong Kong, Singapore, and the UK.
During the first Trump administration, many international students were already more reluctant to come to the US because of the president’s hostile rhetoric. With the return and rapid escalation of that rhetoric, Esaki-Smith is hearing from some universities that international students are declining their admissions offers.
Add to that the other recent developments—the potential dramatic cuts in federal research funding, the deletion of scores of rich public data sets on health and the environment, the clampdown on academic freedom for research that appears related to diversity, equity, and inclusion and the fear that these restrictions could ultimately encompass other politically charged topics like climate change or vaccines—and many more international science and engineering students could decide to head elsewhere.
“I’ve been hearing this increasingly from several postdocs and early-career professors, fearing the cuts in NIH or NSF grants, that they’re starting to look for funding or job opportunities in other countries,” Coleman told me. “And then we’re going to be training up the US’s competitors.”
The attacks could similarly weaken the productivity of those who stay at American universities. While many of the Trump administration’s actions are now being halted and scrutinized by US judges, the chaos has weakened a critical prerequisite for tackling the toughest research problems: a long-term stable environment. With reports that the NSF is combing through research grants for words like “women,” “diverse,” and “institutional” to determine whether they violate President Trump’s executive order on DEIA programs, a chilling effect is also setting in among federally funded academics uncertain whether they’ll get caught in the dragnet.
To scientists abroad, the situation in the US government has marked American institutions and researchers as potentially unreliable partners, several federal workers told me. If international researchers think collaborations with the US can end at any moment when funds are abruptly pulled or certain topics or keywords are suddenly blacklisted, many of them could steer clear and look to other countries. “I’m really concerned about the instability we’re showing,” another employee at the State Department said. “What’s the point in even engaging? Because science is a long-term initiative and process that outlasts administrations and political cycles.”
Meanwhile, international scientists have far more options these days for high-caliber colleagues to collaborate with outside America. In recent years, for example, China has made a remarkable ascent to become a global peer in scientific discoveries. By some metrics, it has even surpassed the US; it started accounting for more of the top 1% of most-cited papers globally, often called the Nobel Prize tier, back in 2019 and has continued to improve the quality of the rest of its research.
Where Chinese universities can also entice international collaborators with substantial resources, the US is more limited in its ability to offer tangible funding, the State employee said. Until now, the US has maintained its advantage in part through the prestige of its institutions and its more open cultural norms, including stronger academic freedom. But several federal scientists warn that this advantage is dissipating.
“America is made up of so many different people contributing to it. There’s such a powerful global community that makes this country what it is, especially in science and technology and academia and research. We’re going to lose that; there’s not a chance in the world that we’re not going to lose that through stuff like this,” says Brigid Cakouros, a federal scientist who was also terminated from USAID. “I have no doubt that the international science community will ultimately be okay. It’ll just be a shame for the US to isolate themselves from it.”
Quantum computing is maturing at a rapid pace, and it is quite plausible that quantum computers capable of solving problems of value to businesses will be available this decade. At the same time, quantum computing likely will not supplant classical computing in the foreseeable future—after all, quantum computing architectures are best equipped to solve certain problems, but not every problem. Quantum computers almost certainly will work in concert with classical processing, where each computing architecture will handle those parts of a calculation that it is best suited to tackle. For that to happen, the quantum computing hardware will require software that combines quantum with classical computing. It also must be devised in a user-friendly way so that nonquantum scientists and software developers working on problems such as modeling molecules with unprecedented accuracy and calculating interesting properties of structured datasets can run quantum computational tasks without having in-depth knowledge of quantum computing.
Briefly, quantum computers solve problems coded in the form of qubits (short for quantum bits; these are units of information). This information is processed by complex hardware consisting of trapped atoms, artificial atoms engineered from superconducting wire—wire that can carry current without resistance—or other physical systems that can be put into quantum states. A qubit chip can look much like a classical computing chip but is capable of different kinds of mathematical operations beyond the binary code and logic operations of classical computing. By incorporating properties of quantum mechanics such as superposition (a set of qubits can be in multiple states at the same time until they are measured), interference (some of those states can cancel out), and entanglement (the ability to create correlations between qubits inaccessible to classical computers), quantum information can be processed in ways that are fundamentally different from how we compute with classical bits, in which the basic units of information exist in one of two distinct states, 1 and 0. This could accelerate the development of computer systems that can exactly predict the behavior of real-world natural systems such as chemical reactions, or perform algebra using a computer with exponentially more computational space than a classical computer (like optimizing energy grids).
The core unit of a quantum computation is the quantum circuit, which refers not to a physical circuit and electronic parts but rather to a computational routine that runs on a quantum processing unit instead of a central processing unit (the core computational unit of a classical computer). A quantum circuit begins when we encode information in qubits, then apply a sequence of operations—the quantum equivalent of a classical computer’s “IF,” “AND,” and “OR— to those qubits. It ends when we “measure” the qubits, receiving the output of the calculation. Measuring a qubit projects it onto classical states (known as “collapsing the wave function”), meaning that quantum circuits can only output a single string of binary code as a result of the execution. A quantum circuit may also include concurrent classical computing. For example, individual qubits might be measured in the midst of a computational procedure with the result stored as classical bits, and then fed back into the same circuit later. Unlike a classical routine, quantum circuits are innately probabilistic—different runs of the same circuit may lead to different output strings of binary digits, based on the probabilities determined at the time of measurement. For example, an algorithm that estimates the energy of a molecule with a quantum computer may result in a distribution of possible energies, which can be averaged into an expectation value.
The complexity of a quantum circuit is determined as a function of the number of qubits (width) and importantly, the number of quantum instructions that the circuit can run before the qubits can no longer accurately store quantum information (depth). In practice, depth is limited by properties such as noise from external sources (a phenomenon known as decoherence), or the process in which qubits “forget” their quantum information. Depth and width are critical for determining the potential types of problems that might be addressed by quantum computers, though these parameters are also moving targets as hardware improves.
On another front, quantum computing researchers are searching for techniques that allow quantum computers to correct errors in real time, called error correction, while working to build hardware less prone to noise and thus capable of running more complex quantum routines. At the same time, thanks to hardware advances and new postprocessing techniques called error mitigation, quantum computers can now use quantum circuits to run calculations that cannot be exactly simulated using more computationally expensive brute-force classical methods for certain types of problems in chemistry. Known as utility-scale problems, these typically require quantum circuits with 100+ qubits and 1000+ quantum gates. Although today’s quantum computers are not mature enough to run certain important quantum tasks such as the famous Shor’s algorithm used to factor numbers, they still have the potential to provide more timely value for research problems with the help of error mitigation. Several such methods are already promising, especially in the field of chemistry, in which techniques such as sample-based quantum diagonalizationmay be used to calculate the properties of molecules. Other algorithms for chemistry, data analysis, and optimization look encouraging for the near future.
These methods necessitate high-performance quantum software. Utility-scale quantum algorithms do not use just one quantum circuit—they typically require running a quantum circuit many times to sufficiently sample from a distribution of possible solutions. Furthermore, most programs incorporating quantum computation require a combination of both quantum and classical processing. Therefore, partitioning problems between quantum and classical processing hardware requires frequent data exchange. This strategy breaks down if the software is too slow. Therefore, software cannot simply be powerful enough to run quantum workloads efficiently on quantum computers. It must also be designed so that it can perform quickly and efficiently when quantum and classical processing are working together.
To make all of this more practical to potential users inside and outside of the field, developers are building and maintaining high-performing quantum software development kits (SDKs) such as IBM’s Qiskit, Quantinuum’s TKET, Google’s Cirq, and others. At the same time, developers have created universal circuits to use as benchmarks in order for these SDKs to track their performance—their ability to run these circuits quickly and efficiently. These benchmarks include QASMBench circuits, Feynman circuits, and Hamiltonian circuits. IBM maintains an open-source package that adapts these circuits to more than 1000 tests for benchmarking the performance of quantum SDKs in order to compare Qiskit to its competitors.
It is important that quantum SDKs remain open and transparent so that users can continue to measure their ability to run these and other circuits. Furthermore, maintainers of these quantum SDKs should use open-source tools for performance comparisons and publish their results publicly, not only so that developers can monitor the continuing development of quantum SDKs, but also so that the quantum community can work together to keep benchmarks relevant.
Software must be more than fast and efficient. In classical computing, software developers don’t have to reach down into the code and program individual bits. They can use a higher-level, more abstract language to harness the computer to carry out the desired tasks. In the same way, maintainers of quantum software development kits must create tools that “abstract away” the details of quantum circuits so that users don’t need to learn the intricacies of quantum computing hardware to write quantum code. These kinds of higher-level abstractions are beginning to emerge in today’s quantum SDK.
Over time, quantum software development kits should broaden their areas of application, moving toward domain-specific libraries of functions akin to those now used in fields such as chemistry simulation, machine learning, and optimization. Again, this will allow domain experts to integrate quantum computing without requiring deep quantum computing knowledge.
All of these requirements are critical in the search for quantum advantage, the point where quantum computers can provide substantial improvement to some problems now only feasible with slower classical computation. It will be then that useful quantum computing is brought to the world.
No single company or individual will bring about this new era. This is a global endeavor requiring a collaboration of physicists, engineers, developers, entrepreneurs, government officials, and more. It’s time to get started.
Amazon Web Services today announced Ocelot, its first-generation quantum computing chip. While the chip has only rudimentary computing capability, the company says it is a proof-of-principle demonstration—a step on the path to creating a larger machine that can deliver on the industry’s promised killer applications, such as fast and accurate simulations of new battery materials.
“This is a first prototype that demonstrates that this architecture is scalable and hardware-efficient,” says Oskar Painter, the head of quantum hardware at AWS, Amazon’s cloud computing unit. In particular, the company says its approach makes it simpler to perform error correction, a key technical challenge in the development of quantum computing.
Ocelot consists of nine quantum bits, or qubits, on a chip about a centimeter square, which, like some forms of quantum hardware, must be cryogenically cooled to near absolute zero in order to operate. Five of the nine qubits are a type of hardware that the field calls a “cat qubit,” named for Schrödinger’s cat, the famous 20th-century thought experiment in which an unseen cat in a box may be considered both dead and alive. Such a superposition of states is a key concept in quantum computing.
The cat qubits AWS has made are tiny hollow structures of tantalum that contain microwave radiation, attached to a silicon chip. The remaining four qubits are transmons—each an electric circuit made of superconducting material. In this architecture, AWS uses cat qubits to store the information, while the transmon qubits monitor the information in the cat qubits. This distinguishes its technology from Google’s and IBM’s quantum computers, whose computational parts are all transmons.
Notably, AWS researchers used Ocelot to implement a more efficient form of quantum error correction. Like any computer, quantum computers make mistakes. Without correction, these errors add up, with the result that current machines cannot accurately execute the long algorithms required for useful applications. “The only way you’re going to get a useful quantum computer is to implement quantum error correction,” says Painter.
Unfortunately, the algorithms required for quantum error correction usually have heavy hardware requirements. Last year, Google encoded a single error-corrected bit of quantum information using 105 qubits.
Amazon’s design strategy requires only a 10th as many qubits per bit of information, says Painter. In work published in Nature on Wednesday, the team encoded a single error-corrected bit of information in Ocelot’s nine qubits. Theoretically, this hardware design should be easier to scale up to a larger machine than a design made only of transmons, says Painter.
This design combining cat qubits and transmons makes error correction simpler, reducing the number of qubits needed, says Shruti Puri, a physicist at Yale University who was not involved in the work. (Puri works part-time for another company that develops quantum computers but spoke to MIT Technology Review in her capacity as an academic.)
The company says it is on track to build a new kind of machine based on topological qubits.
“Basically, you can decompose all quantum errors into two kinds—bit flips and phase flips,” says Puri. Quantum computers represent information as 1s, 0s, and probabilities, or superpositions, of both. A bit flip, which also occurs in conventional computing, takes place when the computer mistakenly encodes a 1 that should be a 0, or vice versa. In the case of quantum computing, the bit flip occurs when the computer encodes the probability of a 0 as the probability of a 1, or vice versa. A phase flip is a type of error unique to quantum computing, having to do with the wavelike properties of the qubit.
The cat-transmon design allowed Amazon to engineer the quantum computer so that any errors were predominantly phase-flip errors. This meant the company could use a much simpler error correction algorithm than Google’s—one that did not require as many qubits. “Your savings in hardware is coming from the fact that you need to mostly correct for one type of error,” says Puri. “The other error is happening very rarely.”
The hardware savings also stem from AWS’s careful implementation of an operation known as a C-NOT gate, which is performed during error correction. Amazon’s researchers showed that the C-NOT operation did not disproportionately introduce bit-flip errors. This meant that after each round of error correction, the quantum computer still predominantly made phase-flip errors, so the simple, hardware-efficient error correction code could continue to be used.
AWS began working on designs for Ocelot as early as 2021, says Painter. Its development was a “full-stack problem.” To create high-performing qubits that could ultimately execute error correction, the researchers had to figure out a new way to grow tantalum, which is what their cat qubits are made of, on a silicon chip with as few atomic-scale defects as possible.
It’s a significant advance that AWS can now fabricate and control multiple cat qubits in a single device, says Puri. “Any work that goes toward scaling up new kinds of qubits, I think, is interesting,” she says. Still, there are years of development to go. Other experts have predicted that quantum computers will require thousands, if not millions, of qubits to perform a useful task. Amazon’s work “is a first step,” says Puri.
She adds that the researchers will need to further reduce the fraction of errors due to bit flips as they scale up the number of qubits.
Still, this announcement marks Amazon’s way forward. “This is an architecture we believe in,” says Painter. Previously, the company’s main strategy was to pursue conventional transmon qubits like Google’s and IBM’s, and they treated this cat qubit project as “skunkworks,” he says. Now, they’ve decided to prioritize cat qubits. “We really became convinced that this needed to be our mainline engineering effort, and we’ll still do some exploratory things, but this is the direction we’re going.” (The startup Alice & Bob, based in France, is also building a quantum computer made of cat qubits.)
As is, Ocelot basically is a demonstration of quantum memory, says Painter. The next step is to add more qubits to the chip, encode more information, and perform actual computations. But they have many challenges ahead, from how to attach all the wires to how to link multiple chips together. “Scaling is not trivial,” he says.
healthcarereimagined 