The 10 most innovative computing companies of 2025

The heated race to develop and deploy new large language models and AI products has seen innovation surge—and revenue soar—at companies supporting AI infrastructure. This year’s Most Innovative Companies in computing include TSMC; the Taiwan-based fabricator’s N3P chip offers the smallest, most densely packed transistor size yet, while the company Chip-on-Wafer-on-Substrate (CoWoS) packaging technology is integral to AI accelerator chips, including Nvidia’s Blackwell GPU. Lambda Labs’ new 1-Click service provides on-demand, self-serve GPU clusters for large-scale model training without long-term contracts. SambaNova Systems takes another tack with its SambaNova Cloud, an “AI inference” service. Powered by the company’s specialized RDU (reconfigurable dataflow unit) processor, the service makes running AI workloads (as opposed to AI model training) faster and more efficient than on GPU-powered systems.

Commercial data centers are also transforming to meet the demands of AI and high-performance computing applications. Aligned Data Centers has rolled out next-generation liquid-cooling systems, which use far less water and energy than air cooling, across its operations. Oxide Computer’s hyperscale cloud computers combine compute, storage, and networking elements in a single plug-and-play package, allowing security-sensitive customers to run their own private cloud server within a data center. And Sima.ai introduced novel multi-modal AI chips for edge computing; they can analyze inputs from text, computer-vision images, and audio for use in self-driving vehicles, robotics, smart retail, healthcare, and other applications.

Apart from AI, some of the most exciting technology developments took place in quantum computing, with companies large and small taking different approaches to building utility-scale systems. They included trapped-ion specialist Quantinuum and “neutral atom”-focused startup Atom Computing. Following its 2023 unveiling of two of the largest quantum computers ever made, IBM expanded and upgraded its cloud quantum service and its open-source Qiskit quantum software platform in 2024.

1. IBM

For growing the quantum computing ecosystem with partners that can broaden access to a large developer community

After its breakthroughs in quantum computing hardware at the end of 2023, Big Blue moved to cement its place as a leader in the quantum ecosystem in 2024, supporting a growing community of developers with software tools and access to computing resources. In December of 2023, IBM debuted the IBM Quantum Heron, a 156-superconducting-qubit system that demonstrated the company’s highest performance metrics to date, offering 16 times better performance and a 25-fold increase in speed over its 2022 quantum systems. It also unveiled its IBM Quantum System Two, a next-generation modular architecture that it hopes to scale in ever larger configurations. In September, IBM expanded its Quantum Data Center in Poughkeepsie, New York, which operates the most utility-scale quantum computers at a single location. And in October, the company opened its first global quantum data center outside the U.S., in Ehningen, Germany.

Aiming to build a critical mass of quantum developers to support the ecosystem, IBM also upgraded its quantum software toolkit Qiskit with user-friendly features in April, and released an open-source benchmarking tool called Benchpress for comparing the performance of different kinds of quantum hardware and software. The same month, IBM delivered its first quantum computer on a university campus, at the Rensselaer Polytechnic Institute in Troy, New York. This November, at its first-ever Quantum Developer Conference , IBM announced that revisions to the Heron system have made it possible to run much bigger quantum “routines,” involving up to 5,000 two-qubit gates, a scale that the company says can deliver real scientific discoveries and push toward “quantum advantage” (solving problems that aren’t feasible with classical computers). IBM says that some 600,000 people have registered to use its quantum systems.

2. Lambda Labs

For accelerating AI development by sharing GPUs with everyone 

AI breakthroughs have traditionally relied on access to lots of GPUs–capacity that has been available mostly through expensive, long-term contracts with long wait times for service. Lambda Labs, which provides computing to support deep-learning applications, aims to change that. In 2024, it launched 1-Click Clusters, a service that provides AI startups and engineers with the first on-demand, self-serve GPU clusters for AI model training. With 1-Click Clusters, users of any size can get on-demand access to state-of-the-art Nvidia H100 Tensor Core GPUs and GH200 Superchips in a public cloud, enabling large-scale model training without having to lock in long-term contracts. Lambda Labs’ AI Cloud is used by companies and research institutions that include Intel, Microsoft, Amazon, Stanford, Harvard, Caltech, and the Department of Defense. In August, the company announced a partnership with SK Telecom to expand cloud services in South Korea.

In December, the company announced the launch of Lambda Inference API, which claims to be the lowest-cost service of its kind on the market, allowing enterprise customers to deploy AI models and applications for end users without having to procure compute power themselves. In February, Lambda Labs raised a $320 million Series C venture round. The company ended 2024 with more than 10,000 customers. 

3. TSMC

For pushing transistor density to the limit. Again.

TSMC is the world’s leading dedicated semiconductor foundry and manufacturer of logic semiconductors, accounting for 28% of global semiconductor output value (excluding memory) as of 2023. With the chips it produces already powering everything from artificial intelligence and smartphones to automobiles, the company continues to deliver breakthroughs.

For AI, transistor density is critical for continued gains in performance and efficiency, and TSMC’s has been relentlessly minimizing. In Q4 2023, TSMC launched its state-of-the-art 3 nm (nanometer) process, which provided an 18% speed improvement and 32% power reduction over earlier 5 nm technology. In 2024, it released an upgrade, the N3P, which offers an additional 4% speed improvement and 9% power reduction. The company expects a majority of its customers’ designs to utilize N3P, which is backward-compatible with its predecessor but provides higher performance efficiency and lower cost.

TSMC is also a major supplier of so-called “CoWoS” Chip-on-Wafer-on-Substrate packaging technologies to designers of AI accelerator chips. NVIDIA’s Blackwell GPU, launched in March 2024, makes use of CoWoS to encase multiple chips together in a single package. Benefits include reducing the distance data must travel between chips, minimizing signal delay and power loss, and enhancing overall system efficiency when running large language models. TSMC had record 2024 annual revenue of $87.8 billion, a 33.9% increase over 2023. 

4. SambaNova Systems

For chips that could deliver exponential AI application gains

SambaNova Systems’ platform, designed for enterprise and government use, combines AI chips with Samba-1, an open-source large language model similar to OpenAI’s GPT-4. The heart of SambaNova Systems’ technology is its own fast, specialized processor called an RDU (reconfigurable dataflow unit) that optimizes data flow and acceleration for large language models and complex AI tasks. With a focus on applying versus training AI, the company is betting that faster, more efficient systems with lower power consumption will shift real-time applications toward RDUs, which it claims deliver 10x the speed of GPUs at 1/10th the cost.

In 2024, SambaNova added Accenture, Analog Devices, NetApp, Aramco, SoftBank, and Los Alamos National Laboratory to its customer list and expanded a partnership with the RIKEN Center for Computational Science. In September, it announced SambaNova Cloud, the world’s fastest “AI inference service” powered by its SN40L AI chip. And in November, the U.S. Department of Energy’s Oak Ridge National Laboratory (ORNL), known for running the world’s most powerful supercomputers, announced that it had deployed SambaNova Suite to assist its research with secure and energy-efficient AI.

5. Aligned Data Centers

For finding a next-gen approach to cooling its data centers

Data centers supporting AI consume around 4% of U.S. electricity, a share expected to double by the end of the decade. Core to this energy demand are cooling systems, which account for at least 40% of energy consumption. In 2024, Aligned Data Centers—which builds adaptive data centers for hyperscale and enterprise customers—rolled out DeltaFlow, a next-generation liquid cooling technology for emerging applications in AI and HPC (high-performance computing). DeltaFlow is a closed-loop water system that directly cools computer chips (“direct-to-chip”) by circulating chilled water through cold plates attached to the components, dissipating heat through a heat exchanger within a single cooling distribution unit (CDU) that integrates seamlessly with existing data center infrastructure.

Aligned has over 5,000MW of data centers in the planning and design phase or under construction. In March 2024, Blackstone loaned the company $600 million to finance the build of Aligned’s newest and largest data center, in West Jordan, Utah. It will join campuses in Chicago, Dallas, Phoenix, Salt Lake City, and Northern Virginia, with five more domestic locations in development. Aligned’s acquisition of Sao Paulo-based Odata last year expanded its footprint throughout Latin America. It has also invested in QScale, a Canadian operator of sustainable AI and HPC data centers.

6. SiMa.ai

For helping industrial robots, self-driving machines, and other machines run AI applications on their own hardware 

A burgeoning subsector of AI hardware is focused on so-called AI-at-the-edge technology, which enables local standalone processing of AI workloads in a wide range of real-world applications, including industrial robotics and autonomous vehicles. The edge AI hardware market is projected to grow from $24.2 billion in 2024 to $54.7 billion by 2029, according to research firm MarketsandMarkets.

Silicon Valley semiconductor startup Sima.ai is a leading developer of embedded machine learning system-on-chip (MLSoC) solutions. After securing $70 million in funding in April, in September the company announced Modalix, the first edge-computing chips targeted at multi-modal AI. The new SoCs, which process text, computer-vision images, and audio, can run cutting-edge reasoning models like Meta’s Llama 2-7B. Energy efficiency is critical in edge applications, and Sima.ai’s chips have demonstrated up to 85% greater efficiency compared to leading competitors. Sima is targeting the embedded edge market, the computing layer that sits between the cloud and personal devices, and is pursuing applications in healthcare, smart retail, self-driving vehicles, government, and robotics.

7. Oxide Computer Company

For giving customers control over their computing clouds 

In 2024, San Francisco Bay Area startup Oxide Computer began commercial sales of its plug-and-play hyperscale cloud computers for private data centers. Designed for customers in government, financial services, and e-commerce who want high-performance computing capabilities along with the security of local control, Oxide’s systems make the transition to the private cloud as easy as possible. Its server racks include 32 compute sleds with more than 2,000 CPU cores, with integrated storage and networking elements, in a single plug-and-play package. Each rack is also equipped with DC power busbars, which are more efficient than standard AC connection. Compared to traditional rack servers, they improve per-watt usage by 70% and energy efficiency as much as 35%, 

Oxide’s computers are delivered with all the software needed to run full cloud computing services and require no assembly, allowing users to get up and running in a matter of hours compared to months for deploying in a data center. Oxide, which cites Shopify as a major ecommerce customer, has raised $78M to date. In November, Lawrence Livermore National Laboratory selected Oxide’s hyperscale cloud system for its supercomputing center in California, providing scientific users with access to virtual machines that can help optimize demanding applications while maintaining the on-premises security necessary for their workloads. 

8. NextSilicon

For creating chips that respond to real-time computing needs

Tel Aviv chipmaker NextSilicon has faced formidable rivals like Nvidia and AMD. But with its innovative Maverick-2 chip, it is targeting the high-performance computing (HPC) niche where it sees less competition and a better fit for its parallel processing technology. Unlike fixed GPUs designed for AI and machine-learning workloads, the Maverick-2—manufactured by TSMC using a 5 nm process—is an “Intelligent Compute Accelerator” that uses advanced algorithms to dynamically reconfigure hardware based on real-time application needs. The algorithm instantly designs a temporary, dedicated chip in the hardware to run the specific software. NextSilicon says the chip offers four times the performance-per-watt of GPUs and 20 times the performance-per-watt of CPUs, while cutting operating costs by more than 50%. 

NextSilicon plans to build on the adoption of the company’s Maverick-1 chips by government agencies and academia. Launched in October, the Maverick-2 is already shipping to dozens of customers, including Sandia National Laboratories, the U.S. Department of Energy, and organizations in financial services, energy, manufacturing, and life sciences.

9. Quantinuum

For bringing quantum computing to the financial sector

Quantinuum takes a different approach to quantum computing than many other companies advancing the technology. While IBM and other leaders in quantum have focused on superconducting qubits as their basic processing unit, trapped ions—charged atomic particles that are confined and manipulated using electromagnetic fields—offer a rival approach with advantages including longer coherence time and high qubit fidelity.

In June 2024, UK and Colorado-based Quantinuum unveiled the industry’s largest-ever “trapped ion” quantum computer, with 56 trapped-ion qubits. In a May publication in Science Advances, researchers from Quantinuum, the U.S. Department of Energy’s (DOE) Argonne National Laboratory, and JPMorgan Chase reported using Quantinuum’s processors to achieve a quantum algorithmic speedup in a benchmark called the quantum approximate optimization algorithm, which has potential applications in logistics, telecommunications, financial modeling, and materials science.

The company has been collaborating with JP Morgan Chase, one of its earliest users, on quantum technology for the financial industry since 2020. In 2024, Quantinuum also partnered with HSBC, the first international bank to offer tokenized physical gold. The companies trialed the first application of quantum-secure protection against a “store now, decrypt-later” (SNDL) attack, in which digital assets protected by today’s encryption are stored with the intention of being decrypted by future quantum computers.

10. Atom Computing

For harnessing light to build high-performance qubits 

Berkeley and Boulder-based quantum computing startup Atom Computing takes a less-common approach to creating qubits—the basic unit of quantum information—using pulses of light to manipulate neutral atoms. Neutral atoms, which can be tightly packed in arrays, enable all qubits to connect with each other and can provide higher performance than fixed-qubit approaches.

In September, Microsoft and Atom announced they would accelerate development of fault-tolerant quantum supercomputers that can solve impactful problems, and that Atom Computing’s neutral-atom hardware would be integrated with the Azure Quantum compute platform. This November, Microsoft and Atom announced rapid progress, setting a record for “entanglement” by creating a quantum connection among 24 logical qubits made from neutral atoms. Entangled qubits, which interact with one another through what Einstein called “spooky action at a distance,” is central to quantum computing’s promise. Even as Microsoft unveils its own cutting-edge quantum hardware, such as its recent Majorna 1 chip that it claims creates a new state of matter, it is working toward 2025 delivery of commercial on-premises quantum systems based on Atom’s technology, with a software stack that allows usability at a range of skill levels.

Explore the full 2025 list of Fast Company’s Most Innovative Companies, 609 organizations that are reshaping industries and culture. We’ve selected the companies making the biggest impact across 58 categories, including advertising, applied AI, biotech, retail, sustainability, and more.