The most innovative companies in computing this year reflect wide-ranging efforts to build the infrastructure that next-generation AI applications require, and to enable their deployment in virtually any imaginable location or scenario. Nvidia’s GB300 NVL72 platform, an “AI supercomputer”—incorporating GPUs, CPUs, power, cooling and networking components—delivers 50x the reasoning power of prior systems. It is already being deployed by hyperscalers and cloud platforms such as CoreWeave and Microsoft Azure.
This year saw a proliferation of data centers to power AI. “Neocloud” providers such as Nebius are commoditizing and democratizing access to GPU clusters with strategic hubs across the U.S. and Europe, serving customers that include Microsoft and Meta. Crusoe began operating the first two buildings of a planned 1.2 gigawatt site for OpenAI’s and Oracle’s $500 billion Project Stargate, The Abilene, Texas facility will be one of the largest in the world, powered by wind with natural-gas turbines for backup. Meanwhile, Armada is building modular, redeployable data centers for remote and rugged industrial applications, serving customers in defense, energy, mining, telecommunications, and public infrastructure.
Tackling bottlenecks that conventional hardware can no longer solve, Ayar Labs is building optical components that replace copper wiring for faster, more efficient computing, while Sandisk’s High Bandwidth Flash memory, built specifically for AI inference workloads, delivers 8x to 16x the capacity of traditional high-bandwidth memory at a similar cost. In semiconductor manufacturing, xLight is leveraging particle-accelerator research in its “free-electron” laser, which could push past current limitations in extreme ultraviolet lithography.
This year’s list also recognizes a maturing quantum computing sector, with Google, Amazon (via AWS), and Quantinuum demonstrating novel architectures, achieving breakthroughs in error correction, and blowing past classical computing benchmarks. And PsiQuantum, having made it to the final stage of DARPA’s Quantum Benchmarking program, scored record-breaking investments.
1. Nvidia
For consistently setting the standard for chips to power AI
Chip designer NVIDIA reached a $4.5 trillion market cap in 2025 by creating the infrastructure powering the AI boom. As companies shift from training AI models to deploying them at scale, it has kept pace by designing high-performing systems that power large-scale reasoning and keep data centers smoothly humming.
In 2025, the company released its GB300 NVL72 platform, merging 72 Blackwell Ultra GPUs and 36 NVIDIA Grace CPUs in a package that functions as a plug-and-play rack-scale AI supercomputer. It integrates innovations in liquid cooling, power management, networking, and software and delivers 50x the reasoning power of previous-generation systems. Tasks that once took weeks can be completed in just hours. And each rack of the GB300 NVL72 also delivers significantly more performance per watt than past systems, while using power management algorithms intended to stabilize the grid.
In July, CoreWeave became the first hyperscaler to deploy GB300 NVL72 in the cloud, giving customers access to AI factories operating at industrial scale. In October, Microsoft Azure launched the world’s first production-scale GB300 NVL72 supercluster—integrating over 4,600 NVIDIA Blackwell Ultra GPUs designed to support OpenAI’s most demanding AI workloads. Additionally, Cisco, Dell, Hewlett Packard, Lenovo, and Supermicro are all expected to deliver a wide range of servers based on Blackwell Ultra products.
Read more about Nvidia, No. 2 on Fast Company’s list of the World’s 50 Most Innovative Companies of 2026.
2. Google
For pushing the boundaries of quantum computation
In December 2024, Google Quantum AI launched Willow —a 105-qubit superconducting quantum processor designed and manufactured at the group’s Santa Barbara lab. Tested against the random circuit sampling (RCS) benchmark (a standard metric for assessing quantum computing chips). Willow excelled—performing a computation in under five minutes that would have taken today’s fastest supercomputers 10 septillion years. More importantly, the processor demonstrated a breakthrough in the critical area of quantum error correction—with results published in Nature showing that, as more qubits are used, error rates are cut exponentially, aka “below threshold.”
While random circuit sampling is a “toy problem” with no valuable real-world applications, Google used Willow last year in what it claimed is “the first verifiable quantum advantage with a real-world application”—computing the structure of a molecule in a physics simulation—with a 13,000x speedup over the world’s fastest supercomputer.
In October, Google Quantum AI announced it was acquiring the team from Atlantic Quantum. The MIT spin-out develops modular quantum computing hardware that combines qubits and superconducting control electronics within a cryogenic system.
Read more about Google, No. 1 on Fast Company’s list of the World’s 50 Most Innovative Companies of 2026.
3. Crusoe
For building the AI factories of the future—fast
Crusoe Energy made its name by capturing wasted natural gas at oil sites and turning it into a power source for bitcoin mining operations. But in March, it sold off its entire crypto business to focus entirely on a much bigger opportunity: building and powering AI data centers.
It’s a dirty business, by and large. Crusoe’s model aims to lower energy costs, reduce emissions, and deploy new energy sources—including natural gas, renewables, and eventually nuclear—more quickly. Crusoe operates as both a power developer and an operator of large-scale “AI factory” campuses.
The company is building the first data center for OpenAI and Oracle’s $500 billion Project Stargate, in Abilene, Texas, which will be one of the largest in the world. The campus leverages abundant West Texas wind complemented by natural gas turbines for backup power. Construction is moving at a searing pace. After the project began in July 2024, the first two buildings of the planned 1.2 gigawatt site began operating in September. The second phase is expected to be completed in mid-2026.
In July, Crusoe announced a deal with energy infrastructure company Tallgrass to build a 1.8 gigawatt campus in Wyoming, with the potential to scale to 10 GW. The campus will integrate multiple energy sources, including natural gas and unspecified future renewable energy developments. The company has also partnered with Redwood Materials to use their repurposed EV batteries to power modular AI factories that integrate power, cooling, remote monitoring, and fire suppression in portable units designed to work on the network’s edge.
4. Amazon Web Services
For bringing down the cost of quantum error correction by up to 90%
In a year of big quantum computing reveals, the unveiling in February of Amazon Web Services’ Ocelot, quantum computing stood out. Developed at the AWS Center for Quantum Computing at the California Institute of Technology, Ocelot is a semiconductor-based chip that employs so-called cat qubits—named after the Schrodinger’s cat thought experiment—that greatly simplifies error correction by intrinsically suppressing some errors.
Combining cat qubits with additional quantum error correction components on its chip allows AWS to reduce quantum error correction costs by up to 90%. This suggests that practical fault-tolerant computers could require far fewer physical qubits than conventionally believed, and cut as much as five years off their development timeline.
Currently, Ocelot is an experimental tool that is not yet available for users of AWS’s Bracket quantum cloud platform. But Bracket customers—including universities, national science institutes, and companies such as Amgen and Volkswagen—can now access a full range of superconducting, trapped ion and neutral atom quantum computers thanks to new machines from Rigetti, IonQ, IQM, and Alpine Quantum Technologies.
5. Ayar Labs
For delivering light-speed connectivity that can drive next-generation compute
As the hardware driving AI and HPC applications grows in size and complexity, connectivity has become a scaling bottleneck. Traditionally, intricate copper wiring shuttles electrons across among GPUs, CPUs, and ASICs (application-specific integrated circuits) as computations unfurl. But the massive processing demands of AI involve so much data moving so swiftly, that electrons are pushed to their limits. That’s why the industry is increasingly integrating optical solutions, replacing electrons with photons that move information at the speed of light.
Ayar Labs has developed one of the leading optical “I/O chiplets” for next-generation computing platforms. The TeraPHY optical engine converts electrical inputs/outputs) into optical signals, which are then transmitted through fiber optic cables, allowing instant communication across millimeters or kilometers. Replacing copper I/O with optical engines is also a critical part of reducing rack power consumption, which NVIDIA estimates could soon reach nearly 600kW per rack—enough to power up to 600 average U.S. homes during peak evening usage—using electrical connections.
In September, Ayar Labs announced a strategic collaboration with Alchip Technologies—a major manufacturing partner of TSMC—to integrate its TeraPHY optical engines into Alchip’s high-performance integrated chips used by major cloud providers and hyperscalers. Its total funding of $370 million comes from AMD Ventures, Intel Capital, and NVIDIA, among others.
6. Nebius
For (nearly) eliminating downtime in fast-scaling AI applications
In 2025, Amsterdam-based Nebius emerged as a top player among a new breed of AI neocloud companies, contributing bits and pieces of the compute needed to deliver AI at scale. The company owns or controls a mix of proprietary data centers, GPU clusters, and hardware optimized for AI workloads, offering “inference as a service.”
In September, Nebius announced a deal to provide Microsoft with additional GPU infrastructure capacity, in a deal worth $17.4 billion over a five-year term. Additional services could bring the total contract value to more than $19 billion. In November, it announced a $3 billion, five-year deal to provide Meta with dedicated GPUs, primarily for training its Llama large language models.
Shoring up its trans-Atlantic foothold in AI compute, Nebius has added strategic U.S. capacity at 300MW facility in Vineland, New Jersey, which will provide services to Microsoft, and a colocation in Kansas City, Missouri, where in December it closed on a deal to build a $6.6 billion, 800MW project linked to a proposed $2 billion nearby power facility. In addition to its own data center in Mäntsälä, Finland, the company is currently leasing or set to lease capacity in the U.K., France, Iceland, and multiple locations in Israel.
7. Armada
For building modular data centers that convert wasted power into compute power
Armada is a full-stack edge infrastructure company delivering compute, storage, connectivity, and AI /machine learning in some of Earth’s most remote and rugged industrial environments. Leviathan—its megawatt-scale, liquid-cooled modular data center—can be installed in just weeks, scale to clusters as large as 50MW (the equivalent of more than 2,000 server rack), and then be redeployed as computing needs evolve.
By colocating with stranded and underutilized energy sources (from flared natural gas to nuclear and renewables), Leviathan transforms wasted power into productive compute. This enables “sovereign AI factories,” that can scale where power is abundant but traditional infrastructure cannot be built.
A notable deployment with the U.S. Navy placed ruggedized versions of its Galleon modular data centers aboard a ship during UNITAS 2025, the world’s longest-running multinational maritime exercise. Adapted for shipboard power, Galleon provided secure, real-time processing of drone and sensor data and ran compute-intensive workloads without relying on land-based hyperscale facilities.
Armada infrastructure is now deployed in more than 100 countries across four continents and covering all 50 U.S. states. In 2025, the company secured multi-year commitments from customers in defense, energy, mining, telecommunications, and public infrastructure, including deployments with Saudi Aramco, Tampnet, the Alaska Department of Transportation, and multiple state and municipal agencies.
8. Lambda
For building AI factories at gigawatt-scale
Founded in 2012 as a photo editing app, Lambda now bills itself as the only AI infrastructure player solely focused on building gigawatt-scale AI factories for training and inference, providing top AI labs, enterprises, and hyperscalers the GPU capacity they crave. Lambda co-engineers with top AI labs and collaborates with data center operators to deploy high-density, liquid-cooled, “energy-optimized” infrastructure. Its recent deployments with Cologix, Aligned Data Centers, and EdgeConneX have brought “superintelligence-grade” compute to strategic hubs in Texas, Ohio, Atlanta, and Chicago. Lambda deployed the first hydrogen-powered NVIDIA GB300 NVL72 systems at the Mountain View campus of data center specialist ECL, demonstrating hydrogen’s potential for gigawatt-scale AI with zero emissions.
In November, Lambda announced a multibillion-dollar partnership with Microsoft to build AI infrastructure powered by Nvidia processors. And investors chipped in nearly $2 billion in total funding in the year—including a February $480 million Series D co-led by Andra Capital and SGW, and a $1.5 billion Series E in November led by TWG Global.
9. Quantinuum
For making the most accurate quantum computations yet
This April, Quantinuum’s Quantum Origin became the first software platform of its kind to receive National Institute of Standards and Technology (NIST) validation. The agency recognized its Quantum Random Number Generator as a critical quantum security tool for federal agencies and their partners required to migrate to post-quantum cryptography (PQC). In March, the company claimed the first commercial application of a quantum computer with partners JPMorgan Chase, Oak Ridge National Laboratory, Argonne National Laboratory, and the University of Texas, solving a known industry challenge of generating the “random seeds” essential for the cryptography behind secure communication.
This November, the company unveiled Helios, its third-generation quantum computer. It features the highest-fidelity physical and logical qubits of any commercial system, and Quantinuum claims that it is the world’s most accurate for general-purpose commercial applications. In September, the company raised $600 million from investors including Nvidia’s venture, at a valuation of about $10 billion.
10. xLight
For rewriting the rules of advanced semiconductor manufacturing
Extreme ultraviolet lithography (EUV)—the use of very short-wavelength light to etch tiny circuit patterns onto silicon wafers—is essential in manufacturing cutting-edge (7-nanometer and below) semiconductors. Dutch company ASML has a de facto monopoly on the enabling equipment, selling its latest EUV machines for over $400 million. ASML’s machines integrate a scanner unit and a light source to create intricate patterns on semiconductor substrates. The current light source comes from laser-pulsed tin (Sn) plasma, which hits hard physical limits as chip makers aim for ever finer chip features.
Palo Alto-based xLight is developing a new light source, called a “free-electron” laser that could push past these light-source limitations and expand the capabilities of EUV machines like ASML’s. Drawing upon decades of research at U.S. national laboratories, xLight uses particle accelerators to generate laser-like X-ray beams that offer narrower spectral width (for finer etching), higher brightness (for faster, more reliable patterning of nanoscale chip features), and femtosecond pulses for more precise control of the photolithographic process. The company aims to build small warehouses adjacent to chipmaking fabs, which will provide “light as service”—delivering laser power on demand across up to 20 ASML scanners, potentially reducing capital and operating expenditures by more than 3X.
In July, the company raised a $40 million Series B funding round led by early-stage VC Playground Global. In December, it became the first company funded through the CHIPS and Science Act in Trump’s second term, receiving $150 million in incentives to help scale its technology, with the government taking an equity stake in exchange.
11. PsiQuantum
For breaking ground on the biggest quantum computing projects in the Western world
While it hasn’t announced the sort of computing speed records that other quantum computer makers have reported, PsiQuantum—one of the few companies building a photonics-based system—has started building out a physical footprint that dwarfs its known competitors (China is a wildcard).
The company was tapped to build the world’s first utility-scale quantum computer: a million-qubit, 540,000-square foot facility in Brisbane backed by $620 million from the Commonwealth and Queensland governments. In September, the company broke ground on a similarly ambitious project in Chicago, where it will be the anchor tenant of the Illinois Quantum and Microelectronics Park, a public-private initiative transforming a long-abandoned U.S. Steel plant on the South Side. The state of Illinois has committed $500 million to the project buildout, with the state, county, and city providing an additional $500 million in incentives.
PsiQuantum, with Microsoft, is one of only two companies so far to advance to the final stage of DARPA’s Quantum Benchmarking program, which aims to identify and support programs capable of building a commercially useful quantum computer by 2033. In a February article in Nature, company researchers documented the performance specs of its novel silicon-photonics platform, co-developed with GlobalFoundries.
In September, PsiQuantum raised $1 billion in a Series E funding round led by BlackRock, Temasek, and Baillie Gifford, with participation from NVIDIA’s NVentures, among many others. It marked the largest-ever funding round for a quantum computing company.
12. Cognichip
For bringing vibe coding to chip design
Launched out of stealth in 2025 with $33 million in seed funding, Cognichip is bringing vibe coding to chip design through what it calls Artificial Chip Intelligence (ACI). Based on a purpose-built, physics-informed foundation model trained on the “abstractions” already used for chip design, ACI transforms traditional serial, “waterfall” workflows into concurrent, adaptive, and highly automated design processes. ACI manages the entire design process as an integrated, intelligent system, identifying trade-offs, refining architectures, and adapting to changes in real time. The system can evaluate architectural options, suggest improvements, and simulate outcomes across various constraints. By removing the bottlenecks that traditional tools leave behind, it lets engineers move from idea to production far faster, with greater efficiency and flexibility.
Cognichip suggests that ACI can slash development costs by 75%, cut completion times in half, and significantly reduce chip size and power. In 2025, Cognichip grew to 55 employees, including engineers and researchers from Apple, Google, Synopsys, Cisco, and KLA, among others.
13. Sandisk
For reinventing memory for the AI era
As AI models grow larger and more complex, inference workloads demand both massive bandwidth and significantly greater memory capacity, requirements that traditional High Bandwidth Memory (HBM) cannot economically or physically deliver at scale. To address this storage bottleneck in AI clusters, Sandisk has developed High Bandwidth Flash (HBF). Unveiled in February 2025, HBF is built specifically for AI inference workloads, delivering 8-16x the capacity of HBM at a similar cost.
In August, Sandisk signed an agreement with SK Hynix, one of the largest suppliers of DRAM and flash memory chips, to establish and standardize the HBF specification and create a technology ecosystem. In late February 2025, nine years after its acquisition by Western Digital, Sandisk Corporation relisted on Nasdaq with its old ticker symbol (SNDK). The stock is up nearly 500% since then.
14. Positron AI
For building an inference-first AI hardware stack in the U.S.
GPUs were not designed to power LLMs and AI inference; they were developed for processing digital images and accelerating computer graphics, and they have become a bottleneck for ballooning inference workloads. Rather than retrofitting graphics hardware for AI workloads, Reno-based Positron AI builds chips specifically to accelerate and serve generative AI applications. Founded in 2023, Positron conceived and launched its debut product, Atlas, in just 15 months. The U.S.-made, air-cooled inference server offers more than 3x better performance per watt and performance-per-dollar compared to NVIDIA’s DGX H200, according to the company. Made up of eight proprietary accelerator chips, the system supports up to 0.5 trillion-parameter models in a single 2kW server and is compatible with Hugging Face transformer models via OpenAI APIs. And by optimizing memory bandwidth, Atlas achieves 93% utilization—far higher than the 10%-30% typical in GPU systems. Customers include web security giant Cloudflare and Parasail, via its AI-native data infrastructure platform SnapServe.
Titan, Positron’s next-generation platform, is due this year. It will offer up to two terabytes of high-speed memory per accelerator and support models of up to 16 trillion parameters, with up to 10x better performance per watt and dollar than Nvidia GPUs. In 2025, the company raised a $23.5M seed round and a $51.6M Series A led by Valor Equity Partners, Atreides Management, and DFJ Growth.
15. Morse Micro
For improving the range of Wi-Fi for IoT
Sydney-based Morse Micro has a fairly simple value proposition: offering IoT-optimized silicon for long-range Wi-Fi connections with very low power consumption. By operating in the sub-GHz band, Wi-Fi HaLow, aka 801.11ah, can extend up to a kilometer or more in open environments. In 2024, Morse Micro set a world record for Wi-Fi coverage by conducting a video call over nearly 10 miles. HaLow is so power efficient that it can maintain connectivity on battery power for years—perfect for smart farms, factories, or city sensors. The tradeoff is speed. While Wi-Fi 7 can transmit data at up to 46 gigabits per second, Wi-Fi HaLow maxes out at 15 megabits per second. But that’s fast enough to easily accommodate the kinds of text and status messages that it is designed for.
The Wi-Fi variant isn’t the only wireless network type vying for IoT applications, though. LoRaWAN, for example, has had a head start, and there are several cellular options. In September, though, market forecaster Omdia predicted that the Wi-Fi HaLow market will grow at a compound annual growth rate (CAGR) of 79% over the next five years, expanding from smart home cameras to smart-city applications and drones. Morse Micro is looking to kickstart the ecosystem with a new qualification program that establishes quality benchmarks for HaLow modules. It also offers two flavors of HaLowLink, gateways that allow easy connection of HaLow-equipped devices.
Explore the full 2026 list of Fast Company’s Most Innovative Companies, 720 honorees that are reshaping industries and culture. We’ve selected the companies making the biggest impact across 59 categories, including advertising, applied AI, biotech, retail, sustainability, and more.
