AI innovation may feel like it’s utterly dominated by a few giants, but the reality couldn’t be more different. These honorees have thrived by zeroing in on specific challenges, including teaching the technology to interpret sensor data, speeding up inference, and giving researchers better insight into how models actually work.
Archetype AI
For turning sensor data into intelligence
Large language models may know a lot, but because they have been trained on text and images, that limits their understanding of our world. By contrast, Archetype AI’s Newton foundation model can interpret data from cameras, lidar, radar, inertial measurement units, and other sensors used in enterprise domains such as construction, logistics, manufacturing, and urban planning. Customers ranging from the city of Bellevue, Washington, to data services firm NTT Data to construction giant Kajima Corporation have adopted the model, which outperforms the major general-purpose LLMs on sensor-related tasks.
Cerebras Systems
For putting inference’s pedal to the metal
If you’ve ever asked AI to do something and then drummed your fingers waiting for it to respond, you’ve run up against slow inference. To pick up the pace, Cerebras developed its own inference platform from the semiconductor wafer up. In January 2026, the company unveiled a partnership with OpenAI to launch a 750-megawatt deployment of its technology. Two months later, it announced that its systems had been deployed in Amazon’s AWS data centers. It’s also working with AWS on an AI architecture that pairs the companies’ technologies to quintuple a hardware footprint’s token capacity.
d-Matrix
For powering real-time AI
In June 2026, d-Matrix’s Corsair AI platform entered full production. It reduces typical latency inherent to AI by placing compute and memory alongside each other, an approach that an independent benchmark showed slashing AI response time to 2 seconds from 24 seconds. The company’s investors including Microsoft, SK Hynix, and Samsung, along with the Qatar Investment Authority and Singapore’s EDBI.
Goodfire
For taking the mystery out of models
Silico, Goodfire’s AI development and interpretability platform, is designed to overcome AI models’ notoriously opaque nature. It allows researchers to examine the innards of a model, make precision adjustments, and steer its behavior in ways that might otherwise be impossible. Mayo Clinic adopted Silico to catalog 4.2 million genetic variants and predict their likelihood of causing disease. Rakuten used it to excise sensitive data from its agent platform, resulting in 58% fewer hallucinations at approximately 90 times less cost than typical approaches.
Liquid AI
For helping AI think small
AI running at massive scale in data centers gets most of the industry’s attention. Liquid AI’s LFM2.5 is a set of AI models optimized to run on processors in laptops, smartphones, vehicles, and IoT devices—no internet connection necessary. Instead of being based on transformers like a conventional model, each one is customized by Liquid AI’s own meta-AI system to operate within specific processor and memory constraints. Chipmakers AMD, Intel, and Qualcomm are shipping LFM2.5 versions for their respective platforms. Last April, Liquid AI also signed a multiyear deal with Mercedes-Benz to embed the technology directly into vehicles.
Merge
For plugging enterprise AI into the real world
Released in October 2025, Merge’s Agent Handler lets organizations use AI to unlock the value of data that lives in their essential applications. Instead of engineering teams having to wire up Model Context Protocol (MCP) connections themselves, they can use Merge’s library, which covers of hundreds of enterprise tools, such as Salesforce, Workday, Gmail, and GitHub. These connectors are prebuilt and designed with security and observability in mind. Some 1,700 paying customers call on the platform for more than 1.1 billion daily API requests; it also powers Perplexity’s Enterprise Pro service.
Runpod
For delivering AI infrastructure as a service
Runpod is a cloud provider focused entirely on helping organizations train and deploy fast, reliable AI without having to invest in GPU hardware or manage it on their own. Its platform provides access to more than 30 Nvidia and AMD GPU models in 31 global regions and offers proprietary technologies such as FlashBoot, which speeds inference by cutting cold-start times to under 200 milliseconds. The company says it’s passed $120 million in annual recurring revenue from more than 1 million customers.
The companies and individuals behind these technologies are among the honorees in Fast Company’s Next Big Things in Tech awards for 2026. Read more about the winners across all categories and the methodology behind the selection process.