Blaize (NASDAQ:BZAI) Chief Financial Officer Harminder Sehmi outlined the company’s edge artificial intelligence strategy, customer use cases, partnerships and financial outlook during a recent company event, emphasizing that the chipmaker is focused on AI workloads that require low power, low latency and efficient processing outside traditional data centers.
Sehmi said Blaize was founded 12 to 14 years ago by Dinakar and other co-founders with the goal of designing “a new type of GPU” that is programmable and optimized for edge use cases. He said programmability is central to the company’s strategy, noting that the major surviving GPU companies shared that characteristic.
Edge AI Focus Includes Smart Cities, Defense and Industrial Uses
Sehmi said Blaize’s chips are designed for use cases where inference happens close to where data is created, such as on drones or boxes mounted on rooftops or lampposts. He said a single chip can handle “four to five high-definition streams,” and that the chip can run multiple algorithms at once, including video and sensor-related workloads.
One customer Sehmi discussed was TCC, which he described as a subsidiary of Saudi Arabia’s Public Investment Fund. He said TCC is using Blaize technology in a highway monitoring application involving license plate recognition and fines.
According to Sehmi, Blaize tested its equipment with TCC last September on a rooftop, where the temperature inside the box reached 75 degrees Celsius while the AI workloads continued running. He said a GPU-based system stopped working at “50 something” degrees Celsius.
Sehmi also pointed to Winmate, a Taiwanese company that develops ruggedized boxes for industrial use and drones. In drone applications, he said Blaize cards can support defensive capabilities by allowing drones to identify incoming threats and run navigation or identification algorithms onboard.
Hybrid GPU-GSP Systems Target Data Center Efficiency
Sehmi said Blaize is “not here to replace NVIDIA” but to complement GPUs by allowing different silicon to be used for different workloads. He said the company’s hybrid approach is aimed at Tier 2 data centers and can support a range of AI workloads beyond video, including text and document processing.
He said smaller AI models, including those in the seven-billion to eight-billion parameter range, can run well on Blaize chips, while larger models can run on GPUs in hybrid systems.
As an example, Sehmi cited Yotta, a data center company in India, and an end customer related to Indian highways. He said Blaize demonstrated that a Supermicro server powered by 24 Blaize cards could run the relevant workloads at two to four times lower total cost of ownership compared with an NVIDIA-based GPU setup.
Sehmi said Blaize’s efficiency advantage comes partly from avoiding the memory-related penalties GPUs can face on small-batch workloads. He said Blaize systems do not require high-bandwidth memory and instead use LPDDR memory.
Partnerships and Go-to-Market Strategy
Sehmi said Blaize uses a mix of direct sales and partnerships. He called the company’s partnership with Nokia a “game changer” because Nokia already has relationships with data centers and provides connectivity. He said Blaize and Nokia can bring AI solutions to those data centers together.
The company also works with independent software vendors, Sehmi said. Blaize has developed some applications internally, but he said it is too expensive to build applications for every use case, so the company works with ecosystem partners that already have relevant software.
Sehmi said Blaize’s chip is taped out by Samsung Foundry in Austin, Texas, a detail he said is important for defense-related applications because it supports “Made in America” requirements. Asked about memory sourcing, Sehmi said he did not yet know the answer and that part of his job is to speak with Samsung and Micron about securing supply.
Revenue Outlook and Margin Mix
On financials, Sehmi said Blaize closed last year at about $39 million in revenue after exceeding a tightened guidance range. He said the company has tightened its 2026 revenue guidance to $130 million, adding that the year is expected to be back-end loaded because contracts require the company to secure chips, cards and servers as proofs of concept move into production.
Sehmi said Blaize has not provided projections for 2027. However, he said the company is considering introducing metrics such as backlog and bookings as its revenue mix begins to include more software and recurring revenue.
Asked about revenue mix, Sehmi said 2025 and 2026 revenue will be mostly hardware, with 2026 including more system-level hardware. He said some deployments, including Yotta, already include a software component, which helped margins. As AI services deployments expand, Sehmi said he expects margins to improve significantly in 2027 and 2028 as software becomes a larger portion of the mix.
Sehmi said hardware will remain important because Blaize expects to continue providing hybrid servers combining GPUs and GSPs. He also cited the company’s relationship with NeoTensr and a white-labeled server manufacturer as factors supporting margins.
Next-Generation Chip Plans
Sehmi said Blaize is on its third chip, after two test chips and one production chip, and that all “lit up within the first hour.” He said the next-generation chip is intended to help run model sizes expected over the next two to three years and strengthen the company’s hybrid strategy by increasing the Blaize component in systems relative to GPUs.
Sehmi closed by saying the company’s focus is on executing existing contracts and ensuring new opportunities do not divert Blaize into unnecessary development efforts. He said the company will rely on its ecosystem for applications outside its core focus areas.
About Blaize (NASDAQ:BZAI)
Blaize (NASDAQ: BZAI) is a fabless semiconductor company that designs and develops hardware and software solutions for artificial intelligence (AI) and machine learning applications at the edge. The company’s core technology is centered on its proprietary Graph Streaming Processor (GSP) architecture, which combines dataflow computing with a highly parallel matrix processing engine to deliver real-time AI inference with low power consumption. Blaize’s platform is aimed at customers seeking to deploy sophisticated AI workloads in environments where power efficiency, latency and form factor are critical.
The company offers a hardware portfolio that includes standalone GSP modules, PCIe cards and M.2 form-factor boards, alongside its Blaize AI software stack.
