Yazılar

AI Chip Startup Positron Raises $23.5 Million to Challenge Nvidia

/in /tarafından

Positron, a startup aiming to rival Nvidia in the artificial intelligence (AI) chip market, announced on Tuesday that it has raised $23.5 million in a seed funding round. Investors in the round included Valor Equity Partners, known for its support of Elon Musk’s ventures, along with Atreides Management, Flume Ventures, and Resilience Reserve.

Focus on Efficiency and Inference:

Positron’s chips are manufactured in Arizona and are designed to use less than a third of the power of Nvidia’s leading H100 graphical processing units (GPUs) while offering similar performance. The company’s chips are specifically intended for AI inference, the phase where AI models are utilized, as opposed to training the models. Although demand currently leans toward training chips, analysts forecast that the need for inference chips will rise as more AI applications are developed.

Industry Shift and Rising Costs:

With major players like OpenAI, Google, and Meta investing heavily in AI infrastructure, the demand for chips is expected to grow significantly. Meta, for example, has pledged to spend up to $65 billion this year, while Microsoft plans to invest $80 billion. OpenAI also announced a $500 billion Stargate infrastructure project. Despite Nvidia’s dominance, holding around 80% of the market, rising costs and concerns about over-reliance on a single supplier have pushed companies such as Microsoft, Meta, and OpenAI to seek alternative solutions, both in-house and externally.

AI Chip Firm Cerebras Partners with Mistral, Claims AI Speed Record

/in /tarafından

Cerebras Systems, a Silicon Valley-based AI chip manufacturer backed by UAE tech firm G42, announced on Thursday its partnership with French AI company Mistral, helping it achieve a new speed record for AI responses.

Mistral, known for its open-source AI technology, aims to rival competitors such as Meta Platforms and China’s DeepSeek, both of which are challenging OpenAI’s dominance in the AI space. On Thursday, Mistral launched a new app, Le Chat, capable of generating responses at a speed of 1,000 words per second.

Cerebras provided the computational power to achieve this speed, which it claimed surpasses the performance of both OpenAI and DeepSeek, making Mistral’s Le Chat the fastest AI assistant globally.

Cerebras, which has filed for an initial public offering currently under U.S. regulatory review due to G42’s involvement, is one of the few major challengers to Nvidia in the AI chip market for training large models. However, its collaboration with Mistral focuses on inference, the process of serving model-based applications to users in real-time.

Andrew Feldman, Cerebras CEO, emphasized that speed in delivering answers has become a key focus as competitors close in on OpenAI’s models. “You want better answers. And to get better answers, you need more compute at inference time,” Feldman told Reuters. He also expressed pride in the collaboration with Mistral, calling it Cerebras’ first major win with a top-tier AI model developer.