Anthropic CEO Dario Amodei is facing challenges in a copyright lawsuit against OpenAI, as reported in new court filings. The Read more
Artificial Intelligence Terminology: A Comprehensive Guide Artificial General Intelligence (AGI) Artificial General Intelligence (AGI) surpasses conventional AI capabilities by excelling Read more
On Sunday, the European Union granted regulators the authority to prohibit the use of AI systems considered to pose an Read more
LexisNexis Risk Solutions, a data broker that collects and uses consumers’ personal data to help its paying corporate customers detect Read more
Andreessen Horowitz general partner and Mistral board member Anjney “Anj” Midha first noticed DeepSeek’s impressive performance six months ago, he tells TechCrunch. DeepSeek introduced Coder V2, which competed with OpenAI’s GPT4-Turbo for coding-specific tasks, according to a paper released last year, putting DeepSeek on a path to release improved models regularly. R1, its new open-source reasoning model, has disrupted the tech industry by offering industry-standard performance at a lower cost.
Despite the sell-off of Nvidia’s stock, Midha explains that R1 doesn’t mean AI foundational models will stop investing billions in GPU chips and data centers. Instead, they will optimize the compute power they have available. Midha emphasizes the value of Mistral’s efficiency improvements and the need to invest in them, even with the billion dollars raised.
Tech and VC heavyweights join the Disrupt 2025 agenda
Midha contends that Mistral remains competitive with rivals like OpenAI and Anthropic due to its open-source nature. Open source models provide access to free technical labor from users, while closed-source competitors guard their secrets and incur costs for labor and compute power. Midha highlights Mistral’s position as the open-source provider with the most compute power.
Facebook’s Llama, Mistral’s largest Western open-source AI model rival, will also receive significant investment. CEO Mark Zuckerberg plans to spend “hundreds of billions of dollars” on AI, including capital expenditures for data centers. Midha, also a board member for other AI companies, explains that the demand for GPUs remains high due to the need for AI model training and product operation.
a16z’s Oxygen GPU sharing program “overbooked”
Midha leads a16z’s Oxygen program, which purchased Nvidia H100 GPUs for its portfolio companies facing a shortage in the market. The demand for GPUs for AI model training and inference remains high, highlighting the continued need for compute power. DeepSeek’s engineering breakthroughs are seen as not impacting Stargate, OpenAI’s partnership for AI data centers.
The recognition that AI is the next foundational infrastructure leads to discussions on “infrastructure independence.” Midha advocates for Western nations to rely on Western AI models that align with Western laws and ethics. However, concerns about Chinese open-source models prompt companies to consider running them locally or using secure cloud services from American providers.
Despite the debate, some companies, like Pat Gelsinger’s startup Gloo, are choosing to build AI services on their version of DeepSeek R1. Midha humorously requests extra GPUs for further development.
