Nvidia: Going All-In on Open-Weights
Nvidia has amassed a vast fortune during the AI surge, but the increasing backlash against open-weights models—models that anyone can modify and utilize on their own systems—might alter the landscape. Nvidia’s main man, Jensen Huang, isn’t foolish enough to rest on past successes. On Friday, a coalition of tech elites penned an open letter as a rebuttal to the anti-open-weights rhetoric from American companies such as Anthropic and OpenAI. Among those who signed were Meta, Microsoft, IBM, and Dell, but Nvidia’s voice was the most prominent. Who would have guessed, right? Nvidia stands to lose the most in this scenario.
Huang: We Must Embrace ‘Open’ Models
Jensen Huang, venturing into the social media landscape on the platform previously known as Twitter, highlighted the letter, advocating for open models: “enhance safety and cybersecurity, promote innovation and distribution, and support sovereignty.” He playfully remarked, “The world requires both advanced closed models and advanced open models.” Nvidia’s trajectory depends on a blend of models, as engaging in a bit of arms trade—selling to both ends—usually secures revenue. If the US chooses to disregard Chinese open-weights models, American clients will have to settle for mediocre models or overpriced alternatives from OpenAI, Anthropic, and Google. This narrows Nvidia’s market and hinders its expansion. So what’s the point of all these “AI factories” Nvidia’s been boasting about if that occurs?
America vs. China: The Open-Weights Confrontation
The US hasn’t endorsed open weights models as strongly as its Chinese counterparts. For firms reluctant or unable to purchase proprietary models, companies like DeepSeek, MiniMax, Z.AI, and Moonshot (Kimi) are their only real alternatives. Thinking Machine Labs’ billion-parameter Inkling model stands as the best open weights option for the US, and by benchmark standards, it pales in comparison to Kimi K3. Huang needs to ignite a spark among US model developers to create open weights models that can compete head-to-head with those from China; otherwise, they will be left struggling.
Flex and Burn: OpenAI and Anthropic
OpenAI and Anthropic are attempting a precarious balancing act at a staggering burn rate, consuming tens of billions annually, hoping their efforts will eventually yield results. Nvidia has invested billions into AI startups like OpenAI and Anthropic, often without conditions, typically linked to infrastructure rollouts. OpenAI and Anthropic find themselves burdened with an immense financial stake tied to Nvidia that may or may not yield positive outcomes. There are whispers that Nvidia may consider a $250 billion lifeline to assist OpenAI in leasing a datacenter from Softbank in Ohio. It’s as if Huang is sending a sly hint to Mr. Altman at OpenAI: “It’d be quite unfortunate if something happened to that funding, wouldn’t it? Best get back on that ‘open’ bandwagon that inspired your company’s name.” OpenAI, SpaceXAI, and Meta seem to have taken the hint loud and clear.
OpenAI and SpaceXAI: Upholding Commitments
OpenAI’s leader, Sam Altman, mirrored Huang’s X post, endorsing open weights models: “I want the US to excel in AI in both open source and proprietary frameworks, and I’m pleased to see this.” OpenAI has a history with open-weights models—recall GPT-2 from 2020? They even launched their first new open model since then, GPT-OSS. However, GPT-OSS is now a relic, preparing to celebrate its first birthday. Even Meta is gearing up to join the open model race, with their AI Officer Alexandr Wang pledging a comeback. Then there’s Elon Musk with his SpaceXAI, supporting Huang’s post, but like many of Elon’s grand promises, execution has never been his forte.
Words or Deeds?
Will these substantial promises translate into reality? We can’t predict just yet. Only time will reveal if new models emerge from SpaceXAI, Meta, or OpenAI, or if they invest sufficiently to compete with Chinese counterparts. If OpenAI has a GPT-OSS-2 in the pipeline, next month could be the perfect time for its release. If not, silence may follow. Anthropic has shown the greatest resistance to China’s open weights models. In a Monday blog post, CEO Dario Amodei expressed concerns about authoritarian regimes exploiting AI for military dominance or citizen oppression, and even the threat of cyberattacks. Nevertheless, Anthropic remains steadfast, avoiding any ventures into open weights models.
Security and Open-Weights: Nvidia’s Initiative
Nvidia is not slowing down, establishing the Open Secure AI Alliance to equip defendents with resources, including open frontier models, against emerging cybersecurity challenges. Regardless of whether companies adopt Nvidia hardware, open ecosystems and choices are always beneficial.
Updated: Is Anthropic the Wisest?
Anthropic’s Monday blog post stirred the conversation, arriving after this article was published, but it’s all part of the ongoing discourse.
Summary
Show Me the Models!
This open-weights narrative has more twists than a Geordie explaining a Newcastle pint. Nvidia’s banking on open models to keep their finances healthy, and the AI rivalry with China appears to be just beginning. While American tech giants make lofty promises, only time will disclose if they’re genuinely willing to take action or remain mere words, ready to challenge China’s progress. Either way, stay tuned at gadgetlad.co.uk for the latest cheeky updates.
