Why Big Tech is giving away open-weight Al models and what companies want in return?
- Beam, Muse Spark Open, Nemotron 3.5 Super VL, Nemotron 3.5 Bolt 7A/8B, Timestral. These may sound like characters from science-fiction franchise. They are, in fact, names of open-weight Al models announced or released by technology firms in U.S. and Europe in October.
Reflection Al released Beam's weights October 16. Meta released Muse Spark Open with NVIDIA made two Nemotron models available October 15. France's Mistral also announced Timestral, family of models designed for time-series forecasting.
Firms are making models available for various tasks viz. from writing software and operating Al agents to interpreting images and forecasting time-series data. But why spend heavily to develop Al models and then allow others to download and adapt them? If models free to download, where does money come from?
Answer lies in race to attract developers, shape how Al is used and earn money from businesses, services and computing infrastructure built around their models.
Read more - China's Gaming Ambition: Soft Power vs. State Control
Open-weight model - Al models learn patterns from data during training. Their trained parameters, called weights, are numerical values that help fix how models process inputs and generate responses. Think of Al model as chef who learned from thousands of recipes. During training, model learns patterns that help it decide what ingredients to combine and in what proportions to make certain dish. Its weights shape these decisions. Making weights available is like sharing chef's acquired knowledge with other chefs, who can then adapt it to their own needs and create dishes of their own. Most people access proprietary Al models via websites or apps while developers may access them via APIs. Provider controls access and may charge for usage. Open-weight model makes trained parameters available for download. Developers can run it on compatible computers or servers, adapt it for particular tasks or integrate into applications. This can give businesses and individual users greater control over how they deploy and customise Al models, and reduce dependence on single provider. However, open-weight does not automatically mean open-source in fullest sense. A firm may release weights without publishing training code or sufficient information about data used to train model. Licences may also curb some uses. Open-weight Al not new. OpenAl released GPT2's weights in 2019. What has changed is capability available via downloadable models and emergence of Chinese developers such as DeepSeek as serious competitors, prompting broader contest over who will shape Al ecosystem.
Chinese competition - Growing availability of Chinese models has intensified competition in open-weight Al market. Firms including DeepSeek and Alibaba released models whose weights developers can download and adapt. DeepSeek-R1, released in January 2025 under MIT licence, made reasoning model available for others to use, modify and commercialise. Alibaba's Qwen family includes models of different sizes for range of applications. Developers choose models based on performance, cost, hardware requirements and licensing. If Chinese model meets their needs at lower overall cost, they may choose it over proprietary American or European service. As more developers build applications and tools around model, its ecosystem can make it more attractive to others.
OpenAl made geopolitical argument explicit when it released its open-weight models in August 2025. It said models could help partner countries build Al infrastructure on what it called U.S.-led rails, giving them incentive to choose democratic Al over autocratic Al. This is company's stated position but illustrates how spread of Al technology also viewed as strategic contest.
Read more - Internet in a Sachet
Where is money? Answer depends on what firm sells. Meta CEO Mark Zuckerberg outlined approach in July 2024 essay, Open Source Al Is Path Forward. Drawing comparison with Linux, he argued releasing Llama's weights may encourage firms to develop tools, integration and hardware optimisation around models, aiding broader ecosystem. Mr. Zuckerberg said selling access to Al models was not Meta's business model, which centred on building products and services. He argued releasing Llama would help Meta avoid dependence on competitors' proprietary technology while encouraging development of tools and services around its models. For NVIDIA, commercial interest different. It sells computing hardware and related software. Wider adoption of Al applications could raise demand for computing power, potentially benefiting NVIDIA, although developers can run open models on competing hardware. Revenue gains from individual model releases not guaranteed. Model developers such as Mistral can offer paid hosting, enterprise support, customised models and private deployments. Alibaba's Qwen models sit alongside its cloud and Al services. Developer may download model's weights at no cost but pay provider to host and run it reliably at scale. Alternatively, developer may run it independently and never pay its creator. Business bet is that wider adoption will generate enough revenue or strategic value elsewhere to justify cost of developing and releasing model.
Shaping Al ecosystem - Longer-term prize is influence. Widely-used model can attract developers who build applications and tools around it. These additions can make model easier to deploy and use, potentially attracting more developers. This reinforcing cycle known as network effect.
Distribution matters too. Making model weights downloadable is only one way to encourage adoption. Offering access via APIs, model hubs and inference platforms can make models easier for developers to discover, test and integrate into applications. Mr. Zuckerberg argued Meta wants Llama to turn into industry standard. Mr. Zuckerberg argued open access to Al models was not Meta's business model... Mr. Zuckerberg argued Meta wants Llama to turn into industry standard.
Read more - Al's Trillion Gamble
For India, for Indian businesses, open-weight models offer alternative to relying on foreign Al services. Organisations can adapt models to Indian languages and local requirements and may choose to run them on own infrastructure, depending on computing resources and model's licence. India has own efforts too. Sarvam Al has released models including Sarvam 30B and Sarvam 105B while AI4Bharat at IIT Madras has made Indian language models and tools available to developers. BharatGen has also released models for Indian languages and domain-specific applications, including agriculture, law and Ayurveda. Objectives, however, vary. Commercial firms may seek venue via applications and enterprise services while research initiatives may prioritise language access and domestic technical capability. Open weights alone do not guarantee lower costs. Hardware, maintenance, performance and licensing still matter and organisations must assess if running model themselves is more economical than using hosted service.
Analytically, open-weight is new oil war - US-China fight for developer mindshare, India caught in middle. For Indian startups, open models reduce dependence but increase risk of misuse - safeguards can be removed. Need India-specific licensing framework.
Tags: #OpenWeightAl #MuseSpark #Mistral #Nemotron #DeepSeek #Meta #NVIDIA #AI4Bharat #SarvamAl #BharatGen