Open-Weight AI Companies Are Silicon Valley’s Hottest Targets

Silicon Valley’s AI acquisition race is moving beyond flashy consumer apps and proprietary foundation models. Some of the industry’s biggest companies are now spending billions to secure the infrastructure, talent and developer ecosystems behind open-weight artificial intelligence.

The latest potential blockbuster is Nvidia’s reported pursuit of Hugging Face, the widely used platform for hosting and distributing AI models, datasets and developer tools. Nvidia has reportedly been discussing a deal worth around $13 billion, although the acquisition had not been officially confirmed as of August 29.

It comes just days after Nvidia struck a roughly $6 billion agreement with open-weight AI startup Poolside and weeks after Stripe agreed to acquire OpenRouter, one of the largest marketplaces and routing platforms for AI models.

Taken together, the transactions show how quickly value is shifting toward companies that make it easier for businesses to access, customize, host and switch between AI models.

Nvidia’s Reported $13 Billion Hugging Face Deal Could Be the Biggest Signal Yet

Hugging Face has become one of the most important pieces of infrastructure in the open AI ecosystem. Developers use the platform to publish, test, fine-tune and distribute machine learning models, giving it a role sometimes compared with GitHub’s position in software development.

Nvidia is reportedly considering paying around $13 billion for the company. Nvidia was already an investor in Hugging Face, participating in its $235 million Series D funding round in 2023, which valued the startup at $4.5 billion.

That means the reported acquisition price would represent almost three times Hugging Face’s 2023 valuation.

The underlying platform has also grown dramatically. Hugging Face reported that its community reached about 13 million users, more than 2 million public models and over 500,000 public datasets in 2025. More recent ecosystem data indicates the Hub crossed 3 million public models in August 2026.

For Nvidia, acquiring that developer network could provide something even more valuable than models themselves: distribution.

Nvidia already offers its own Nemotron open-weight models, but controlling a platform used by millions of AI developers could give the chipmaker a much larger channel for promoting its software stack, inference infrastructure and GPUs.

Nvidia Is Also Spending Billions on Poolside

Hugging Face is not Nvidia’s only open-weight AI move.

The chipmaker recently struck an approximately $6 billion agreement with Poolside, an AI startup focused heavily on coding and open-weight models. The transaction involves technology licensing and bringing more than 100 Poolside employees into Nvidia’s broader AI efforts.

Nvidia is also expected to make a separate $1 billion equity investment in Poolside, giving the startup an estimated valuation of around $12 billion.

The technology and engineering talent are expected to contribute to Nvidia’s Nemotron initiative as the company attempts to build increasingly capable open models.

The strategy puts Nvidia in an unusual position.

It remains the dominant supplier of AI accelerators to many of the world’s largest proprietary AI labs, but it is simultaneously building an alternative ecosystem in which businesses can deploy open models using Nvidia infrastructure.

That diversification could become increasingly important as companies such as Google and OpenAI develop more of their own AI silicon.

Stripe Paid More Than $8 Billion for OpenRouter

The acquisition wave is not limited to Nvidia.

Stripe announced on August 19 that it would acquire OpenRouter, a marketplace and gateway that allows developers to access hundreds of different AI models through a single interface.

The companies did not publicly disclose the purchase price, but Reuters reported that the deal was valued at slightly more than $8 billion.

The scale behind OpenRouter helps explain the price.

OpenRouter says it now processes more than 10 trillion tokens per day across more than 400 AI models and serves a community exceeding 10 million developers and companies. It also says inference volume has increased by at least 10 times annually since the platform was founded.

For Stripe, the acquisition connects AI consumption directly with the financial infrastructure used to charge for it.

As AI applications increasingly combine models from OpenAI, Anthropic, Google, DeepSeek, Qwen and smaller providers, routing systems can automatically choose models based on cost, speed, availability or performance.

Stripe CEO Patrick Collison described tokens as becoming a central economic unit for companies building AI products, reflecting the company’s bet that managing AI compute could become closely tied to billing and payments.

Open-Weight Adoption Is Small, but It Is Growing Fast

The billions being spent on open-weight AI might suggest the technology already dominates enterprise deployments. The actual numbers show a more complicated picture.

Ramp’s August AI Index found that 6.1% of businesses using AI were paying for model-serving platforms that provide access to open-source, open-weight and some Chinese models. That figure increased by 0.2 percentage points from the previous month.

Jellyfish sees an even smaller but rapidly accelerating trend among software engineering organizations.

Its August 2026 research, covering roughly 276,000 engineers, found that fewer than 2% of companies regularly use open-weight models or model routers. However, adoption approximately doubled over the previous six weeks.

Among engineers already using open-weight models, Jellyfish found that Kimi from Moonshot reached 51.8% usage and GLM from Zhipu reached 51.2%, followed by DeepSeek at 8.6%, Alibaba’s Qwen at 7.9% and MiniMax at 5%.

That suggests Chinese-developed open models are becoming an increasingly important part of the enterprise AI conversation even as American proprietary platforms continue to dominate overall spending.

Businesses Are Starting to Question the Cost of Frontier AI

One reason open-weight models are attracting so much attention is economics.

Running millions or billions of inference requests through premium proprietary APIs can become expensive, particularly for repetitive workloads such as customer support, classification, extraction and automated content processing.

Ramp’s latest spending data suggests businesses may already be reaching limits on how much they are prepared to pay for additional model performance.

In July, the top 1% of companies in Ramp’s dataset spent a median of approximately $7,400 per employee on AI, while the top 10% spent around $650 per employee. The median company spent just $11.95 per employee.

The same research found that 43.5% of U.S. businesses in its AI dataset were paying for Anthropic, compared with 39.7% for OpenAI.

But open-model infrastructure is steadily gaining ground as businesses look for cheaper alternatives.

Open-weight models can be downloaded, fine-tuned and sometimes self-hosted, allowing companies to optimize them for narrow workloads instead of repeatedly paying for the capabilities of the largest frontier models.

Fireworks AI Shows How Valuable the Infrastructure Layer Has Become

Another company attracting attention is Fireworks AI, which provides infrastructure for deploying and customizing open models.

Fireworks raised approximately $1.5 billion in July at a $17.5 billion valuation. The company also announced that it had crossed a $1 billion annualized revenue run rate.

Perhaps more striking is its inference volume.

Fireworks says its infrastructure serves more than 40 trillion tokens every day, with over 95% coming from models specialized using customers’ proprietary data and optimized for particular tasks.

That is four times the daily token volume OpenRouter reported when announcing its Stripe deal.

Ramp’s August data also identified Fireworks as the fastest-growing company in its model-serving and inference category, while OpenRouter remained the most widely adopted provider within that segment.

Fireworks has not announced an acquisition, but its scale and valuation demonstrate why inference providers and model-routing platforms have become strategically important assets.

Open Weight Does Not Necessarily Mean Open Source

There is also an important distinction behind the trend.

“Open-weight” models are not necessarily fully open-source models.

A company may release the numerical weights required to run and customize an AI model while keeping training datasets, training code or other parts of the development process private. Licenses can also impose restrictions on how models are commercially deployed.

Despite those differences, open-weight models give businesses substantially more control than APIs where the underlying model remains entirely controlled by the provider.

That control matters for companies concerned about data governance, latency, customization, vendor lock-in and long-term inference costs.

Why Big Tech Is Buying Before Adoption Peaks

The most interesting part of the acquisition boom may be its timing.

Enterprise adoption of open-weight AI remains relatively small, yet Nvidia, Stripe and investors are assigning multibillion-dollar values to the companies building its infrastructure.

They appear to be buying ahead of the curve.

Frontier models from OpenAI, Anthropic and Google still dominate many complex coding, reasoning and agentic workloads. But as open models improve, companies may no longer need the most expensive frontier model for every task.

An enterprise could use a premium model for difficult reasoning, a customized open-weight model for customer support and another smaller model for classification or data extraction, with a router automatically deciding which one handles each request.

That architecture makes the companies controlling model marketplaces, inference systems and developer ecosystems increasingly valuable.

The AI industry may therefore be moving from a market dominated by a handful of model makers toward one where thousands of specialized models coexist.

If that transition continues, Nvidia’s interest in Hugging Face, its multibillion-dollar Poolside agreement and Stripe’s purchase of OpenRouter may look less like isolated deals and more like the beginning of a race to control the infrastructure underneath the open AI economy.