# Open-Weight AI: Why Big Tech Pushes It — and What SMEs Gain

> Author: Chris Jon Graf (AI Strategist & CEO)
> Updated: 2026-07-28
> URL: https://ai-outsourcing.ch/insights/open-weight-ai-why-big-tech-pushes-it-and-what-smes-gain

## Summary

Microsoft and NVIDIA back open-weight models not out of altruism, but because open weights drive demand for their chips and cloud platforms. For Swiss SMEs, that same self-interest delivers cheaper, more transparent, more adaptable AI—without vendor lock-in.

Microsoft and NVIDIA together rank among the most valuable companies on earth, with a combined market capitalisation north of 8 trillion dollars. And these two giants are precisely the ones now championing open, freely usable AI models with full force. The reason is not philanthropy but hard-nosed economics: every open model trained on their hardware and run on their cloud increases demand for exactly those chips and platforms. For Swiss SMEs, this is good news—the very same self-interest delivers cheaper, more transparent, more adaptable AI tools, without forcing you to give up your independence.

## The mechanism: self-interest as the engine behind open models

The logic is simple once you see it. Microsoft and NVIDIA do not primarily earn from the model itself but from the infrastructure underneath it—data centres, GPUs, cloud contracts. An open model that any company can download and run on its own does not sell as a product. But it generates demand for the compute it runs on and for the tools used to fine-tune it. The trend shows up even outside the major platform vendors: Mira Murati's startup recently released Inkling, a 975-billion-parameter model, under an Apache 2.0 licence—free to use, free to modify, free to commercialise. Openness is no longer a niche pursued by idealists; it is a strategic weapon wielded by the largest and the most ambitious players at the same time.

## What 'open-weight' concretely means for your business

Open-weight is not the same as open-source in the classical sense, but for practical purposes one thing matters most: you gain access to the trained model parameters and can run, adapt and further develop them on your own or rented infrastructure with your own data—without depending on a single vendor.

- No obligation to send data to a single closed-source provider
- Full control over hosting location—relevant for data protection and sovereignty
- Fine-tuning on your own business data is possible
- Switching infrastructure providers without rebuilding the entire AI application
- Transparency into model behaviour, which simplifies audits and compliance

## The cost curve already backs up the thesis

**90%** — Cost reduction in accessing capable AI models since 2022

This dynamic is not a promise for the future; it is already measurable. Since 2022, the cost of accessing capable AI models has fallen by roughly 90 percent—an effect driven substantially by competition between open and closed models. For decision-makers, this means the right time to start with AI is not 'eventually', but now, while the curve still points downward.

## Vendor lock-in is the bigger risk, not the AI itself

Anyone relying entirely on a single closed platform today takes on a risk that is often underestimated: dependence on the pricing, roadmap and business decisions of one vendor. Recent standards conflicts among agent frameworks illustrate how quickly ecosystems can wall themselves off from each other once commercial interests collide.

How much AI resilience against vendor lock-in matters in a crisis becomes clear when looking at how Chinese companies responded to US sanctions: organisations that had already invested in open, portable architectures were able to react far faster than those fully committed to a closed system.

## Diversification as a principle, not an exception

Large technology companies themselves increasingly follow this principle of diversification. Vendor diversification in the cloud business, as seen at Meta, shows that even companies worth billions are increasingly running multiple infrastructure partners in parallel rather than relying on a single provider. The same principle applies to a Swiss SME on a smaller scale: open models combined with several possible hosting options mean you can actually switch when prices rise or performance disappoints—not just in theory.

## Sovereignty is becoming a regulatory obligation

The question of vendor dependence is no longer just strategic; it is increasingly regulatory. The planned EU Cloud and AI Development Act introduces tiered sovereignty requirements—depending on the sensitivity of data and application, companies will need to demonstrate where and how their AI systems are operated. Open models that can run on European or Swiss infrastructure are considerably better positioned to meet these requirements than closed systems that must run through a single US provider's cloud.

## The Swiss AI paradox—and the opportunity within it

**8%** — Share of Swiss SMEs using AI productively in their core business, despite high talent density

Switzerland has one of the highest densities of AI talent in the world—yet according to recent surveys, only around 8 percent of SMEs already use AI productively in their core business. This paradox is largely explained by uncertainty: concerns about lock-in, opaque costs, and dependence on a single vendor. The open-weight movement is directly shrinking these barriers. Companies that start today with an experienced partner no longer have to choose between control and speed.

> Incentives aligned. Progress accelerated. Capitalism still delivers—when the interests line up.
>
> — Commentary on Big Tech's open-model strategy, 2026

## What this means concretely for your AI strategy

1. Check for every AI project whether an open model alternative exists before committing to a closed solution.
2. Negotiate contracts so that switching providers remains technically and contractually feasible.
3. Choose infrastructure partners who actively support openness and portability rather than making it harder.
4. Build internal competence to evaluate models—not just consume them.
5. Take advantage of the current cost curve now instead of waiting for a supposedly 'better' moment.

> **Openness as a selection criterion**
>
> When you buy in external AI expertise, explicitly ask about the underlying model and infrastructure strategy. A good partner works with open, portable standards—not with hidden dependencies that become expensive over time.

## FAQ

### What is the difference between open-weight and open-source AI models?

Open-weight models make the trained parameters (weights) openly available so they can be downloaded, run and adapted. Full open-source status would additionally require open training data and methods, which is not the case for most current models.

### Why do Microsoft and NVIDIA promote open models if it creates competition for their own products?

Because their core business is not the model itself but the infrastructure underneath it—chips, data centres, cloud capacity. Open models increase demand for that infrastructure regardless of who ultimately deploys the model.

### Is open-weight AI worthwhile for smaller Swiss SMEs without an in-house IT department?

Yes, precisely because operating costs have dropped sharply due to competition, and specialised outsourcing partners can handle the technical complexity without requiring the SME to build its own infrastructure.

### What risk does an SME take on by relying entirely on a closed AI system?

The main risk is vendor lock-in: price increases, changed terms of use, or a forced provider switch can become costly and disruptive because data, workflows and integrations are tightly bound to a single system.

### What does the EU Cloud and AI Development Act have to do with this trend?

The planned Act introduces tiered sovereignty requirements that determine, based on data sensitivity, where and how AI systems may be operated. Open, portable models tend to meet these requirements more easily than closed systems tied to a fixed cloud provider.

## Sources

- [Microsoft and NVIDIA back open-weight models](https://www.linkedin.com/posts/alvinfsc_the-invisible-hand-just-flexed-microsoft-activity-7486563372210061312-yamL)