Can AI Thrive Behind Closed Doors? The High Stakes of Openness

Here is what’s new in the AI world.

Following the Code: Will AI Repeat the Open-Source Revolution?

The central value in enterprise AI does not lie in the foundational models themselves, but in the proprietary guardrails, intelligent agents, and unique data that are built on top of them.

While these systems may use open-source models as a base, companies prefer to outsource the complex, "undifferentiated heavy lifting" of core model development and maintenance.

A common narrative in tech suggests that open-source solutions inevitably win due to inherent moral or collaborative superiority.

However, history shows that open source dominates only when a technology becomes a universal infrastructure, something everyone needs but no one gains a competitive advantage from building privately.

This is what happened with Linux in the server operating system market; it became a shared commodity so companies could compete on higher-value applications and data.

A similar dynamic is unfolding in artificial intelligence. While open-source "open weights" models like Meta's Llama or DeepSeek's offerings are rapidly achieving performance close to that of closed models like GPT-4 at a fraction of the cost, a significant market gap persists.

A recent analysis estimates that by continuing to use expensive closed models, the global market is overpaying by roughly $24.8 billion annually.

This is not simply a market inefficiency, but a reflection of what enterprises are truly buying: not just model output, but convenience, reliability, legal indemnification, and integrated safety features.

You cannot sue a GitHub repository, but you can hold a service provider like OpenAI accountable.

The analogy of AI to the Linux model also breaks down on a structural level. Contributing to an open-source software project like the Linux kernel has a low barrier to entry.

In contrast, modifying a massive large language model requires immense computational resources and access to proprietary training data, effectively preventing true community-led development.

Major AI players release powerful models not to foster open collaboration, but as a strategic move to commoditize a competitor's core product while shifting value to their own proprietary platforms and services.

Consequently, the future of AI is moving toward a hybrid, pragmatic model:

In essence, the $24 billion "premium" paid for closed models is not wasted but is being reallocated to purchase reliability, risk mitigation, and specialised capabilities.

The winners in the AI economy will not be ideological purists of open or closed systems, but the pragmatists who expertly combine open-source model infrastructure with proprietary data, services, and guardrails to solve the critical "last mile" of enterprise adoption.

What If You Said No to Musk and Built World-Beating AI?

Two 22-year-old friends from Michigan are gaining prominence in the AI industry after choosing to pursue their own startup over a multimillion-dollar acquisition offer from Elon Musk.

William Chen and Guan Wang, the co-founders of Sapient Intelligence, met in high school and shared ambitious visions for AI.

Their shared belief that an AI surpassing human intelligence is inevitable drove their collaboration:

Their academic partnership continued at Tsinghua University in Beijing, where they earned the support of professors for an ambitious project aimed at overcoming the structural limitations of large language models (LLMs).

Their initial breakthrough was OpenChat, a small but efficient LLM trained on high-quality conversations that gained significant recognition in academic circles.

The success of OpenChat attracted an acquisition offer from Elon Musk's company, xAI. However, Chen and Wang declined the multimillion-dollar deal to focus on their own venture, leading to the creation of Sapient Intelligence.

Their new creation is the Hierarchical Reasoning Model (HRM), an architecture they claim surpasses major AI systems in abstract reasoning tasks.

In a breakthrough this June, a prototype HRM with just 27 million parameters outperformed models from leading companies like OpenAI, Anthropic, and DeepSeek on complex benchmarks, including advanced puzzles and the ARC-AGI test.

The key innovation, according to Chen, is that HRM's two-part recurrent structure mimics human thought by combining deliberate reasoning with fast reflexive responses.

This architecture reportedly leads to far fewer hallucinations than conventional LLMs and already delivers state-of-the-art performance in specialized areas like weather prediction and medical monitoring.

Sapient Intelligence is now planning to open an office in the United States.

Nomura Taps OpenAI's AI to Transform Its Asset Management Division

Japanese financial giant Nomura Holdings has entered into a strategic collaboration with OpenAI. As part of this partnership, Nomura will adopt OpenAI's Deep Research technology and receive strategic support to develop and deploy new AI-powered services.

The goal is to combine Nomura's extensive proprietary financial data with OpenAI's advanced AI models to deliver higher-value investment advice, market analysis, and data solutions to its clients.

Nomura's leadership stated that generative AI has the potential to fundamentally transform financial services beyond just improving efficiency.

The collaboration aims to leverage this technology to create more accessible and secure services for clients while also opening up new revenue opportunities.

In a separate but significant move, Nomura has successfully completed its previously announced acquisition of Macquarie's U.S. and European public asset management business for US$1.8 billion.

This acquisition adds approximately US$166 billion in client assets to Nomura's global asset management platform, significantly expanding its reach in equities, fixed income, and multi-asset strategies.

The newly acquired business will be integrated with Nomura's existing private markets and high-yield units to form a new entity called Nomura Asset Management International. The leadership team will be headed by former Macquarie executive Shawn Lytle as CEO.

Furthermore, Nomura and Macquarie have formalized a strategic partnership that includes the distribution of select Macquarie private funds to Nomura's U.S. clients and collaboration on developing new investment strategies.

A joint working group has also been established to explore further collaborative opportunities.

Together, these initiatives, the OpenAI partnership, and the Macquarie acquisition represent major steps in Nomura's strategy to build a global, technology-driven investment management platform.

Why is OpenAI Buying Equity? A Stake in Thrive for Enterprise Dominance.

OpenAI has announced it is taking an ownership stake in Thrive Holdings, an investment firm launched by its existing backer, Thrive Capital.

Under the strategic partnership, OpenAI will embed its engineering, research, and product teams within Thrive Holdings' portfolio companies to accelerate their adoption of AI technology and improve operational efficiency.

Thrive Holdings acquires and operates businesses in foundational industries, starting with accounting and IT services, that it believes can benefit significantly from AI integration.

The deal represents another example of OpenAI's growing trend of taking equity positions in its partners. OpenAI, which did not disclose financial terms of the agreement, structured the partnership so its stake in Thrive Holdings will grow as the portfolio companies succeed.

This arrangement also serves as a means for OpenAI to gain compensation for the services it provides.

In a related announcement on the same day, OpenAI also expanded its enterprise reach through a collaboration with the consulting giant Accenture. OpenAI will roll out its ChatGPT Enterprise offering to tens of thousands of Accenture employees globally.

Your opinion matters!

Hope you loved reading our piece of newsletter as much as we had fun writing it. 

Share your experience and feedback with us below ‘cause we take your critique very critically. 

Thank you for reading

-Shen & Towards AGI team