NORBERT BIEDRZYCKI

The New AI Power Struggle: OpenAI, Microsoft, Google and the Battle for the AI Platform

The New AI Power Struggle: OpenAI, Microsoft, Google and the Battle for the AI Platform

Within a single year, generative AI has rearranged the competitive map of the technology industry. What is at stake is not a product category. It is control of the next platform.

The speed of strategic moves among the largest technology companies in 2023 has been remarkable.

Microsoft deepened its partnership with OpenAI at the start of the year and began embedding AI assistants across its products, from search to office software to developer tools. Google, which invented the transformer architecture behind most of these models, declared an internal “code red”, launched Bard, merged its two AI research units into Google DeepMind and this month unveiled Gemini. Amazon committed up to $4 billion to Anthropic, while Google also agreed to invest up to $2 billion in the same company. Meta took a different route and released its Llama models openly, betting that an open ecosystem would weaken the advantage of closed competitors.

And in November the world watched the most dramatic governance episode in recent technology history, when OpenAI’s board dismissed its CEO, Sam Altman, only to reinstate him five days later after most of the company’s employees threatened to leave.

What the OpenAI crisis revealed

For investors, that weekend was more than corporate drama. It exposed a structural question that had been easy to ignore: what exactly do investors own when they invest in a frontier AI company?

OpenAI is governed by a non-profit board whose mandate is not to maximise returns. Its largest investor had no board seat. Its value proved to be concentrated almost entirely in its people, who could have moved to another employer within days. When I sat on boards, governance was something investors scrutinised before they wrote a cheque. In this sector, the rush to secure access to frontier models has sometimes put strategic urgency ahead of conventional governance.

Four sources of control

When I try to map who will capture value from generative AI, I see four sources of control.

The first is models: the ability to train state of the art systems. This requires talent, capital and compute, and is concentrated in a small number of players.

The second is compute and cloud infrastructure. Every model, open or closed, runs on someone’s hardware. The hyperscalers sell capacity to everyone, including their competitors’ partners.

The third is distribution: the products that hundreds of millions of people and businesses already use every day. Office software, search, operating systems, mobile devices, enterprise platforms. Adding AI to an existing product with a huge installed base is a much easier route to revenue than persuading customers to adopt a new tool.

The fourth is data and customer relationships in specific industries, which no general purpose model owns.

The companies that combine several of these sources of control are in the strongest position. That is why the partnerships between model developers and cloud and software platforms matter so much. They are not just commercial agreements. They are attempts to assemble a complete stack.

Many winners or a few?

The question for investors is whether this market will support many dominant platforms or consolidate around a handful of winners. Historically, platform markets tend toward concentration, because scale economies and network effects reinforce each other. On the other hand, open models are improving quickly, and it is possible that model capability will become less differentiated over time, shifting power back to distribution and applications.

My view at the end of 2023 is that we will see an oligopoly at the model and infrastructure layer, and much more fragmentation at the application layer. For most investors, the practical implication is to avoid competing head on with the platforms, and instead to back companies that benefit from the competition between them.

Investment question: Who will capture the most economic value from generative AI: model developers, cloud platforms or companies controlling customer distribution?

Read next

  1. The Business of AI 2020–2026: about this series

  2. The AI Investment Paradox: Record Capital, But Where Are the Returns?

    The Business of AI series, written in 2026

  3. The Physical AI Opportunity: Investing Beyond Software

    The Business of AI series, written in 2026