Meta spends $14B on AI without significant progress: strategic failure?

What happened?

According to a report by Xataka, Meta spent at least $14 billion on its bet on artificial intelligence, including a 49% investment in Scale AI valued at $14.3 billion (according to Reuters) and the hiring of its founder, Alexandr Wang, to lead the company's superintelligence efforts. However, a year later, Meta has not achieved significant progress in the chatbot and language model market, where it still lags behind ChatGPT and Gemini.

The investment in Scale AI, announced on June 13, 2025, valued the startup at around $29 billion. Scale AI focuses on data labeling and curation for training AI models, a critical but low-profile business. Meta not only injected capital but also integrated Alexandr Wang, founder and CEO of Scale, to lead its superintelligence efforts. This move reflects a dual strategy: securing access to high-quality data and attracting top-tier talent.

However, the results have not followed. Meta's Llama models, although open-source and popular among developers, have failed to capture mass consumer attention. OpenAI's ChatGPT surpasses 400 million weekly active users (according to OpenAI data from May 2025), while Google's Gemini is integrated into the Android and Workspace ecosystem. Meta AI, the assistant based on Llama, is available on Facebook, Instagram, and WhatsApp, but its usage remains marginal compared to its competitors.

Why is this important?

This situation shows that money and resources do not guarantee success in the AI race. Meta, with billions of users and enormous distribution capacity, has failed to translate its investment into differentiated products. The lack of significant progress casts doubt on its strategy and ability to compete in a market dominated by OpenAI and Google.

Historically, Meta has proven to be an effective imitator: it copied Snapchat with Stories, TikTok with Reels, and Twitter with Threads. But generative AI seems to be a different challenge. It is not enough to distribute a product; innovation in models and user experience is needed. While OpenAI launches GPT-4o with multimodal capabilities and Google deploys Gemini 1.5 Pro with a 1-million-token context window, Meta has focused on open models that, although useful for the research community, do not generate direct revenue or retain users.

Moreover, the investment in Scale AI underscores the importance of training data but does not solve the differentiation problem. Having the best data does not guarantee having the best product if it is not combined with a solid product strategy. This is a pattern we have seen in other companies: Google+ failed despite Google's resources, and Microsoft lost the smartphone war despite its investment in Nokia.

Consequences for Meta and the industry

For Meta, the consequences could be a rethinking of its AI strategy, possibly betting on acquisitions or deeper collaborations. Investor pressure is growing: Meta's AI spending in 2024 exceeded $30 billion (according to its annual report), and it is expected to increase in 2025. If it does not translate into revenue, cuts or changes in direction could occur.

For the industry, it shows that the advantage in data and users is not enough without innovation in models and products. Companies like Anthropic (with Claude) and Mistral are proving that small, agile teams can compete with giants. Additionally, the investment in Scale AI highlights the importance of training data but does not solve the differentiation problem.

Another consequence is the impact on the AI job market. The hiring of Alexandr Wang can be interpreted as a sign that Meta desperately needs external talent, which could trigger a war for salaries and conditions. However, it could also indicate that Meta does not fully trust its internal AI team, which could affect morale and employee retention.

What readers should know

Meta is not the only tech giant struggling to find its place in AI. Unlike Microsoft (with OpenAI) or Google (with Gemini), Meta has attempted its own approach, but without visible results. The lesson is that in AI, the speed of innovation and the ability to launch attractive products are more important than the budget.

Furthermore, the open-source nature of Llama, although celebrated by the community, has not generated a clear competitive advantage. While OpenAI and Google protect their models as trade secrets, Meta gives its away, hoping the community will improve it. But so far, no external improvement has led to a flagship product.

Finally, users should be aware that Meta's AI is integrated into the apps they use daily, but they may not notice it. Meta AI can suggest responses in Messenger or generate images on Instagram, but these features are discreet and have not changed brand perception. For Meta to win the AI race, it needs a product that users want to actively use, not just a passive feature.