Google Cuts Meta Off Gemini Capacity as $500 Billion AI Push Stalls

Q: Why do Meta's AI engineers call their half-trillion-dollar project 'the gulag'?

Meta engineers call it “the gulag” because many say the work is compulsory, repetitive and demoralizing despite the project’s enormous budget. Reporting on Meta’s 6,500-person Applied AI unit describes engineers being reassigned into the group with little real choice — effectively “join or leave” — leading some to call themselves “draftees.” Inside the team, employees say they were pulled away from the kind of product engineering they were hired to do and instead given labor-intensive tasks such as generating puzzles and difficult coding prompts to train and evaluate AI systems. Workers quoted in the coverage described that work as “menial,” “mechanical,” “soul-crushing,” and offering “zero purpose.” One employee said the experience felt isolating: little interaction, constant weekly tasks, and limited sense of impact. A very high manager-to-employee ratio in the unit reportedly added to the feeling of being processed rather than supported. The nickname also reflects the broader climate around Meta’s AI push. The company has committed more than half a trillion dollars over two years, but the effort has been marked by low morale, internal conflict, layoffs, leadership churn, and pressure to deliver after setbacks such as Google cutting off access to Gemini capacity. That combination — forced placement, drudgework, weak morale, and intense pressure inside a troubled reorganization — is why some engineers use the term.

3 articles · Updated · Futurism · Jul 2

Google earlier this year stopped supplying Gemini AI capacity to Meta after Meta’s demand became exceptionally high, according to the Financial Times.
That cutoff slowed Meta projects including data moderation and customer service, underscoring how much the company had relied on Google’s models rather than its own Llama systems.
Meta chose Gemini because it outperformed its in-house tools, but Google was also struggling to satisfy other customers’ AI needs amid broader capacity constraints.
More than $500 billion in planned AI spending over two years has not prevented internal strain at Meta, where morale, infighting, executive departures and token-use cutbacks have clouded its AI push.
Muse Spark, Meta’s new multimodal reasoning model, is now being positioned as a way to reduce dependence on outside models as AI access costs rise across the industry.