NEW DELHI — In March 2026, Google placed a hard cap on Meta’s access to its Gemini artificial‑intelligence models after Meta asked for additional computing capacity that Google said it could not provide. Internal communications obtained by the Times of India show the restriction forced Meta’s engineering teams to ration “AI tokens” and slowed several internal AI projects, prompting project leads to prioritize only the highest‑impact use cases while postponing lower‑priority experiments. The move came just weeks before Meta’s AI chief Alexandr Wang publicly teased an upcoming Muse Spark update that the company says will reduce its reliance on external AI models for coding and agentic tasks.
What happened
According to the internal documents cited by the Times of India, Meta approached Google in early 2026 requesting a larger share of the cloud‑based compute that powers Google’s Gemini family of large language models. Google responded that it could not accommodate the request and imposed a usage ceiling on Meta’s access. The cap applied to the number of “AI tokens” – a unit Google uses to meter consumption of its generative‑AI services – that Meta could consume each day. Meta engineers were instructed to allocate those tokens to projects deemed “high‑impact,” while other experiments were placed on hold.
The timing of the restriction is notable because, on 12 April 2026, Alexandr Wang announced that Meta would soon roll out Muse Spark, an internal model update designed to cut the company’s dependence on rival AI services for tasks such as code generation and autonomous‑agent functions. Wang’s teaser framed Muse Spark as a step toward greater self‑sufficiency in Meta’s AI stack.
Google declined to comment on the specific capacity request but released a brief statement that the company “prioritises equitable access to its AI infrastructure for all partners.” Meta has not issued an official comment on the token‑rationing measures, though the internal documents show that senior engineers were briefed on the new limits and asked to re‑evaluate project road‑maps.
Why it matters
The episode highlights the growing tension between the world’s largest cloud providers and the tech firms that depend on them for AI compute. As generative‑AI models become larger and more compute‑intensive, access to high‑performance infrastructure is increasingly a strategic asset. By capping Meta’s usage, Google effectively limited the pace at which Meta could experiment with Gemini‑based applications, potentially slowing product roll‑outs that rely on the model’s capabilities.
For Meta, the restriction underscores the risk of relying on a competitor’s AI platform for core functions. Muse Spark, if successful, could give Meta a proprietary alternative for coding assistance and agentic workloads, reducing its exposure to external capacity constraints and to any future policy or pricing changes Google might impose. A shift toward internal models would also alter the competitive dynamics of the AI services market, where Google, Microsoft, Amazon and others vie for enterprise customers.
Background and context
Google’s Gemini models are part of the company’s broader generative‑AI offering, marketed to enterprise partners through Google Cloud. The models are accessed via an API that charges by the number of tokens processed, a common pricing structure in the industry. Meta has historically leveraged external AI services to supplement its own research, especially for tasks that require massive compute resources beyond its in‑house data centres.
Alexandr Wang, Meta’s head of AI, is the company’s highest‑paid employee and has been a vocal advocate for building “self‑sufficient” AI capabilities. In recent months, Meta has announced a series of internal initiatives aimed at reducing reliance on third‑party models, citing concerns over cost, data privacy and strategic independence. Muse Spark, the update teased in April 2026, is positioned as the latest milestone in that effort, targeting “coding and agentic tasks” that currently lean heavily on external services.
Competing claims and uncertainty
Google’s public statement frames the cap as a matter of “equitable access,” suggesting that the limitation was driven by resource constraints rather than a deliberate attempt to hinder a competitor. However, the internal timing—coinciding with Meta’s imminent Muse Spark announcement—raises questions about whether the decision also served a competitive purpose.
Meta’s internal response, as reflected in the token‑rationing directives, indicates that the company was forced to re‑prioritize projects, but the documents do not reveal the extent of the disruption. No quantitative data on delayed milestones or financial impact has been disclosed. Likewise, the efficacy of Muse Spark remains untested; while Wang’s teaser promises reduced dependence on rival models, the update has not yet been released, and independent assessments of its performance are unavailable.
Analysts familiar with cloud‑AI partnerships note that capacity caps are not uncommon when demand outstrips supply, especially during periods of rapid AI model rollout. Yet they also point out that large tech firms can leverage such caps to exert pressure on rivals. Without concrete evidence of intent, the precise motivations behind Google’s decision remain ambiguous.
What to watch next
– Muse Spark rollout – Meta is expected to launch the Muse Spark update in the coming weeks. Independent benchmarks of its coding and agentic capabilities will indicate whether the company can meaningfully reduce its reliance on Gemini.
– Google‑Meta cloud relationship – Follow any further statements from Google regarding capacity allocations for partners, as well as any renegotiated terms that might emerge after the Muse Spark launch.
– Industry response – Competitors such as Microsoft and Amazon may comment on the broader issue of AI‑compute scarcity, potentially influencing pricing or capacity‑sharing policies across the sector.
– Regulatory scrutiny – Given growing attention to competition in the AI market, regulators could examine whether capacity caps constitute anti‑competitive behavior, especially when they affect major rivals.
Conclusion
The March 2026 cap on Meta’s access to Google’s Gemini models illustrates the fragile interdependence of today’s AI ecosystem, where cloud providers control the compute that powers large language models and AI‑first companies depend on that compute to accelerate product development. While Google attributes the limitation to resource constraints, the proximity of the cap to Meta’s Muse Spark announcement fuels speculation about strategic maneuvering. The coming weeks will test whether Meta’s push for self‑sufficiency can offset the short‑term disruption caused by the token‑rationing, and whether the broader market will see a recalibration of power between cloud AI providers and the firms that build on their platforms.
Sources
Times of India, “Weeks before Meta’s highest‑paid employee Alexander Wang told the world that the company’s AI model update is coming, Google sent Meta a message ‘We cannot’,” https://timesofindia.indiatimes.com/technology/tech-news/weeks-before-metas-highest-paid-employee-alexander-wang-told-the-world-that-the-companys-ai-model-update-is-coming-google-sent-meta-a-message-we-cannot-/articleshow/132215347.cms
Story synopsis gathered from: Times of India – Top Stories — source
Corrections
If you believe this article contains an error, contact Herald Express with the source URL and supporting evidence.
