Dive Brief:
- Meta released a terminal coding agent Wednesday called Muse Code, another step toward the company’s pursuit of enterprise spend amid ongoing competition with leading frontier model developers. Muse Code, now in beta, was designed for software engineering tasks such as writing code, validating results, coordinating subagents for tasks and solving engineering problems.
- The Facebook, Instagram and WhatsApp owner also updated its Muse Spark foundation model — which powers Muse Code — boosting its code generation, debugging codebase understanding and end-to-end developer capabilities.
- The release is another way Meta aims to generate revenue from its AI offerings as it heavily invests in compute capacity. Meta upped its capital expenditure forecast for the year to $130 billion, up from $125 billion, executives said during its Q2 earnings call last week. The move is mainly driven by investments in servers, data centers and network infrastructure. CEO Mark Zuckerberg said during the call that Meta was considering selling computing power from its data centers.
Dive Insight:
Meta is diversifying its AI offerings and pursuing enterprise spend amid a sectorwide rush to supply businesses with AI tools and supporting services. Enterprises spent $143 billion on cloud infrastructure in the second quarter of 2026.
Although its core business is in digital advertising for its collection of social apps, leadership has been heavily investing in infrastructure investments to power AI models, business tools and compute capabilities
The social giant entered into a long-term compute agreement with AI cloud platform Nebius in March and expanded its existing partnership with cloud provider CoreWeave in April.
Meta’s pursuit into compute would make the company a competitor to those same vendors, Nick Patience, VP and practice lead for AI at The Futurum Group, told CIO Dive in an email.
“It fits a wider pattern,” Patience said. “Infrastructure owners monetizing excess capacity to each other, reinforcing that power and physical capacity, not model quality, are the scarce assets right now.”
Meta is among a mix of providers that have huge infrastructure requirements, but may have extra capacity for compute, Ed Anderson, distinguished VP analyst at Gartner, told CIO Dive.
Hyperscalers Amazon, Microsoft and Google have built full-service, multicapacity offerings that will account for 67% of data center capacity by 2031. They service a broad set of enterprise needs, Anderson said. What Meta could be offering is likely within the market of AI-optimized infrastructure — supporting AI workflows and model training.
“There’s effectively, insatiable demand for capacity, all the providers are reporting growing backlogs because of their inability to actually fulfill some of their opportunities,” Anderson said. “So there is, for the foreseeable future, lots and lots of demand for this type of capacity.”
Many companies are looking for a mix of cloud offerings, Anderson said; about 80% of enterprises are using a multicloud strategy, with 75% having one major cloud provider and a mix of smaller offerings.
Enterprises in this position could be looking to diversify their cloud offerings, Anderson added. They could also be seeking data centers physically closer to their operations or seek more data sovereignty.
Anderson said the demand-capacity ratio for AI infrastructure is so skewed in favor of cloud providers at the moment that it’s creating opportunities for anyone interested in entering the market.
“If and when we ever reach the point where demand matches capacity, we’re going to see the market really shake out, and that’s when the competitive intensity will really heat up,” he said
