This Week in AI: July 4 – 10, 2026
According to multiple reports, Mark Zuckerberg broke his silence on X to launch Muse Spark 1.1 and Meta's first paid developer API, priced below Anthropic's listed rates. The same week, Meta confirmed plans for a cloud business, OpenAI made GPT-5.6 public, DeepSeek was reported to be building its own AI chip, and Anthropic signed a long-term data center lease with TeraWulf.
On this page+
- The Big Story: Zuckerberg Breaks His Silence on X to Launch Muse Spark 1.1 and Meta's First Paid Model API
- Source
- Meta Confirms It's Building a Cloud Business, According to Reports
- Source
- OpenAI Ships GPT-5.6 Publicly, Ending Government-Gated Preview
- Source
- DeepSeek Is Reportedly Building Its Own AI Chip
- Source
- Anthropic Signs a Long-Term Data Center Lease With TeraWulf
- Source
- Strategic Takeaway
The Big Story: Zuckerberg Breaks His Silence on X to Launch Muse Spark 1.1 and Meta's First Paid Model API
According to Bloomberg, Mark Zuckerberg posted on X for the first time in years on July 9 to unveil Muse Spark 1.1 and the Meta Model API, Meta's first paid platform for outside developers, notably choosing X over Meta's own Threads to make the announcement. Per Meta's own blog post, Zuckerberg described the model as strongest at "agentic performance, tool use, and computer use," with a large context window and the ability to split work across parallel helper agents. Fortune reported that Muse Spark 1.1 beats Gemini on coding and reasoning benchmarks and surpasses older OpenAI and Anthropic models on several verticals, though according to that same reporting it still trails the current flagships, Claude Fable 5, Mythos 5, and GPT-5.6, on top-end coding tasks. Bloomberg noted that Meta priced API access well below Anthropic's listed rates for Claude Sonnet 5, landing closer to Anthropic's cheapest model, Claude Haiku 4.5.
Source
Meta Confirms It's Building a Cloud Business, According to Reports
The Motley Fool and CNBC reported that Meta confirmed plans to sell its surplus AI computing power as a standalone cloud offering, internally called Meta Compute, putting it in direct competition with Amazon, Microsoft, and Google for the first time. Per Meta's own newsroom announcement, the company followed up on July 9 by breaking ground on a large AI data center in Sturgeon County, Alberta, described as Meta's largest facility outside the United States.
Source
OpenAI Ships GPT-5.6 Publicly, Ending Government-Gated Preview
According to CNBC, OpenAI publicly released its GPT-5.6 family, Sol, Terra, and Luna, on July 8, after weeks of limiting the rollout to a small group of government-approved companies at the request of the US government. Per OpenAI's release notes, Sol targets frontier reasoning and long-horizon agentic work, Terra is positioned to match the prior flagship's performance at a lower cost, and Luna is the fastest, cheapest tier of the family.
Source
DeepSeek Is Reportedly Building Its Own AI Chip
Citing sources familiar with the matter, Reuters reported that DeepSeek is developing an inference-focused AI chip of its own, aiming to cut its dependence on Nvidia and Huawei hardware. That reporting says the Hangzhou-based company has been hiring chip-design engineers and holding talks with foundry and memory partners. The same reporting describes Huawei as holding a dominant share of China's domestic AI chip market, with Alibaba and Baidu also fielding their own chips.
Source
Anthropic Signs a Long-Term Data Center Lease With TeraWulf
Per a TeraWulf regulatory filing and CNBC's reporting, Anthropic agreed to a long-term lease of a TeraWulf-built data center on a site in Hawesville, Kentucky, previously used by Century Aluminum. Multiple outlets described it as one of the largest single-site compute commitments disclosed by any AI lab to date.
Source
Strategic Takeaway
By most accounts, Meta had the loudest stretch of any lab recently. In short order it reportedly confirmed a new cloud business aimed at AWS and Azure, broke ground on its largest data center outside the US, and shipped a model priced to undercut Anthropic directly. That same pattern shows up at OpenAI, DeepSeek, and Anthropic: each appears to be betting that owning the full stack, chips, compute, and now distribution, matters as much as any single model's benchmark scores.
