AI Competition Intensifies as Giants Adjust Strategies and Forge Alliances
In early July 2024, the global artificial intelligence industry saw a series of major developments, from model security controversies and landmark open-source collaborations to massive chip manufacturing deals, all signaling a new phase of technological and commercial competition. The conflict between Alibaba and Anthropic has become public, while Moonshot AI’s Kimi has partnered with Microsoft’s GitHub, illustrating the complex landscape of fragmentation and consolidation within the industry.
Alibaba-Anthropic Tensions Escalate: Claude Code Banned Over Backdoor Risks
On July 3, Alibaba Group internally confirmed that it has added the code generation model Claude Code to its high-risk software list due to a reported security risk involving a potential backdoor. According to an internal memo, starting July 10, 2024, Alibaba will completely ban employees from using Claude Code in the work environment and recommends its self-developed intelligent programming assistant, Qoder, as an alternative.
The immediate cause for the ban was an analysis that found Claude Code contained code snippets capable of detecting if a user is from China, raising concerns about data security and potential backdoors. This incident follows Anthropic’s accusation in June that Alibaba was ‘distilling’ its models—a technique for training a smaller, more efficient model by learning from a larger, complex one, which is typically restricted by terms of service. The tension between the two companies has further escalated. Alibaba’s decision highlights the high level of vigilance that enterprises maintain regarding data security, code transparency, and supply chain risks when using third-party AI services.
Open-Source Milestone: GitHub Copilot Integrates Kimi K2.7 Code
In contrast to the friction between giants, the open-source community is making collaborative strides. On July 3, Moonshot AI announced that its large-scale programming model, Kimi K2.7 Code, which was open-sourced on June 12, has been officially integrated into Microsoft’s GitHub Copilot. This marks the first time the GitHub Copilot platform has integrated a third-party open-source model, having previously relied mainly on closed-source models from OpenAI.
According to the official announcement, Kimi K2.7 Code is hosted by GitHub on the Microsoft Azure cloud platform and will operate on a pay-as-you-go basis, rolling out in phases to Copilot Pro, Pro+, and Max subscribers. It will later be extended to Copilot Business and Enterprise versions. This move is seen as a major breakthrough in the commercial application of open-source AI models, proving their performance has met the integration standards of a top-tier industry platform. Compared to its predecessor, Kimi K2.7 Code boasts significant improvements in long-context programming and instruction-following capabilities, while also being optimized for token consumption, reducing inference costs by 30%.
The AI Hardware and Chip Race: Samsung Lands Major Meta Deal as Apple’s Project Stalls
The competition in artificial intelligence is rooted in powerful computing support, making the race in the chip sector particularly fierce. According to a July 3 report, Samsung Electronics’ foundry business has made significant progress, reportedly securing a deal with Meta to design and produce its next-generation AI Application-Specific Integrated Circuits (ASICs) worth over 10 trillion South Korean won (approximately $7.2 billion). It is understood that future versions of Meta’s self-developed AI accelerator, ‘MTIA,’ are planned for mass production using Samsung’s cutting-edge 2-nanometer process. ASICs are chips customized for specific applications like AI computing and offer higher power efficiency than general-purpose GPUs. If realized, this partnership would significantly boost Samsung’s position in the global AI chip foundry market.
In contrast, Apple is taking a more cautious approach to its AI hardware explorations. According to leaks, Apple’s ‘Project H90,’ an AirPods Pro project featuring a camera, has been ‘paused.’ The project had garnered significant attention and was expected to combine AI visual capabilities to offer users innovative interactive experiences. The project’s suspension may be due to challenges Apple faces with Siri’s visual intelligence features and the optimization of related AI models. The original plan might be postponed until 2027, reflecting that even a company like Apple faces immense engineering and technical hurdles in transforming cutting-edge AI technology into mature, reliable consumer products.
Model Releases and Strategic Reflection: Industry Adjusts Pace Amidst a Boom
AI models themselves continue to iterate rapidly. ByteDance’s Doubao video generation model, Seedance 2.5, is expected to launch on its experience center on July 6, supporting 30-second native video generation and multi-modal inputs, showcasing its continued efforts in the text-to-video field. Tencent Cloud also announced that the official version of DeepSeek-V4 will be available on its model service platform in mid-July, introducing a peak-valley pricing model to explore more flexible commercialization strategies.
However, alongside this technological frenzy, industry leaders are also engaging in sober reflection. Meta CEO Mark Zuckerberg admitted in an internal meeting that despite the company’s full-scale efforts, the development of AI Agents is progressing slower than he anticipated. AI Agents are systems capable of autonomous perception, decision-making, and task execution, and are considered a key path toward higher-level artificial intelligence. Zuckerberg’s statement injects a dose of realism into the currently overheated AI market, acknowledging that the leap from existing large language models to super-intelligent agents capable of autonomously completing complex tasks is still a long road ahead.
