More Meta Partnerships, Alibaba's New AI Chip, Microsoft Challenges OpenAI, xAI's New AI Model, Apple's AI Chatbot, & AI Transforms Drugmaking
Meta in talks with Google & OpenAI, Alibaba's new AI chip could fill China void, Microsoft releases new AI models, Grok Code Fast 1 is released, Apple has an internal AI chatbot, and AI for drugmaking
Meta’s AI Strategy Shifts with Google and OpenAI Partnership Talks
Meta is in talks to integrate Google's Gemini and OpenAI's models into their AI products, following internal chaos marked by the brief departure and rehiring of ChatGPT co-creator Shengjia Zhao as chief AI scientist, exposing ongoing struggles with their AI development.
This pivot prioritizes rapid deployment over proprietary innovation, a pragmatic response for a $1.8 trillion giant that's lagging in the AI race despite aggressive talent grabs like Scale AI's CEO Alexandr Wang.
By partnering with rivals, Meta could quickly enhance offerings like WhatsApp and Instagram AI features, closing the gap on leaders like Google and OpenAI while buying time to stabilize Llama. We previously noted Zuckerberg's hiring spree was an admission that external help was needed. This extends that logic, as partnerships could help acquire users amid Llama's delays.
However, relying on other models risks diluting Meta's open-source ethos and could make their key hires second-guess joining. Can Meta keep developing Llama after these partnerships? If not, this could accelerate a talent exodus and investor skepticism.
Alibaba's New AI Chip Targets Nvidia's Void in China
Alibaba has developed a new AI chip in response to ongoing US export controls and geopolitical tensions with China. This move is a direct challenge to Nvidia’s dominance in the Chinese market and signals a major strategic pivot for the Chinese tech giant.
The new chip is designed for a broader range of AI inference tasks, such as running chatbots and image generators, making it more versatile than Alibaba's more specialized chips. It is also being manufactured by a Chinese company, a departure from an earlier Alibaba chip that was fabricated by Taiwan Semiconductor Manufacturing.
While the chip's performance benchmarks are not yet public, its compatibility with existing Nvidia software frameworks is a key strategic advantage, as it makes it easier for developers to transition. This move positions Alibaba as a serious contender to fill the void left by Nvidia’s export restrictions.
However, the chip is designed for inference, not the more compute-intensive task of model training, where Nvidia’s chips still have a commanding lead. This highlights the ongoing challenge for Chinese companies to create a full-stack, end-to-end AI ecosystem capable of competing with the most advanced US systems. It’s a significant step, but the gap in high-end training chips and the broader software ecosystem remains a major hurdle.
Microsoft’s Escalates OpenAI Competition with New AI Models
Microsoft debuted MAI-1-preview and MAI-Voice-1, signaling a pivot from its heavy reliance on OpenAI’s tech.
Microsoft has access to OpenAI’s IP until 2030, and they aren’t waiting to diversify their AI offerings. MAI-1-preview powers Copilot’s text capabilities, while MAI-Voice-1 generates a minute of audio in under a second on a single GPU, targeting on-device efficiency for Microsoft’s software ecosystem.
This move is positioning Microsoft as a direct competitor to OpenAI, despite their $14 billion partnership. We wrote in June about OpenAI’s success seizing Microsoft accounts, and Microsoft is responding.
MAI-Voice-1’s efficiency could enable AI-driven features in edge devices, reducing cloud dependency and costs compared to OpenAI’s compute-heavy models.
However, Microsoft’s slower integration of new OpenAI models, as noted in June, suggests internal friction and a strategic bet on self-reliance. Can Microsoft’s models match OpenAI’s performance, or will they struggle to differentiate in a crowded enterprise market?
The shift underscores Microsoft’s need to control its AI destiny while leveraging its massive user base across Windows and Edge.
xAI’s Grok Code Fast 1 Targets Developers with Speed and Savings
xAI launched Grok Code Fast 1, a lean AI model for autonomous coding tasks, touted as “speedy and economical.” Designed for agentic programming, it handles multi-step coding jobs and integrates with tools like GitHub Copilot, offering a free trial to attract developers.
At $1.50 per million output tokens, Grok Code Fast drastically undercuts Grok 3 ($15 per million output tokens), and OpenAI’s GPT-5 ($10 per million output tokens).
Its efficiency could disrupt the market for startups and developers seeking affordable AI tools.
And adoption has been high according to OpenRouter, where it is ranked #1 and taking significant share from Claude Sonnet 4.
But that could be due to xAI giving it away for free. The real test will come once xAI starts charging.
Apple Launches Internal AI Chatbot for iPhone 17
Apple launched Asa, an internal AI chatbot for retail staff, to boost product knowledge ahead of the iPhone 17 launch. Asa delivers conversational access to specs, use cases, and sales tips, enhancing employee expertise without public-facing AI like Siri.
This internal focus contrasts with Apple’s stalled “Apple Intelligence” and rumored $14 billion Perplexity acquisition. Asa’s practical application could streamline sales, but its limited scope suggests Apple is still cautious about consumer AI.
AI Transforms Drugmaking, but Hallucinations Pose Risks
A PharmaVoice report details AI’s transformation of drugmaking, with companies like Novo Nordisk and Immunai slashing R&D timelines. Successes like Insilico Medicine’s AI-designed pulmonary fibrosis drug and DeepMind’s AlphaFold highlight AI’s potential to cut costs and accelerate discoveries.
AI’s ability to analyze vast datasets outpaces traditional methods, but the report warns of “hallucinations” and data quality issues risking faulty outcomes.
Unlike software, where bugs can be patched, drugmaking errors could be catastrophic, demanding rigorous validation. Compared to Google’s Genie 3, which simulates physics-aware environments, drugmaking AI needs similar precision.
Can the industry balance speed with safety, or will overreliance on AI lead to costly mistakes?