PLUS: Alibaba's new 2.4T parameter model and the debate over AI-written sermons
Happy reading
The race to build AI's hardware foundation is escalating, with TSMC doubling down on its U.S. expansion. The chipmaking giant is injecting another $100 billion into its Arizona facilities to meet the relentless demand for next-generation AI hardware.
This monumental investment is a major step toward onshoring the AI chip supply chain, despite much higher construction costs in the U.S. But is this massive bet enough to secure the hardware future of AI, or simply a sign that the infrastructure race has no finish line?
In today’s Next in AI:
TSMC's $100B bet on U.S. AI chip manufacturing
Alibaba’s 2.4T parameter open-weight model
AMD’s next-gen AI accelerator
The debate over AI-written sermons
TSMC's $100B AI Bet
Next in AI: TSMC is pouring an additional $100 billion into its Arizona chip factories, accelerating a massive U.S. expansion to meet what its CFO calls a "multi-year demand mega trend" for AI hardware.
Explained:
The fresh commitment brings TSMC's total investment pipeline in Arizona to $265 billion, supported by an increased full-year capital expenditure forecast of up to $64 billion.
This investment will fund at least four additional 2-nanometer fabs and advanced packaging facilities, which are essential for creating the next generation of powerful AI chips.
Despite U.S. fab construction costs being four to five times higher than in Taiwan, the company sees the move as critical for capturing overwhelming demand from its leading U.S. customers.
Why It Matters: This monumental investment is a major step toward building a more resilient AI semiconductor supply chain on U.S. soil. It also signals that the hardware infrastructure race powering the AI boom is nowhere near slowing down.
Alibaba's AI Answer
Next in AI: Alibaba has announced Qwen3.8, a colossal model that will be released as open-weight. This signals another major entry from China in the global race to build frontier AI.
Explained:
At 2.4 trillion parameters, it's the first Qwen model to cross the trillion-parameter threshold and is claimed to be second only to Anthropic's Fable 5.
While the full model is promised as open-weight "soon," a preview version is currently accessible only through Alibaba's paid APIs like Token Plan and Qoder.
The announcement currently lacks a model card, license details, or independent benchmarks, leaving the community to rely on internal claims for now.
Why It Matters:
Alibaba's move to release its largest model as open-weight signals a significant strategic shift and intensifies the global competition in frontier AI development. This could provide developers worldwide with access to an exceptionally powerful model, assuming the final release lives up to its initial claims.
AMD's Next-Gen AI Chip
Next in AI: A deep-dive into LLVM commits gives us the first detailed look at AMD’s upcoming GFX1250 AI accelerator. The new architecture reveals a ground-up redesign focused squarely on high-performance computing and machine learning workloads in the datacenter.
Explained:
The new chip design features a major upgrade in processing power, allowing each wave to address up to 1024 vector registers—a fourfold increase over previous consumer architectures that enables more complex operations.
It introduces new tensor operations that combine the simpler programming model of AMD's consumer GPUs with the high-performance capabilities of its datacenter line, streamlining development for AI tasks.
AMD is also adding features that mirror NVIDIA's, like Thread Block Clusters for managing large parallel workloads, and pairing it with a massive 432 GB of HBM4 memory to handle enormous AI models.
Why It Matters: AMD's architectural overhaul signals a serious, direct challenge to NVIDIA's dominance in the AI accelerator market. This level of competition is set to accelerate innovation and provide more powerful, efficient hardware choices for developers and enterprises.
AI in the Pulpit
Next in AI: A growing number of pastors are using AI assistants like ChatGPT to research and outline sermons, saving valuable time but sparking a debate over technology's role in ministry. The practice is quickly becoming more common, as new research finds that nearly a quarter of U.S. pastors now use AI for sermon prep.
Explained:
The adoption rate has doubled since early 2024, with 24% of pastors now actively using AI to help write or edit sermons, primarily for brainstorming and initial research.
Real-world financial and time pressures are a major driver. One pastor, a top 1% ChatGPT user, cut his sermon preparation time from 15 hours to just a few by using AI for first-pass research.
Significant accuracy issues remain a concern, with the CEO of YouVersion warning that AI often misquotes Scripture in as many as 60% of cases, suggesting the technology is not yet reliable for theological questions.
Why It Matters: This trend demonstrates AI's expansion into deeply human and spiritual domains, testing the boundaries between technological assistance and authentic expression. It also shows how AI is becoming an essential productivity tool for professionals in every field, even those you'd least expect.
AI Pulse
Researchers found that giving people AI advice caused their accuracy on difficult questions to drop from 27% to 9%, while their confidence nearly tripled, a phenomenon known as cognitive surrender.
An interdisciplinary report from leading AI and neuroscience researchers concludes there are no obvious technical barriers to creating conscious AI systems, prompting calls for an urgent ethical framework to manage the possibility.
A widely-circulated essay argues the tech industry is in a state of mass psychosis over AI, claiming that nearly all corporate AI projects are failing and it has become impossible to have rational conversations about the technology's actual impact.