USD.AI (CHIP) Surge

There’s a two tier world for High Performance Compute (HPC). Hyperscalers (OpenAI, XAi, OpenAI, Google, etc), have access to near infinite financing, but are constrained by power, water, and land.

Elon Musk's "Terafab" Chip Factory Vision Begins with a $55 Billion Bet SpaceX has formally proposed investing $55 billion to initiate construction of a "Terafab" chip manufacturing facility in Grimes County, Texas, with the total cost potentially reaching $119 billion in later phases. This massive project, a joint initiative by SpaceX and Tesla, marks a pivotal step in Elon Musk's strategy of vertical integration for his company ecosystem. The core logic is that Musk's ventures—SpaceX, Tesla, xAI, and future projects like the Optimus robot—consume enormous amounts of AI computing power. Terafab is envisioned not merely as a factory but as a "full-stack AI infrastructure strategy," aiming to bring chip production, energy sourcing, and compute deployment under one umbrella to secure a self-sufficient supply of this critical resource. Analysts describe this as a bold "15-year strategy" with significant execution risks. Building a leading-edge semiconductor fab requires 3-5 years, specialized equipment like ASML's EUV lithography machines, and a skilled workforce, with the earliest chip output not expected until mid-2028 at best. It mirrors a broader industry trend where giants like Microsoft and Google are also pouring billions into custom AI chips, driven by the belief that in the AI era, controlling computing power means controlling the future. Timed alongside SpaceX's impending IPO, the Terafab announcement also serves as a powerful narrative, linking Tesla to SpaceX's and AI's growth story. Whether the vision translates into a functioning foundry remains uncertain, but Musk's move to have a rocket company build chips is redefining industry boundaries once again.

The 2026 Q1 earnings season reveals a paradox: while AI helps companies freeze hiring and boost productivity, the soaring costs of AI inference—token consumption and GPU depreciation—are eroding savings from workforce reductions. Shopify reported that AI now writes over 50% of its code, enabling significant output with stable headcount. However, LLM costs, driven by heavy usage of its AI assistant Sidekick, are pressuring its subscription毛利率. Similarly, Roblox attributed a quarter of its full-year利润率下调 to increased AI investment. The article highlights a broader industry imbalance: combined AI capital expenditure for Amazon, Meta, Microsoft, and Google is projected to reach $725 billion in 2026, vastly outpacing potential savings from layoffs. For instance, Meta's planned裁员 would save about $2.4 billion annually, offsetting only ~12% of its incremental AI depreciation. While底层 model and chip suppliers like NVIDIA maintain high profitability, application-layer companies face a pricing squeeze. Their strategies now involve要么 tightly linking AI costs to user engagement (like Shopify) or introducing fees for advanced AI features (like Roblox), as covering AI bills with裁员 savings alone is financially unsustainable.

"Cerebras, a pioneering AI chip company, successfully debuted on NASDAQ (CBRS) on May 14, 2026, with its stock price surging approximately 68% on the first day. This marks a significant milestone following a decade-long journey, as recounted by early investor Steve Vassallo. The story begins not in 2016, but with the deep, 19-year relationship between Vassallo and founder Andrew Feldman, which started with Feldman’s previous company, SeaMicro (acquired by AMD in 2012). In 2016, Feldman and a core team of chip and system experts sought to challenge the emerging consensus. At a time when AI’s practical utility was still debated and GPUs were becoming the default hardware, they envisioned a fundamentally new computer architecture purpose-built for AI workloads. They identified memory bandwidth, not raw compute power, as the critical bottleneck for neural networks. Defying industry inertia, Cerebras pursued a radical, wafer-scale chip design—58 times larger than the biggest existing chips. This meant confronting and solving a cascade of unprecedented engineering challenges: power delivery, thermal management, and maintaining electrical continuity across tens of thousands of connections. It required reinventing nearly every aspect of modern computing—semiconductors, systems, data structures, software, and algorithms. The path was fraught with setbacks, including a prototype that caught fire on its first power-up. Progress was marked by intense, iterative problem-solving, with the board meeting every 6-8 weeks to tackle the latest technical frontier. Through disciplined perseverance and deep trust within the team, they achieved a breakthrough in August 2019 when their first wafer-scale computer successfully operated. Feldman’s drive for a 1000x leap, his formative upbringing among intellectual giants who modeled both brilliance and kindness, and his belief in building a loyal, mission-driven team were central to Cerebras’s culture. His competitive strategy was that of David vs. Goliath—finding innovative, human-centric approaches that larger incumbents would overlook. From the symbolic delivery of the first term sheet over a backyard fence in 2016 to the NASDAQ bell ringing in 2026, Cerebras’s journey is a testament to long-term vision, technical audacity, and the power of foundational founder-investor relationships. It stands as a reminder that the computing revolution can come not just from more GPUs, but from a complete reimagining of the architecture itself."

The global AI chip market is undergoing a significant paradigm shift, with ASICs (Application-Specific Integrated Circuits) emerging from a niche to a mainstream force, challenging the long-held dominance of GPUs in AI training. This "golden era" for ASICs is primarily driven by the industry's pivot from training to inference, where the cost and energy efficiency advantages of custom chips become critical for scaling to billions of users. Key signals include Google's TPU capturing 78% of its AI server shipments in Q1 2026, OpenAI's plans for a massive custom ASIC cluster with Broadcom, and cloud providers (CSPs) increasingly favoring in-house or custom designs for supply chain control and cost efficiency. Market forecasts are bullish: AI ASIC revenue is projected to hit $300 billion by 2027, with a 34% CAGR, potentially reaching a 45% share of the AI chip market. The competitive landscape is expanding beyond traditional leaders Broadcom and Marvell. MediaTek is aggressively targeting the data center ASIC market, projecting over $10 billion in 2026 revenue, while Qualcomm, leveraging its AlphaWave acquisition, is launching customized inference chips. These mobile chip giants are leveraging their SoC design expertise for a cloud-side transition. In China, companies like VeriSilicon and ASR Microelectronics are capitalizing on this trend as pivotal "enablers," providing full-stack ASIC design services and experiencing explosive order growth, particularly for cloud-side AI projects. However, challenges remain: high development costs, software ecosystem gaps compared to NVIDIA's CUDA, dependency on advanced packaging capacity (like TSMC's CoWoS), and the fundamental trade-off between customization and flexibility. The future is not a simple replacement of GPUs by ASICs but a more specialized coexistence. The consensus points toward "GPUs for training, ASICs for inference," or hybrid clusters. Ultimately, the rise of ASICs represents a democratization of computing power, shifting definition authority from a single chip giant to a broader ecosystem of cloud providers and end-users, offering the industry more choice in the silicon that powers AI.