--- title: "The \"maneuvering\" of American tech giants" type: "Topics" locale: "en" url: "https://longbridge.com/en/topics/26449503.md" description: "$Microsoft(MSFT.US) $NVIDIA(NVDA.US) The US tech industry has entered a phase of "maneuvering and strategizing". 01 NVIDIA is the hottest player right now. Jensen Huang said, everyone wants to be the first to receive (NVIDIA GPUs). Everyone wants to get the most products. However, it's common knowledge that tech giants have long suffered under NVIDIA. Given any opportunity, they're figuring out how to break free from NVIDIA. The basic strategy is: self-developed chips + externally customized chips + large model agents. Microsoft in November 2024..." datetime: "2025-01-06T03:52:51.000Z" locales: - [en](https://longbridge.com/en/topics/26449503.md) - [zh-CN](https://longbridge.com/zh-CN/topics/26449503.md) - [zh-HK](https://longbridge.com/zh-HK/topics/26449503.md) author: "[萍股相逢](https://longbridge.com/en/profiles/16460454.md)" --- # The "maneuvering" of American tech giants $Microsoft(MSFT.US) $NVIDIA(NVDA.US) The U.S. tech industry has entered a phase of "maneuvering and strategizing." 01 NVIDIA is the hottest player right now. Jensen Huang said everyone wants to be the first to receive their (NVIDIA GPUs). Everyone wants the most products. However, it's no secret that tech giants have long been frustrated with NVIDIA. Given the chance, they're all figuring out how to break free from NVIDIA. The basic strategy is: self-developed chips + externally customized chips + large model proxies. In November 2024, Microsoft launched its first data processor for internal operations, the Azure Boost DPU; in December 2024, Amazon announced it would mass-produce its next-gen AI training chip, Trainium 3, in 2025; Google's TPU has always been high-profile, and in early 2025, it even poached a processor executive from Microsoft. To accelerate chip development, finding custom developers is a common approach. Broadcom, which develops custom chips for Google, ByteDance, and Meta, claims that these three clients could spend $60-90 billion annually on custom chips within two years. Another custom chip company, Marvell, signed a five-year deal with Amazon for chip development. Iteration requires real-world scenarios; success hinges on large models willing to use the chips. Google has its own large models, solving the problem internally, while Microsoft and Amazon have launched a "proxy war." Microsoft heavily invested in OpenAI, while Amazon poured resources into Anthropic. OpenAI accelerates Microsoft's AI product integration, while Anthropic claims it will use a computing cluster with hundreds of thousands of Amazon's self-developed chips to ensure the success of Amazon's chip R&D. Jensen Huang once said he doesn’t pick fights but isn’t afraid of them either. Watching clients turn into competitors and alternatives gaining momentum, NVIDIA isn’t sitting idle—it’s pushing into the "client’s territory" by expanding into cloud services. Huang stated that NVIDIA could aggressively develop its own cloud services and AI software business (DGX Cloud and NVIDIA AI Enterprise) to encroach on cloud service providers' turf, aiming for a future scale of $150 billion. Meanwhile, industry leader Amazon’s cloud revenue in FY2024 approached $100 billion. Huang’s ambitions are big, but it’s also somewhat a forced move. Rather than letting clients steal his home, he’d rather steal theirs. 02 Jensen Huang could be called obsessive, and so could Elon Musk. In the automotive chip space, NVIDIA and Qualcomm battled, pushing Intel’s Mobileye out of the picture, while Tesla built a wall with its self-developed chips and FSD, preserving its "pure land." Since 2022, Musk has tried to shed the "car manufacturer" label, replacing it with "artificial intelligence." Although Tesla’s latest earnings report showed its first sales decline in over a decade, as a company Wall Street calls "an undervalued AI play," its tech prospects drove Tesla’s stock to rebound from a 6% drop to an 8% gain. Tesla’s AI ecosystem includes self-developed chips, FSD, the Dojo supercomputer, the Cortex supercomputing cluster, and the humanoid robot Optimus. But Musk’s AI ambitions go further. Frustrated by OpenAI’s deviation from his original vision as an early investor, he not only sued OpenAI and sparred on Twitter but also founded X.AI, with NVIDIA as a strategic investor. When Musk tried training large models with Dojo, the results were poor. So, he had to turn to NVIDIA. In September 2024, Musk announced the launch of Colossus, a super AI training cluster built with 100,000 NVIDIA H100 chips. He personally contacted Jensen Huang to secure another batch of NVIDIA products, hoping massive computing power would help X.AI compete with Google and OpenAI. Musk and Huang also intersect in humanoid robotics. If Tesla is the "godfather" of humanoid robots, NVIDIA is poised to become the "godmother." NVIDIA plans to launch Jetson Thor, its next-gen compact computer for humanoid robots, in H1 2025, creating a "full-stack" solution: AI model training software (DGX systems), simulation environments (Isaac Lab), and chip hardware (Jetson series). Beyond selling shovels, NVIDIA, alongside Microsoft and OpenAI, invested in humanoid robotics company Figure AI. This mirrors NVIDIA’s investment in cloud service firm CoreWeave: provide money and ammunition to help conquer the market. The benefit of this approach is staying focused on core competencies. Even someone as energetic as Musk faces "succession struggles" across multiple fiercely competitive fields, making quick, decisive counterattacks difficult. 03 Though Microsoft and OpenAI often appear as a pair, their relationship has grown somewhat distant. According to The Information, the two sides began new negotiations in October, focusing on Microsoft’s equity stake, its exclusive cloud provider status for OpenAI, and whether Microsoft would continue taking 20% of OpenAI’s revenue. Given OpenAI’s push to become a for-profit business, tensions with Microsoft are a classic storyline: Microsoft is a superpower, but OpenAI is a rising force that naturally seeks greater autonomy once it gains strength. Exclusive cloud services and a 20% revenue cut could hinder OpenAI’s ambition to change the world. Moreover, as large model tech advances rapidly, Microsoft’s lead via OpenAI seems less certain. Microsoft has since partnered with Meta, Mistral, and even OpenAI’s rival Anthropic in early 2025, planning to offer their models on Azure. Sam Altman initially praised the Microsoft-OpenAI partnership as rock-solid. But The Information reported that in December 2024, Altman admitted, "I won’t pretend there aren’t disagreements or challenges between us and Microsoft; obviously, there are." In the adult world, there’s no eternal friendship—only negotiations and renegotiations over interests. Microsoft grew under IBM’s shadow before surpassing it; it’s unlikely to let OpenAI replicate that path. 04 U.S. antitrust actions against tech giants haven’t stopped, but actual breakups are rare. Beyond political lobbying, regulators likely recognize that as low-hanging fruit disappears, tech’s deep-water phase demands massive funding and expertise for breakthroughs. Training OpenAI’s models costs nearly $100 million per run; a 5nm chip tape-out runs $30-40 million. Small firms can innovate in applications, but infrastructure—hardware, architectures, large models—requires scale to compete. Thus, U.S. tech giants both compete and collaborate. Whether skeptics call AI a bubble or not, these companies are charging ahead like: ### Related Stocks - [MSFT.US](https://longbridge.com/en/quote/MSFT.US.md) - [NVDY.US](https://longbridge.com/en/quote/NVDY.US.md) - [NVDL.US](https://longbridge.com/en/quote/NVDL.US.md) - [NVDA.US](https://longbridge.com/en/quote/NVDA.US.md)