文章以 Elon Musk 与 Sam Altman 的诉讼为线索，回溯 OpenAI 由「非营利」走向「营利」的关键驱动：竞赛逻辑与资源规模。1945 年 Robert Oppenheimer 提到「敌方先做出武器」的忧虑，约 70 年后（即 2015 年）Altman 在寄给 Musk 的电邮中以「AI 的 Manhattan Project」类比，主张既然先进 AI「几乎肯定」无法被阻止出现，且最可能由已于前一年收购 DeepMind 的 Google 做出，那么「由 Google 以外的人先做出来」更可取，显示道德诉求与先发制人动机交织。
诉讼核心在于组织型态与资金流向：Musk 指控 Altman 与 Greg Brockman 诱使其提供 3,800 万美元（$38mn）慈善捐款，却意图日后将 OpenAI 转为营利以「套现」；被告否认并反控 Musk 亦寻求从 OpenAI 获利，未果后以诉讼打击竞争者（包括其 AI 公司 xAI）。文章指出 2017 年后竞争动态改变：Google 在把大量 GPU 晶片整合为单一系统上取得突破，使「单次大型实验」可在更短时间内完成；Ilya Sutskever 直言过去多晶片仅利于「多个并行小实验」而非「加速单一大实验」，因此「学术实验室」或纯非营利体制原可竞争的前提消失，资本重要性大幅上升。
至 2018 年初，Musk 判定必须改采营利模式以取得投资，甚至提议由 Tesla 并购，称其为唯一能与 Google 竞逐的路径；董事会反对后他切断正式关系，但非营利架构已进入淘汰。经多次重组（最近一次在去年 10 月），OpenAI 营运现置于估值 8,520 亿美元（$852bn）的营利公司并据称准备上市；商业投资人（Nvidia、Amazon、SoftBank、Microsoft 等）持股居多，其中 Microsoft 为最大股东，而原非营利实体仍持有约五分之一（约 20%）股份并依法可任免董事。惟 2023 年末董事会曾以「沟通不够坦诚」解任 Altman，引发 Microsoft 与员工反弹、几乎整个董事会被替换且 Altman 回任，显示治理权力的实际约束有限；此外 2019 年为安抚疑虑而设的投资报酬上限，亦于去年 10 月改制时被取消，使股东面临「无上限报酬」的激励，与 Brockman 私下提及「让我到 10 亿美元（$1bn）」的文件一起，构成所谓使命漂移的量化线索。
The article frames the Musk–Altman lawsuit as a post-mortem on OpenAI’s founding promise that frontier AI could be built by a non-profit rather than a return-seeking corporation. Echoing Robert Oppenheimer’s 1945 logic that speed matters because an adversary might get there first, Sam Altman wrote to Elon Musk in 2015—roughly 70 years later—arguing that advanced AI was “almost definitely” unstoppable and most likely to be created by Google, which had bought DeepMind the year before; therefore it would be better for someone other than Google to build it first.
The legal dispute centers on money, intent, and organizational design. Musk alleges Sam Altman and Greg Brockman induced him to provide $38mn in philanthropic funding while planning to “cash in” by converting OpenAI into a profit-making enterprise; the defendants deny this and argue Musk sought to profit himself and sued only after failing, also to harm a rival to his own AI company, xAI. A technical inflection point is placed in 2017, when Google’s ability to link a huge number of GPU chips into one system shifted progress toward fast, single large experiments; Ilya Sutskever said that earlier, many chips mainly enabled many small experiments in parallel, letting an “academic lab” compete, but that advantage faded once scaling a single run became decisive and capital-intensive.
By early 2018 Musk concluded OpenAI needed a for-profit model to raise investment, even proposing a Tesla takeover, but the board refused and Musk severed formal ties; nonetheless, restructuring proceeded. After multiple reorganizations, most recently last October, OpenAI’s operations sit in a for-profit company valued at $852bn, largely owned by commercial investors (including Nvidia, Amazon, SoftBank, and Microsoft as the largest shareholder), while the original non-profit retains roughly one-fifth (~20%) of the equity and formal board powers. Yet the late-2023 firing and rapid reinstatement of Altman after staff and Microsoft backlash highlighted weak practical constraints, and the removal last October of a 2019 cap on investor returns—alongside a disclosed note about a path to $1bn—underscored how governance and incentives drifted toward unlimited upside in the race for AGI.