OpenAI Distances Itself From Microsoft, Goes All-In on Amazon

TECH BRIEF

Hello and welcome to Front Research, here's what's happening in tech today.

• OpenAI's revenue chief sends a leaked memo saying Microsoft has "limited our ability" to reach customers, openly tilting the company toward Amazon

• Stanford's 2026 AI Index puts China within 2.7% of the US on top model performance, the gap is effectively gone

• SoftBank, Sony, Honda, and NEC launch a trillion-parameter "physical AI" foundation model company backed by ~$6.3B of Japanese government support

• Treasury Secretary Bessent and Fed Chair Powell push the largest US banks to test Anthropic's Mythos model after it found thousands of zero-day flaws

• Nvidia denies it is buying a major PC maker, sending Dell and HP shares whipsawing after a 5-7% intraday surge

Let's jump in.

OpenAI Distances Itself From Microsoft, Goes All-In on Amazon

OpenAI's newly appointed revenue chief, Denise Dresser, sent an internal memo to staffers on Sunday that openly criticized the company's flagship partnership with Microsoft and positioned Amazon as the new growth engine for its enterprise business. In the memo, obtained by CNBC, Dresser wrote that the Microsoft tie-up "has also limited our ability to meet enterprises where they are, for many that's Bedrock," referring to Amazon's enterprise AI platform. (CNBC)

Dresser said inbound demand since OpenAI announced its Amazon partnership in late February has been "frankly staggering." That partnership includes a planned investment of up to $50 billion from Amazon and gives AWS customers native access to OpenAI models on Bedrock. The memo also went after Anthropic, claiming its strategy is built on "fear, restriction, and the idea that a small group of elites should control AI," while OpenAI's "positive message" will win out over time. (Axios, Yahoo Finance)

The memo is the bluntest public signal yet that the OpenAI-Microsoft relationship is fraying. Microsoft has been OpenAI's exclusive cloud partner since 2019, and that exclusivity was loosened earlier this year as part of OpenAI's restructuring. Dresser's framing makes clear OpenAI now views Microsoft as a constraint rather than a partner, and the company is actively steering enterprise pipeline toward AWS.

Why it matters: This is a reordering of the cloud and AI market. Microsoft has built much of its enterprise AI thesis on OpenAI exclusivity, and Azure's premium valuation depends on it. If OpenAI is now actively routing its biggest enterprise deals through Bedrock, the AWS narrative strengthens materially while the Azure-OpenAI moat narrows. Watch how Microsoft responds, the company is rumored to be accelerating its own in-house model strategy.

Stanford AI Index 2026: The US-China Model Gap Has Effectively Closed

Stanford HAI released its 2026 AI Index Report on April 13, and the headline finding is that China has erased almost the entire performance gap with the US on frontier AI models. Anthropic's top model now leads the closest Chinese model by just 2.7% as of March 2026, and on Arena's community-driven leaderboard, US and Chinese models have traded the top spot multiple times since early 2025. (Stanford HAI, SiliconANGLE)

The report frames the narrative as a clean reversal from the dominant 2024 view, where the US lead was thought to be measured in years. DeepSeek-R1 briefly matched the top US model in February 2025, and the gap has been razor-thin ever since. The US still leads on capital and infrastructure (private AI investment hit $285.9B in 2025, more than 23x China's $12.4B), and produces more top-tier models and high-impact patents. China leads on publication volume, citations, patent output, and industrial robot installations. (MIT Technology Review, IEEE Spectrum)

One striking secondary finding: the number of AI researchers and developers flowing into the US has dropped 89% since 2017, with 80% of that decline happening in the last year alone. Stanford flags this as a long-tail risk to US leadership even if capital advantages persist. (Implicator AI)

Why it matters: This reframes the whole policy debate around AI export controls and chip restrictions. If China is keeping pace on model quality with a fraction of the capital and constrained access to top-end Nvidia chips, the case for export controls as a competitive moat weakens. It also pressures US labs to sustain their lead through deployment and distribution rather than raw model performance.

Japan Forms a Trillion-Parameter "Physical AI" Consortium

SoftBank, NEC, Honda, and Sony jointly established a new company on April 12 called Japan AI Foundation Model Development, with the goal of building a trillion-parameter foundation model trained on Japanese data and tuned for real-world control tasks like robotics and factory systems. Japan's New Energy and Industrial Technology Development Organisation has earmarked roughly ¥1 trillion (about $6.3B) in AI support over five years starting fiscal 2026, and the new firm is expected to apply. (Japan Times, SiliconANGLE)

The investor base is notable for its breadth: Nippon Steel, Kobe Steel, MUFG, Sumitomo Mitsui Banking, and Mizuho are all participating, alongside the four core tech companies. SoftBank and NEC will lead AI development, Honda will be the first deployer in autonomous vehicles, and Sony brings robotics and gaming hardware. The consortium is targeting "physical AI applications" by 2030, with the model designed to perceive and act in the physical world rather than purely generate text or images. (TechWire Asia, Decrypt)

The framing is explicitly sovereign: Japan wants a model it owns, trained on Japanese data, deployed in Japanese factories, without routing through US or Chinese cloud platforms.

Why it matters: This is the most coordinated sovereign AI push outside the US-China duopoly to date. It validates the "physical AI" thesis (robotics, embodied agents) as a distinct frontier from generative AI, where Japan has structural advantages in industrial automation, cars, and consumer hardware. For Nvidia, Honda, and Sony shareholders, it is a clear signal that automotive and industrial AI is becoming a distinct layer of the stack. For the EU, it is a template for how to organize a national AI champion.

Treasury and Fed Push Banks to Test Anthropic's Mythos for Cyber Vulnerabilities

Bloomberg reported on April 13 that Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell summoned executives from JPMorgan, Goldman Sachs, Citi, Bank of America, and Morgan Stanley to a meeting urging them to test Anthropic's new Mythos model for finding cybersecurity vulnerabilities. All five banks are now reportedly running tests. (Bloomberg, PYMNTS)

Mythos has uncovered thousands of zero-day flaws across major operating systems and browsers in early testing, despite not being trained specifically for cybersecurity. Anthropic is distributing it through a restricted program called Project Glasswing to roughly 50 organizations. The concern from regulators is that if defenders can use Mythos to find vulnerabilities, attackers will be able to use equivalent capabilities at the same speed, dramatically compressing the patch window for critical financial infrastructure. (NBC News, Anthropic)

The political subtext is striking. Even as the Pentagon is in court fighting Anthropic over its refusal to remove safety guardrails for autonomous weapons and surveillance, Treasury and the Fed are now actively recommending the same company's models to the largest banks in the country. (Fortune)

Why it matters: This is the first time US financial regulators have formally pushed a specific AI vendor onto the systemic banks, and it signals that the Fed views AI-driven cyber risk as a near-term financial stability issue. For Anthropic, it is a major enterprise validation that lands directly inside its competitive fight with OpenAI. For the broader cybersecurity sector (CrowdStrike, Palo Alto, SentinelOne), an AI model that can find zero-days at scale changes the economics of vulnerability management.

Nvidia Denies It Is Buying a PC Maker, Dell and HP Whipsaw

Dell and HP shares jumped sharply on April 13 after a SemiAccurate report claimed Nvidia had been in negotiations for over a year to acquire a major PC-oriented firm. Dell closed up 6.7% and HP rose 5.3% on the day. Nvidia issued a denial in an emailed statement: "The media report is false; NVIDIA is not engaged in discussions to acquire any PC maker." Both stocks dropped more than 3% in post-market trading after the denial. (Yahoo Finance, The Tech Portal)

The episode underscores the ongoing market obsession with how Nvidia, now valued at roughly $4.3 trillion, will deploy its cash and strategic optionality. Even an unverified rumor moved two large-cap PC names by mid-single-digit percentages within hours, and forced an unusual public denial from Nvidia within the same trading session. (FinancialContent)

Why it matters: The market is pricing in the assumption that Nvidia will eventually buy its way down the stack, into PCs, networking, or systems integration. The speed and magnitude of the Dell and HP move shows how thin the float of credible AI hardware acquisition targets has become. For investors, it is a useful reminder that Nvidia's strategic choices now move adjacent multi-billion-dollar names on rumor alone.