SSPAI Morning Brief: OpenAI Partners with Pentagon as NVIDIA Prepares AI Inference Chips

Morning Brief

OpenAI replaces Anthropic to partner with the Pentagon
Paramount set to acquire Warner Bros., Netflix out
South Korea lifts data restrictions on Google Maps
Ultrahuman launches Ring Pro
NVIDIA to introduce AI inference-focused chips
Rising memory costs may eliminate entry-level PCs
Rumors Worth a Quick Look

OpenAI replaces Anthropic to partner with the Pentagon

According to reports from The Wall Street Journal and other outlets, on February 27, the Trump administration announced a comprehensive ban on federal agencies using technology from AI startup Anthropic, designating it as a “supply chain risk.” At the same time, the U.S. Department of Defense reached an agreement with Anthropic’s competitor OpenAI, allowing its AI models to be deployed in classified military environments.

Previously, the Pentagon had requested absolute usage rights for the Claude model within all legal boundaries, while Anthropic insisted on maintaining red lines prohibiting use in domestic mass surveillance and autonomous weapons. After negotiations broke down, Trump accused Anthropic of being a “radical left-wing” company on social media and ordered all federal agencies to fully discontinue use of its models within six months. The U.S. Department of Defense subsequently placed the company on a “supply chain risk” list typically reserved for foreign adversarial firms. This move not only cut off approximately $200 million in government contracts but may also compel other U.S. defense contractors to divest from the technology. Anthropic has stated it will challenge the decision in court.

Following Anthropic’s exit, OpenAI quickly filled the gap in military contracts. Sam Altman stated that the new agreement with the military includes safety provisions that similarly prohibit use for mass surveillance or autonomous weapons. OpenAI published the specific wording of these clauses on its website, noting that it is unclear why Anthropic was unable to reach a similarly secure agreement.

OpenAI emphasized that the partnered models remain cloud-hosted and therefore retain autonomous controllability — a solution that Anthropic had previously evaluated and explicitly rejected during negotiations, arguing that modern military systems are already highly networked, rendering any distinction between cloud and weapon endpoints largely meaningless. Additionally, OpenAI’s terms prohibit AI from directly controlling weapons only in scenarios where laws or departmental policies require human oversight. However, current Department of Defense policy relies on vague language such as requiring an “appropriate level of human judgment,” which could potentially be modified to bypass such restrictions. Safeguards related to mass surveillance largely restate existing applicable laws and do not prohibit governments from obtaining surveillance data indirectly or through third parties, whereas Anthropic had sought to explicitly bar the government from using AI to analyze “non-classified bulk commercial data.”

Media reports indicate that OpenAI executives donated tens of millions of dollars last year to political action committees supporting Trump, while Anthropic’s CEO reportedly had personal conflicts with the Department of Defense’s CTO. By the day negotiations concluded, Trump had already prepared social media posts criticizing Anthropic. While Anthropic representatives were still attempting to continue talks, a replacement agreement between the Department of Defense and OpenAI was already in place.

Founded in 2021 by several core researchers who left OpenAI due to concerns over AI safety, Anthropic is currently among the most highly valued AI companies in the world. The company has maintained a strained relationship with the current U.S. government, having drawn criticism for hiring former Biden administration officials, advocating for stronger AI regulation, and opposing restrictions on AI chip exports to China. In January this year, the U.S. military reportedly used the Claude model in operations in Venezuela. Anthropic subsequently inquired with partners about the model’s usage details, which the Pentagon viewed as a challenge to military decision-making authority — an incident that ultimately contributed to the breakdown in relations. Just hours before negotiations collapsed, Claude was reportedly used in a U.S. airstrike operation in the Middle East.

The incident has sparked a wave of U.S. users uninstalling OpenAI’s ChatGPT in favor of Claude. By Saturday, Claude had risen to the top of the U.S. Apple App Store’s free app rankings, surpassing ChatGPT.

Paramount set to acquire Warner Bros., Netflix out

According to The Wall Street Journal, on the evening of February 26, after six months of intense bidding, Paramount — led by David Ellison — successfully acquired Warner Bros. Discovery in a deal valued at up to $81 billion. Streaming giant Netflix officially withdrew from the bidding, abandoning its previously agreed $72 billion acquisition proposal with Warner.

David Ellison is the son of Oracle founder Larry Ellison. Prior to this, he had secured control of Paramount through his company Skydance Media. Following the acquisition of Warner, Ellison will preside over an unprecedented media empire, bringing HBO, CNN, CBS News, Warner Bros. Pictures, as well as major IP such as DC Comics and Harry Potter, under his control.

The bidding battle proved exceptionally complex, with Ellison submitting nine separate offers over the course of six months. Warner initially rejected multiple proposals and even announced in December last year that it would accept Netflix’s acquisition plan. However, Ellison refused to back down, raising his bid to $31 per share — a 63% increase over his initial offer — while pledging to cover Warner’s $2.8 billion termination fee to Netflix. He also threatened to challenge control at the shareholder level and engaged in extensive political lobbying in Washington, ultimately persuading Warner’s board to reverse its decision.

The news triggered a strong reaction in the capital markets, with Paramount’s stock surging more than 20% on Friday, while Netflix shares — having exited the bidding war — also rose nearly 14%. The transaction is expected to take between 6 to 18 months to close. The merged Paramount is projected to carry approximately $100 billion in debt, with management facing the challenge of cutting $6 billion in costs.

South Korea lifts data restrictions on Google Maps

According to The New York Times, on February 27, South Korea’s Ministry of Land, Infrastructure and Transport approved Google’s request to export high-precision map data from the country. This decision overturns longstanding regulatory restrictions and clears the way for Google Maps to offer full functionality in South Korea, including real-time navigation.

For years, citing national security concerns, the South Korean government strictly limited the overseas transfer of mapping data. As a result, Google Maps has been unable to provide real-time navigation, walking directions, or detailed business listings within the country, forcing both residents and foreign visitors to rely on domestic navigation apps. Under the newly approved conditions, Google will be permitted to transfer government-reviewed map data to its global data centers, provided that sensitive security information such as military base locations is strictly excluded. Google executives welcomed the decision but have yet to disclose a timeline for the rollout of full navigation features.

Since introducing its map services in South Korea in 2008, Google has faced ongoing limitations due to these policies, and first publicly requested permission to export mapping data in 2016. According to the Ministry, the policy shift is intended to “boost tourism and stimulate economic growth,” though it has also been influenced by U.S.–South Korea trade negotiations. Previously, the Trump administration had reportedly pressured Seoul to cease regulatory practices that disadvantaged U.S. technology firms in areas such as location data.

The full deployment of Google Maps is expected to directly challenge the dominance of domestic players such as Naver and Kakao in South Korea’s digital mapping sector. The country’s spatial information industry has expressed concern that the entry of a global platform could ultimately reduce local firms to subcontractors for Google. However, some tourism and business stakeholders remain optimistic, believing that the move will help local merchants expand into international markets while encouraging domestic mapping services to improve through healthy competition.

Ultrahuman launches Ring Pro

According to The Verge, on February 27, health tech brand Ultrahuman unveiled its third-generation flagship smart ring, Ring Pro, along with the AI health platform Jade. Ring Pro is available in four color options and is priced at $479. It is currently open for global pre-orders and will begin shipping in March. However, due to an ongoing patent dispute with leading smart ring brand Oura, the new device is not yet available in the U.S. market.

In terms of hardware, Ring Pro’s battery life has been extended from the previous 4 to 6 days to a maximum of 15 days on a single charge, and it now supports offline storage of up to 250 days of health data. The device also features a faster dual-core processor, an upgraded heart rate sensor, and safety-focused physical design improvements to ensure it can be more easily removed from the finger in emergency situations.

Also introduced is a new Pro charging case, capable of providing an additional 45 days of battery life while storing up to one year of device data. The charging case supports wireless charging, includes an LED battery indicator, and comes equipped with a built-in speaker and proximity tracker.

The Jade bio-intelligent AI platform is currently capable of delivering real-time health insights, guiding breathing exercises, and detecting atrial fibrillation (AFib). Ultrahuman plans to expand its capabilities through future updates, enabling proactive health risk alerts, smart home temperature adjustments, and even online food ordering — with the goal of transforming Jade into an always-on “autonomous health agent.”

NVIDIA to introduce AI inference-focused chips

According to The Wall Street Journal, NVIDIA plans to unveil a new chip specifically designed for AI inference — the process by which models respond to user prompts — at its upcoming GTC Developer Conference in San Jose next month. The product integrates chip designs from startup Groq, enabling more efficient handling of complex AI requests. OpenAI, the developer of ChatGPT, has agreed to become one of the largest customers for the new chip.

As enterprises increasingly deploy AI agents capable of autonomously executing tasks and automated programming tools, demand for inference computing power has surged. Although NVIDIA currently holds over 90% of the global GPU market share with its Hopper and Blackwell series, traditional GPUs remain costly and energy-intensive when handling real-world inference workloads.

To address competition from in-house chips developed by Google and Amazon, as well as startups such as Cerebras, NVIDIA invested heavily late last year to bring Groq’s technology and core team under its umbrella. Groq utilizes a distinct architecture known as a Language Processing Unit (LPU), which delivers exceptional efficiency during the token-by-token decoding phase of AI model response generation. OpenAI is reportedly planning to leverage NVIDIA’s new system to enhance its automated programming tool Codex in response to competing solutions such as Claude Code, which primarily run on Amazon and Google cloud chips.

In an effort to retain key clients, NVIDIA recently invested $30 billion in OpenAI and secured substantial orders for dedicated inference compute capacity, offsetting OpenAI’s recent procurement of chips from Amazon and Cerebras. Additionally, NVIDIA expanded its partnership with Meta this month, deploying large-scale server clusters powered exclusively by its self-developed Vera CPUs for the first time, supporting Meta’s AI agents for advertising-related computational workloads.

Rising memory costs may eliminate entry-level PCs

According to The Register, citing forecasts from research firm Gartner, surging memory demand driven by the AI boom is creating a severe cost crisis for the global PC and smartphone markets. As DRAM and NAND flash prices continue to climb, global PC shipments may decline by more than 10% this year, while smartphone shipments are expected to drop by around 8%. Entry-level devices priced below $500 may disappear from the market entirely.

Unlike previous price hikes caused by production cutbacks, the current surge is largely driven by massive demand from hyperscale cloud providers for AI computing infrastructure, directly pushing up component costs across the consumer electronics sector. Some memory prices have already multiplied several times since last year, and Gartner predicts that the cost of memory and storage used in personal devices could rise by a further 130% by the end of 2026. Under this pressure, device manufacturers are finding it increasingly difficult to maintain profit margins on budget PCs, while raising prices risks alienating price-sensitive consumers.

The adoption of AI PCs is also being hindered by elevated memory costs. For instance, Microsoft’s Copilot+ platform requires a minimum of 16GB of memory, significantly raising manufacturing thresholds. HP recently revealed that DRAM now accounts for 35% of total PC bill-of-materials costs. As a result, AI PCs are likely to remain positioned in the premium segment in the near term, with market share not expected to exceed 50% until 2028. The smartphone sector faces similar challenges, with low-end models being hit particularly hard due to the higher proportion of storage components in their overall cost structure.

As the cost of new devices rises across the board, both enterprises and consumers will be forced to extend the lifespan of existing hardware. Replacement cycles for commercial and consumer PCs are projected to lengthen by 15% and 20%, respectively. Analysts warn that this demand-driven memory shortage may persist for the long term, potentially lasting through the end of 2027.

Rumors Worth a Quick Look

On February 28, Google announced that it will automatically restore developer accounts for Gemini CLI and Antigravity that were recently suspended for violating service terms, and has introduced a self-service account recovery process. Affected accounts are expected to regain access within one to two days. Going forward, Google will no longer impose unannounced bans. Users flagged for violations will receive email notifications with clear error messages and a dedicated appeal form link. To restore access, users must reaffirm compliance with the terms of service, including a commitment not to use third-party tools to bypass system restrictions. Once submitted, accounts will be automatically reinstated through periodic synchronization. However, a second violation will result in permanent suspension. Google emphasized that any attempt to use third-party software, tools, or services (such as OpenClaw) to “harvest” or “piggyback” Gemini CLI’s OAuth authentication in order to access Google backend services or usage quotas constitutes a serious violation.
On February 26, Restaurant Brands International (RBI), the parent company of Burger King, announced plans to roll out an AI assistant named Patty across its U.S. locations. The system is designed to support employees and monitor customer interactions in order to assess overall team friendliness. Addressing concerns over workplace surveillance, Burger King stated that the AI is not intended to monitor or evaluate individual staff, nor will it require employees to follow scripted responses. Instead, it aggregates the frequency of keywords such as “welcome,” “please,” and “thank you” to help managers understand overall service patterns. Previously, brands such as McDonald’s and Taco Bell experimented with AI-powered drive-through ordering systems, but these were discontinued due to frequent errors.
According to The Register, Microsoft’s HoloLens has found a new role within the U.S. military. Initially tested in 2018 as battlefield equipment, the device was shelved after causing severe physical discomfort among soldiers, including headaches, visual fatigue, and nausea. It is now being repurposed by the U.S. Army and Air Force to assist with remote aerial cargo inspections, with early success reported. In practice, Army personnel responsible for loading supplies wear the HoloLens headset, while Air Force experts connect remotely via laptops and Wi-Fi hotspots. Using a first-person perspective and AR visual prompts, Air Force staff guide soldiers in adjusting rigging and repositioning cargo to ensure proper weight distribution for safe airdrops. According to military officials, the system has reached a “plug-and-play” level of readiness after a year of refinement. Microsoft ended further development of HoloLens in late 2024, with official support for existing models set to conclude by the end of 2027. It remains unclear whether the headsets used in these trials are drawn from military inventory or newly procured.
On March 1, Xingji Meizu Group’s China CMO Wan Zhiqiang stated during a livestream on Meizu’s JD flagship store that existing Meizu models will continue to be sold at current prices, though some are already sold out or nearing depletion. The Meizu 23 has completed development but will not be released commercially. Future updates to the Flyme operating system will be limited to essential maintenance, with no major version upgrades planned. Previously, on February 27, Meizu Technology announced a strategic transformation, stating that it will suspend domestic self-developed hardware projects for new smartphone products.
Mark Gurman reports that —
Apple plans to release a touchscreen MacBook Pro by the end of 2026, while maintaining a clear distinction between Mac and iPad product lines. The new MacBook Pro will retain its traditional design and macOS interface, with touch input offered as an optional feature rather than a primary interaction method. A large foldable iPad is also reportedly in development for release later in the decade, and will continue to run iPadOS. A redesigned MacBook Air is not expected until around 2028.
At this year’s WWDC, Apple is also expected to rebrand the Core ML framework as Core AI.

SP-Life