Samsung Workers Forge Profit-Linked Pay.

Mazen Al Sabbagh

— 6 min read
Share

TL;DR

  • Samsung's workers ratified a landmark compensation structure with 73.7% approval, averting a significant strike. This agreement introduces profit-linked bonuses for semiconductor employees, redefining South Korean corporate labor relations.
  • Anthropic is projected to outpace OpenAI in revenue by at least 35%, according to The Information. Their annualized run rate nears $50 billion, indicating a strategic shift in market leadership and enterprise adoption.
  • Fireworks AI is pursuing a $15 billion valuation, nearly quadrupling its worth in seven months. This reflects surging investor confidence in inference-focused startups, recognizing the strategic importance of efficient model deployment.
  • OpenRouter secured $113 million in Series B funding at a $1.3 billion valuation, led by Alphabet's CapitalG. The investment validates multi-model routing as a distinct and essential layer within the AI infrastructure stack.
  • The Pentagon requested $29.5 billion for fiscal 2027 to establish AI supercomputing infrastructure via its "AI Arsenal initiative," signaling a strategic push for government-owned, secure AI capabilities.

Lead Story: Samsung Workers Forge Profit-Linked Pay

Samsung Electronics' tentative 2026 wage agreement was approved by its union Wednesday morning in Suwon. With 73.7% of 62,616 participating members voting in favor, at a 95.5% turnout, the immediate threat of an 18-day general strike was neutralized following a last-minute intervention by South Korea's labor minister.

The terms of this agreement are notable by Korean corporate standards. Semiconductor division employees are set to receive cash bonuses of up to 50% of their base pay, alongside a special stock-based performance bonus equivalent to 10.5% of annual operating profit. This is contingent on the chip business achieving 200 trillion won ($132 billion) in operating profit from 2026 to 2028, and 100 trillion won from 2029 to 2035. For top memory engineers, this translates to potential payouts of up to $400,000.

Reuters described the deal as "seismic" for South Korea, establishing a precedent where the world's largest memory chipmaker directly links worker compensation to division-level profitability on this scale. The Korea Herald confirmed that management and the joint bargaining group formally signed the agreement Wednesday afternoon. This structure unveils fundamental tensions within traditional compensation models.

The victory, however, is not without internal divisions. A significant compensation disparity persists between chip and non-chip workers: Device Experience division employees receive approximately 6 million won in company shares, a stark contrast to the semiconductor team's potential hundreds of millions. The Donghaeng Labor Union, representing non-chip workers, withdrew from joint bargaining entirely and filed a court injunction, which was rejected Monday. Union chairman Choi Seung-ho, who had pledged to resign if the deal failed, ultimately signed it into force. The structural tension within Samsung, driven by differential profitability from AI memory demand, has been formalized rather than resolved.

In Other News

Anthropic is likely outperforming OpenAI in revenue generation. The Information reported Tuesday night that Anthropic's revenue now surpasses OpenAI's by at least 35%, a notable reversal from the beginning of the year. While OpenAI recorded $5.7 billion in Q1 revenue, approximately $1 billion more than Anthropic in the same period, Anthropic subsequently informed investors that its Q2 revenue would reach $10.9 billion. This projection indicates more than a doubling in a single quarter, with an annualized run rate set to exceed $50 billion by June's end. OpenAI's growth, though strong at over 50% for the first five months of 2026, has not matched this acceleration. Anthropic's ongoing $30 billion-plus funding round, expected to close this week at a valuation above $900 billion, is poised to coincide with the company establishing both its safety credentials and a clear revenue lead.

Fireworks AI seeks a $15 billion valuation, quadrupling in seven months. Bloomberg reported Tuesday that the Redwood City inference startup is in discussions for a new funding round, co-led by Index Ventures. This represents a significant increase from its $4 billion Series C valuation in October 2025. Founded in 2022, Fireworks AI specializes in helping enterprises deploy and scale AI models, prioritizing speed and cost optimization for open-source large language models. The rapid ascent from $552 million (Series B, July 2024) to $4 billion, and now a potential $15 billion, underscores the growing strategic importance of the inference layer – the compute-intensive process of running AI models – as training becomes increasingly commoditized. Key backers include Sequoia, NVIDIA, and AMD.

OpenRouter secured $113 million and surpassed a $1.3 billion valuation. The AI model marketplace announced its Series B funding Tuesday. The round was led by Alphabet's CapitalG, with participation from NVentures (NVIDIA), ServiceNow Ventures, MongoDB Ventures, Snowflake Ventures, Databricks Ventures, and existing investors Andreessen Horowitz and Menlo Ventures. OpenRouter currently processes 25 trillion tokens per week, a substantial increase from 5 trillion just six months ago, across over 400 models from providers like Anthropic, Google, OpenAI, xAI, and DeepSeek. The New York Times highlighted Uber's CTO noting that the company had expended its entire 2026 AI budget months into the year, pushing enterprises toward multi-model strategies and open-source alternatives for cost efficiency. As CEO Alex Atallah stated, "Running inference at scale is fundamentally a multi-model problem."

X / Social Pulse

The Samsung vote garnered measured responses from labor analysts, who pointed to the DX division's withdrawal and injunction attempt as indicators of unresolved structural inequality within the conglomerate. The Information's report on Anthropic's revenue leadership generated more incisive reactions, with observers noting the perceived irony that the company positioned as a "safety lab" is simultaneously becoming a revenue leader, challenging the narrative that safety commitments inherently impede competitive advantage. Fireworks AI's $15 billion valuation prompted critical inquiry into the sustainability of inference-layer startup margins against hyperscalers who control their own silicon. Separately, the Starlette "BadHost" vulnerability disclosure (detailed in Quick Hits) raised significant concerns within the agent-building community regarding the inherent fragility of the open-source dependencies underpinning millions of AI agents.

One to Watch

The Pentagon targets $29.5 billion for AI supercomputing under an "Arsenal" initiative. The Department of Defense's fiscal 2027 budget request includes a new "AI Arsenal initiative," designed to construct secure data centers, procure next-generation GPUs, and centralize supercomputing assets across military operations. This substantial spending is allocated within the reconciliation portion of the $1.45 trillion defense proposal. A Pentagon official informed DefenseScoop that this initiative represents "an investment in foundational, government-owned AI infrastructure to maximize federal buying power." This development occurs while the Pentagon's relationship with at least one major AI lab, Anthropic, whose $200 million contract was rescinded, remains adversarial. Concurrently, a separate $54.6 billion request for the Defense Autonomous Warfare Group has drawn Senate warnings that doctrine may be lagging behind expenditure.

Quick Hits

  • A critical "BadHost" vulnerability in Starlette, an open-source package with 325 million weekly downloads, threatens millions of AI agents reliant on its web serving infrastructure.
  • Trump appointed former Florida AG Pam Bondi to PCAST, the AI advisory panel co-chaired by David Sacks and White House science adviser Michael Kratsios.
  • ByteDance is extending special stock packages to select AI staff to counter poaching efforts from rival labs, illustrating the escalating compensation costs in the AI talent market.
  • Demis Hassabis suggested AGI could emerge by 2029, a tighter timeframe than his previous estimates, in comments reported this week—aligning Google DeepMind's internal forecast with Anthropic's Jack Clark.
  • Anthropic's Mythos model identified over 10,000 critical vulnerabilities across widely utilized open-source software, expanding the Project Glasswing security initiative launched last week.

The pattern this week illuminates the growing disparity in value capture within the AI ecosystem. Samsung's chip workers secured substantial bonuses, while their counterparts in consumer electronics received a fraction. Anthropic surpassed OpenAI in revenue by focusing on enterprise deals over consumer subsidies. Fireworks AI nearly quadrupled its valuation by betting on efficient inference as a core business driver, while OpenRouter reached $1.3 billion by making the underlying model itself an interchangeable commodity. Concurrently, the Pentagon requested $29.5 billion to establish its own AI infrastructure, rather than depend on commercial labs it views with increasing skepticism. The winners in AI are increasingly not those who build the largest models, but those who control the strategic points of leverage around them—the chips, the inference capabilities, the routing layers, and the direct contracts. The Samsung vote, in particular, crystallized this: even within a single corporation, the question of who profits from AI-driven demand and who does not has become too significant to be mitigated by a single agreement.

Sources

Lock in. M. mazen@thorterminal.com

Read more