The Big Sunday Politics Section: Trump, Trump, Trump

Anthropic Blacklisting (I): Trump's Pentagon Deal Divides the AI World

Sam Altman announced a deal with the Pentagon on Friday evening to use OpenAI models in classified military networks — with the same red lines for which Anthropic was blacklisted hours earlier. The deal prohibits domestic mass surveillance and requires human control over autonomous weapons, reports the New York Times. However, OpenAI has neither the necessary security certification nor the infrastructure in the Pentagon's military cloud systems. What Altman actually signed and when it will take effect remains unclear — the Wall Street Journal reported as recently as Thursday that there was no signed contract. Anthropic, whose Claude model already runs on classified systems for the CIA and NSA, was classified as a “Supply Chain Risk” by Defense Secretary Pete Hegseth after CEO Dario Amodei rejected the “for all lawful purposes” clause. OpenAI employees learned in an all-hands meeting that the dispute was personal: Amodei had “insulted” the Pentagon leadership with public blog posts. → Techpresso

Synthszr Take: Altman is signing a contract for infrastructure that doesn't exist, on systems to which OpenAI has no access. Anthropic has been running Claude on classified Pentagon networks for months — OpenAI would first have to go through security certifications that typically take years. Amazon's $50 billion investment could provide the technical foundation, but there's a world of difference between an announcement and operational reality. What's really happening here is the Pentagon demonstrating it can kick any AI provider out of the market at any time if the CEO doesn't fall in line. IT service providers should reconsider their dependencies on individual model providers — today's preferred partner could land on a blacklist via a tweet tomorrow.

Anthropic Blacklisting (II): Gary Marcus Suspects Deception by Sam Altman

Gary Marcus reveals a maneuver that seems remarkably cynical even by Silicon Valley standards: Sam Altman publicly supported Dario Amodei's Anthropic while secretly working on a deal since Wednesday that would take Anthropic's business away. The timeline speaks volumes: First, Greg Brockman donated $25 million to Trump's PAC, then came Altman's public declaration of solidarity, followed by Trump's denunciation of Anthropic as a 'supply chain security risk.' On the same day Altman declared his support, he signed a deal with similar conditions – the play was perfectly staged. Marcus sees this as the U.S. transitioning from a market economy to an oligarchy, where connections and donations determine business success. Anthropic was not just rejected but permanently banned, while OpenAI was awarded the contract with virtually identical terms. → Gary Marcus from Marcus on AI

Synthszr Take: A $25 million PAC donation as an entry ticket for government contracts – the new reality of American tech politics turns enterprise sales into a political show. Technical excellence alone no longer wins major strategic contracts when competitors have mastered the game of political influence. The incident reveals the Achilles' heel of the AI market: a few providers control critical infrastructure, and those excluded for political reasons have no alternative. For consulting firms, this means multi-vendor strategies are becoming mandatory, not optional. Banning Anthropic as a 'supply chain risk' is absurd – but it's precisely this arbitrariness that makes vendor lock-in with AI models an existential business risk. Sam Altman is playing the old game: it's not enough to win, the opponent must also lose.

US Government Meets with Robot Manufacturers: China in Focus

The U.S. Department of Commerce is inviting American robot manufacturers to a roundtable discussion on March 10. According to an invitation seen by Semafor, the goal is to identify 'key challenges in supply chains and policy for American robotics manufacturing.' Both industrial robot manufacturers and humanoid robot developers have been invited. A department spokesperson emphasized that they want to gather industry feedback on federal robotics policies — the discussion will not overlap with existing work on tariffs or import/export restrictions. Robotics is becoming a central stage in the U.S.-China technology competition, with Commerce Secretary Howard Lutnick having already met with robotics industry CEOs last year. American companies are calling for increased government support as Chinese manufacturers benefit from massive state subsidies. According to Politico, the Biden administration is considering an executive order on robotics later this year. → Semafor Technology

Synthszr Take: Washington understands robotics for what it is: the critical infrastructure of the 21st century. China isn't subsidizing its robot manufacturers for the love of innovation, but because autonomous manufacturing defines the next industrial revolution — whoever controls the robots determines global supply chains. For IT service providers, this means robotics integration is shifting from a nice-to-have to a mandatory program, as clients will soon have to choose between a 'China stack' and a 'US stack.' The real question isn't whether America will catch up, but whether European providers will have any role to play at all. Without its own robotics platforms, the EU faces complete dependence on two competing technospheres.

Trump Invites Big Tech to the White House: Energy Self-Sufficiency for AI Data Centers

Tech giants are being summoned to the White House in March to sign Trump's 'Rate Payer Protection Pledge' — committing to build their own power plants for their AI data centers. Amazon, Google, Meta, Microsoft, xAI, Oracle, and OpenAI are expected to secure their own energy supply so that American electricity customers don't have to pay for the AI industry's exploding energy demand. Trump justifies the initiative by citing an outdated power grid that cannot handle the enormous capacity requirements of modern data centers. The measure is positioned as protection against rising electricity prices while simultaneously securing American AI dominance over China. Energy Secretary Chris Wright and tech advisor Michael Kratsios are orchestrating the meeting on March 4, after Trump had already announced the initiative in his State of the Union. → Casey Newton

Synthszr Take: With their own power plants, Big Tech is buying regulatory freedom — a classic deal between Silicon Valley and Washington. Corporations are not only bypassing grid bottlenecks but also political debates about energy distribution, creating their own infrastructure islands. For IT service providers, this means data centers are becoming fully integrated energy-tech complexes that require new competencies in power plant technology and energy management. German providers without these skills will lose touch with hyperscale AI projects. The move showcases Trump's transactional political style: tech gets freedom to develop in exchange for shouldering societal costs — a model that Europe cannot replicate in the same way.

Perplexity Computer: An Agent for All AI Cases?

Perplexity is launching Computer, a cloud-based AI agent that orchestrates 19 different models and can independently create sub-agents for specific tasks. The service runs exclusively for Max subscribers ($200 per month) and is designed to autonomously handle complex workflows from data collection to final visualization. The company had to cancel the planned live demo hours before the press conference due to technical issues. Perplexity is repositioning itself: instead of ads, it's focusing on high-paying enterprise customers who make 'GDP-moving decisions' — a deliberate move away from the mass market. The company now operates its own AI-optimized search API and relies on multi-model orchestration: Gemini Flash for visual outputs, Claude Sonnet 4.5 for software engineering, and GPT-5.1 for medical research. The Comet browser app will be released on iOS in February, followed by a developer conference in San Francisco in March. →  Perplexity.ai

Synthszr Take: Perplexity is building the Salesforce for AI workflows — $200 a month for orchestrated agent work instead of individual API calls. The canceled demo highlights the core problem: agent reliability remains the crucial technical question, while Perplexity is already defining the pricing structure for enterprise automation. A new business area is emerging for service providers: integration and governance of multi-model workflows that juggle different providers. The pivot away from advertising toward premium enterprise makes strategic sense — you can't beat OpenAI's 800 million users with volume, but you can with specialized orchestration. German mid-sized companies will need these agent layers to securely use various AI models with their own data. The question is no longer which model, but who masters the orchestration.

Salesforce & The Agentic Cannibalization

Salesforce CEO Marc Benioff proclaims a new era of the 'agentic revolution' and promises to bring billions of autonomous AI agents into companies. These agents are intended to independently handle customer inquiries, orchestrate sales processes, and execute complex workflows without human intervention. The CRM giant is positioning its Agentforce platform as a fundamental shift: away from assisting copilots and toward autonomous digital workers. Benioff goes so far as to declare traditional SaaS models obsolete — a remarkable statement from a provider whose core business is based on precisely that model. The agents promise a 90% cost reduction in customer service interactions and are meant to 'augment' human employees, not replace them. Salesforce sees itself in direct competition with Microsoft and its Copilot strategy, with Benioff framing the difference between 'assisting' and 'automating' as a key competitive advantage. → The Business Engineer

Synthszr Take: Benioff is deliberately cannibalizing his own business model here — a classic case of disruption from within. Agents that act autonomously rather than just assisting will, in the long run, make large parts of traditional CRM license fees obsolete: Why pay for 100 sales users when 10 agents can do the same job? For IT service providers, this means a fundamental shift: instead of selling process automation, they will be selling agent orchestration. The real challenge isn't the technology, but the governance of autonomous systems — who is liable if a sales agent independently grants discounts or modifies contracts? Salesforce is betting that companies will trade this loss of control for massive efficiency gains. A risky, but probably correct, calculation.

Jack Dorsey's Drastic Cuts

Jack Dorsey announces the layoff of 4,000 employees at Block — 40 percent of the workforce. The fintech conglomerate, which includes Square, Cash App, and Tidal, will shrink to fewer than 6,000 employees. Dorsey explicitly justified the move with AI integration: smaller, flatter teams could work more effectively with 'intelligence tools.' Block continues to post rising profits; the decision is said to be proactive, not reactive. Analysts point out that Block had tripled its workforce between 2019 and 2022 — and that Dorsey had already struggled with profitability at Twitter. The stock market rewarded the move: Block shares rose after the announcement. Dorsey predicts that 'the majority of companies will come to the same conclusion within a year.' → Tech Brew

Synthszr Take: Block isn't laying off 4,000 employees due to losses, but because of AI — a precedent that legitimizes CEOs worldwide to make similar cuts. The tripling of the workforce from 2019-2022 was classic tech hubris; now AI serves as an elegant excuse for the correction. Dorsey provides no details on how exactly 'intelligence tools' will replace 40 percent of jobs — he doesn't have to, as long as the stock market applauds. Anyone still selling 'AI consulting' as an add-on instead of a fundamental organizational transformation hasn't read the signs. The truth, however, could also be that Jack Dorsey massively overstaffed Block — similar to what happened at Twitter.

The Citrini Study: A Case for Dystopian AI

In his AI Supremacy newsletter, Michael Spencer analyzes the controversial Citrini Research study, 'The 2028 Global Intelligence Crisis,' which caused significant market turmoil in late February. The speculative 'memo from the future' outlines a scenario where AI agents become so successful by 2028 that they trigger an 'Intelligence Displacement Spiral' — leading to mass layoffs in the white-collar sector, a rise in unemployment to 10.2%, and a 'Ghost GDP' where productivity gains don't reach the real economy. The study triggered the largest stock drop for IBM in 20 years (down 13%), while Accenture, American Express, and DoorDash also recorded double-digit losses. Spencer himself is skeptical of the dramatic 2028 scenario but points to the real risks: Jeremy Ney of American Inequality is already documenting measurable job losses in AI-exposed sectors, with young graduates (ages 22-25) in tech positions experiencing a 13% decline in employment since ChatGPT. Spencer sees the real danger less in technical disruption and more in the emerging 'AI Aristocracy': while the top one percent of households saw a five trillion dollar increase in wealth in the second quarter of 2025 alone, the bottom 50% gained only 150 billion — a 33-fold difference. → Michael Spencer from AI Supremacy

Synthszr Take: Citrini Research accidentally created the perfect weapon for short sellers: a viral dystopian fan-fiction that triggers immediate, measurable market movements. IBM didn't lose 21% YTD because of a real AI threat, but because a Substack post claimed Anthropic could modernize COBOL systems. For IT service providers, this reveals the new reality: markets react faster to AI narratives than companies can implement AI. The actual disruption is happening more slowly — in 2026, enterprise AI will still be stuck in pilot hell. What Spencer correctly identifies is that wealth concentration from AI-driven stock prices is already creating a 'Ghost Economy' where productivity gains only land with the top 10%. While US markets react to hype cycles, there is still room for substantial transformation projects — as long as you don't completely ignore the narrative battle.

Unitree's Most Powerful Robot Dog Yet

Unitree has unveiled the As2, its most powerful quadruped yet, which can sprint up to 11 mph and carry heavy equipment through rough terrain. The robot withstands rain, dust, and extreme temperatures — designed for industrial inspections, outdoor deliveries, and emergency response where traditional wheeled robots fail. In parallel, Chinese smartphone giant Honor is entering the robotics market with a humanoid home assistant, which debuted at the Mobile World Congress in Barcelona. The system works in tandem with Honor's 'Robot Phone,' creating an AI ecosystem beyond traditional mobile devices. Meanwhile, Google is pulling its robotics software company Intrinsic back into the parent company after five years as a standalone 'Other Bet' — a signal of direct competition with Amazon and Tesla in industrial automation. → Superhuman – Zain Kahn

Synthszr Take: Unitree is currently redefining what 'consumer robotics' means — not with cuter pet simulations, but with industrial-grade hardware at a consumer price point. An As2 costs less than a premium e-bike but performs inspection work that currently requires specialized teams. Meanwhile, Honor is copying Xiaomi's playbook: using smartphone margins to finance entry into adjacent markets while leveraging its own customer base as a distribution channel. With Intrinsic, Google is doing what Microsoft did with GitHub — bringing a standalone developer community back into the core business to generate cloud revenue. The common denominator: robotics is becoming a platform battle where Chinese hardware excellence meets American software dominance.