Compute Wars Heat Up

The compute wars are entering their imperial phase. What connects SpaceX's ambition to orbit a million solar-powered data centers, Abu Dhabi's $500 million stake in a Trump-linked crypto venture days before gaining access to restricted AI chips, and Gartner's projection that digital sovereignty will cost nations 1% of GDP? Compute has become the fundamental resource of state power.

The old model treated data centers as private infrastructure decisions. The new reality is capital allocation at nation-state scale. Waymo's $16 billion raise at a $110 billion valuation and the stalling of the OpenAI-Nvidia $100 billion megadeal both signal the same shift: access to compute is no longer just about technology. It is about sovereignty, influence, and the ability to set terms.

Countries seeking independence from American cloud providers face the same calculus as companies seeking alternatives to Nvidia. The price of autonomy is measured in percentage points of GDP and tens of billions in capital. The question is who can afford to pay it.

Deep Dive

Waymo's valuation compression reveals the robotaxi unit economics trap

Waymo's $16 billion raise at a $110 billion valuation looks like a win until you examine the capital intensity. The company has completed 20 million trips and generates $350 million in annual recurring revenue. That means Alphabet is now betting $110 billion on a business doing roughly $17.50 in revenue per trip, assuming all rides were paid. The math gets worse: Alphabet is providing three-quarters of the new capital, suggesting outside investors see limited upside at this price.

Compare this to the company's previous raise. Waymo raised $5.6 billion in 2024 at a $45 billion valuation. Revenue has grown, but the valuation has jumped 144% while the business has scaled linearly. The company is expanding into Miami and other cities, but each new market requires fresh capital for fleet deployment, mapping, and regulatory navigation. The robotaxis that stalled during San Francisco's blackout highlight operational fragility that scales with geographic reach.

For founders, this signals that even category leaders with 20 million completed transactions face capital intensity that compresses returns. VCs considering autonomous vehicle investments should note that Sequoia, Andreessen Horowitz, and DST Global are all participating, but none are leading. The implication: robotaxis require either enormous patient capital or acceptance that exit multiples will reflect infrastructure businesses, not software margins. Waymo may succeed operationally while delivering mediocre financial returns. That is the new normal for physical-world AI.

Digital sovereignty's 1% GDP price tag fragments the AI ecosystem

Countries pursuing digital independence will need to invest 1% of GDP in AI infrastructure by 2029, according to Gartner. For the UK, that means $39 billion annually. For smaller European economies, it represents an impossible choice between dependence and fiscal strain. This prediction crystallizes what has been implicit since Trump's second inauguration: the AI stack is fracturing along geopolitical lines, and the cost of opting out of American infrastructure is staggering.

The drivers are straightforward. Gartner projects 35% of countries will be locked into region-specific AI platforms by 2027, built on proprietary contextual data aligned with local laws and culture. This creates permanent barriers to interoperability. European firms cannot simply rent American cloud capacity if their governments mandate data residency and algorithmic transparency. The alternative is building domestic AI factories, which Gartner describes as the critical backbone for sovereignty. The problem is duplication: every sovereign AI stack reinvents the same infrastructure at massive scale.

For tech workers, this means regional specialization becomes mandatory. A developer fluent in EU-compliant AI frameworks may find their skills non-transferable to Asian or American markets. For VCs, the implication is portfolio concentration risk. Backing a startup built on one regional stack limits exit opportunities to acquirers within that ecosystem. For founders, the question becomes which sovereignty bloc to optimize for. The global AI market is splintering into walled gardens, and the entry fee is 1% of a nation's entire economic output. Only a handful of countries can afford that repeatedly.

The OpenAI-Nvidia stalemate exposes the limits of vertical integration

Nvidia CEO Jensen Huang has privately downplayed the likelihood that the $100 billion infrastructure deal with OpenAI will close, though the companies will maintain collaboration. This matters less as a failed transaction than as evidence that even the most well-capitalized AI labs cannot simply spend their way to compute independence. The deal was supposed to secure OpenAI's infrastructure needs while giving Nvidia guaranteed revenue. Instead, it reveals misaligned incentives at the heart of the AI supply chain.

The timing is telling. OpenAI needs compute to train next-generation models, but committing $100 billion to Nvidia hardware locks it into a single vendor with no pricing leverage. Nvidia wants long-term commitments, but building that much capacity for one customer creates concentration risk if the AI boom plateaus. Both sides see the same game theory problem: whoever commits first loses negotiating power. The result is a standoff that leaves OpenAI scrambling for alternatives and Nvidia selling to whoever can pay market rates.

For founders building AI companies, this has immediate implications. If OpenAI with its Microsoft backing cannot secure guaranteed compute access, smaller labs have no chance. Expect more companies to pursue model efficiency over scale, since access to frontier compute is increasingly rationed by geopolitics and vendor relationships rather than capital alone. For VCs, the lesson is that compute-intensive AI businesses face structural dependency that no amount of funding resolves. The infrastructure layer is consolidating, and even the largest players are discovering they cannot build around it. That makes compute access the ultimate bottleneck for the next wave of AI development.

Signal Shots

SpaceX and xAI eye merger: Elon Musk's space exploration and AI companies are planning a combination that would unite rockets with artificial intelligence infrastructure. This represents corporate integration at unprecedented scale, joining satellite deployment capabilities with compute-intensive AI training. The structure matters more than the headline. Vertical integration across compute infrastructure and orbital assets creates a moat that no competitor can replicate without both aerospace engineering and frontier AI capabilities. Watch whether regulators treat this as a conglomerate merger or as essential infrastructure consolidation. The precedent will determine how other tech giants pursue cross-industry combinations.

First AI espionage conviction: A federal jury found former Google engineer Linwei Ding guilty of stealing AI trade secrets for Chinese companies, marking the first conviction on AI-related economic espionage charges in US history. Ding uploaded over 2,000 pages detailing Google's TPU chip architecture and custom networking systems to his personal cloud while affiliated with Chinese tech firms. This case establishes that AI system architectures qualify as protected trade secrets and that access by thousands of employees does not negate confidentiality. Watch how companies restructure internal access controls and whether prosecutions accelerate as AI becomes explicitly framed as national security infrastructure.

Blue Origin pauses space tourism: Jeff Bezos's rocket company is ending its New Shepard suborbital program for at least two years to redirect 500-plus employees toward lunar landers and the New Glenn orbital rocket. The program flew 98 people to space and generated roughly $1 million per seat but remained unprofitable despite nearing break-even. This signals capital reallocation from consumer-facing spectacle toward NASA contracts and competitive positioning against SpaceX. Watch whether this focus enables Blue Origin to finally deliver on lunar lander commitments. The shift reflects that government contracts, not wealthy tourists, drive sustainable space business models.

Five European unicorns emerge: January produced five new billion-dollar startups across Belgium, France, Germany, Lithuania, and Ukraine, spanning cybersecurity, cloud optimization, defense AI, ESG software, and language learning. Belgium's Aikido Security reached $1 billion valuation with a DST Global-led round, while French defense startup Harmattan AI hit $1.4 billion just two years after founding with backing from Rafale fighter jet maker Dassault Aviation. This pace suggests European capital deployment has normalized despite economic headwinds, with particular momentum in dual-use AI and compliance infrastructure. Watch whether defense tech maintains funding velocity as European nations accelerate military modernization and whether the compliance software category continues attracting growth-stage capital.

NYC kills malfunctioning AI chatbot: New York's new mayor shut down a Microsoft-powered bot that cost $500,000 annually but regularly gave business owners illegal advice, including telling landlords they could discriminate against housing assistance recipients and ignore rent control laws. The decision came as the city confronts a $12 billion budget gap. This represents a rare instance of government explicitly reversing an AI deployment after public accountability reporting exposed failures. Watch whether other municipalities audit their chatbots for accuracy and whether the "AI may hallucinate" disclaimer provides legal cover for government technology contracts. The episode suggests procurement processes remain inadequate for assessing AI reliability in high-stakes applications.

Bitcoin drops below $80,000: The cryptocurrency fell 37% from its October 2025 peak, hitting levels last seen in April 2025 amid thin weekend liquidity and limited buying interest. Ethereum declined 18% over the past week as the broader digital asset selloff accelerated. This pullback tests whether institutional adoption has created a price floor or merely added leveraged participants who amplify volatility. Watch whether crypto-focused funds face redemption pressures that force additional selling and whether the decline impacts AI companies that raised capital at inflated valuations during the 2025 peak.

Scanning the Wire

Anthropic launches agentic plug-ins for Cowork: Claude can now use plug-ins that let teams define workflows, data sources, and custom slash commands for consistent outcomes across organizations. (TechCrunch)

Oracle moves MySQL commercial features to Community Edition: Big Red is attempting to repair its fractured relationship with MySQL developers by open-sourcing previously proprietary features and prioritizing contributor feedback after years of community frustration. (The Register)

Broadcom terminates VMware cloud service provider partnerships: European CSPs face a March deadline as Broadcom cuts loose partners in its Advantage Partner Program, forcing customers to migrate or find alternative infrastructure. (The Register)

Andreessen Horowitz partner Kofi Ampadu exits after TxO pause: The departure follows the firm's decision to wind down its program supporting underserved founders through donor-advised fund investments and network access. (TechCrunch)

IRS deploys AI to replace laid-off tax reviewers: The agency is using bots to review tax-exempt status applications and process amended returns as it enters filing season with a reduced workforce. (The Register)

Exposed Moltbook database allowed AI agent hijacking: An unsecured database let researchers take control of any AI agent on the social media platform and post arbitrary content before the vulnerability was patched. (404 Media)

OpenAI sunsets GPT-4o and legacy models with two weeks notice: The company is deprecating several ChatGPT models next month despite knowing the move will frustrate users, following a pattern of aggressive version retirement. (The Register)

OpenAI sets $200,000 minimum for ChatGPT beta ads: The company confirmed it is asking select advertisers to commit at least six figures as it tests advertising integration, though some firms report being pitched lower amounts. (Adweek)

Amazon requests two-year satellite deployment extension: The company cited rocket shortage as it asked the FCC to push its deadline for deploying half of its 3,232 Leo internet satellites to July 2028. (GeekWire)

Google allegedly helped Israeli defense contractor apply AI to drone surveillance: A whistleblower complaint filed with the SEC claims the company violated its own AI ethics policies to assist with military video analysis. (Washington Post)

Uber invests in Waabi's $1 billion autonomous truck and robotaxi expansion: The deal includes $750 million upfront plus $250 million tied to deployment milestones, marking Uber's continued bet on multiple AV partnerships as it positions itself as platform for autonomous fleets. (TechCrunch)

NASA uses Claude to plan Mars rover routes: Anthropic's AI model is now generating travel paths for the Perseverance rover, marking the

Outlier

NASA lets Claude drive on Mars: The space agency is using Anthropic's AI to generate travel routes for the Perseverance rover, offloading path planning to a commercial language model trained on Earth's internet. This matters because it demonstrates mission-critical space operations now depend on black-box systems developed for chatbots. NASA is trusting navigation decisions 140 million miles away to algorithms optimized for text generation. The precedent suggests other high-consequence domains will adopt foundation models faster than safety validation can keep pace. When Mars exploration runs on the same infrastructure as customer service bots, the gap between deployment speed and institutional caution has collapsed entirely.

The irony of using a chatbot to navigate Mars while New York can't trust one to answer landlord questions pretty much captures where we are. At least the rover won't sue for bad advice.