The week’s theme: big bets and brittle systems. Between sky‑high private valuations for AI labs, government use of investigative databases, and drones punching holes in server campuses, the thread is the same — enormous digital value sitting on physical and regulatory fault lines.

In Brief

ICE reportedly using a Palantir database to compile dossiers on protestors

Why this matters now: Department of Homeland Security agents allegedly used Palantir’s Investigative Case Management to log and “flag observers” of ICE operations, raising First Amendment and oversight questions.

Recent reporting says a partially unsealed court filing from a January operation in Maine alleges DHS agents entered photos, license-plate numbers and other identifying details about people who observed or protested ICE activity into Palantir’s Investigative Case Management (ICM). Plaintiffs contend the entries treated constitutionally protected observation and protest as criminal suspicion; the government calls the claim “meritless.” Civil‑liberties advocates are now pushing for clearer purpose limits and auditing on tools that let agencies aggregate and share sensitive personal data across systems. Read the reporting at Engadget for the initial coverage and the filing details.

“Agents used ICM to ‘flag observers.’”

Resistance to data centers spreads across Europe

Why this matters now: Local governments and activists are blocking or delaying new data centers on climate, water and tax‑benefit grounds, shifting where cloud and AI infrastructure can be built.

Across Belgium and other parts of Europe, campaigns argue that local communities shoulder environmental and grid burdens while profits flow to large U.S. tech firms. Complaints point to heavy electricity and water demands, long waits for grid connections, and weak local economic upside. Policymakers are responding with tighter scrutiny and sometimes project rejections, which could meaningfully reroute where next‑generation cloud and AI facilities land. See the Belgian reporting for examples and local quotes.

“The population and environment bear all the burden. The profits go to the US.”

Deep Dive

Anthropic: ~$145B raised, $2T IPO target. What profit makes that work?

Why this matters now: Anthropic’s reported fundraising totals and reported aim for a >$2 trillion public valuation would recast how markets price frontier AI businesses and expose their real revenues, compute costs, and risks.

The pitch floating through press and investor circles — that Anthropic could target a valuation north of $2 trillion — is the kind of number that forces a simple but brutal exercise: what revenue and profit justify that market cap? The discussion has moved from headlines into arithmetic and skepticism. Some outlets report mid‑2026 annualized revenues in the “tens of billions,” while analysts quoted elsewhere say $100–$120 billion of revenue would make a $2 trillion market value look like roughly a 20x sales multiple.

“The filing warns advanced models could pose ‘catastrophic or existential risks to humanity.’”

Put the numbers into context. At a 20x sales multiple, $2 trillion implies about $100 billion in annual sales. If Anthropic eventually ran with that revenue and achieved a 20% net margin — already optimistic for a company bearing huge compute and R&D bills — net income would be about $20 billion, implying a price/earnings ratio around 100x. Push margins to 40% (a level more typical of mature, highly profitable software franchises but unlikely for compute-heavy AI providers) and net income reaches $40 billion, yielding a 50x P/E. Both outcomes require either sustained hypergrowth and clear dominance, or an investor tolerance for eye‑popping multiples that assume future profitability rather than present cashflow.

There are practical headwinds to those optimistic scenarios. Large‑scale models are expensive to train and to serve: ongoing costs include GPUs or custom accelerators, enormous electricity and cooling bills, specialized data‑center capacity, and constant model upkeep. The prospectus warnings about societal risks add another layer — reputational problems or regulatory constraints could bump up costs, slow enterprise adoption, or prompt new compliance requirements. Customer concentration is another risk repeated in investor conversations: if a few hyperscale buyers or a single platform partner account for a big share of revenue, loss of those relationships would damage top‑line stability.

Reddit commentators and market skeptics ran through similar math and flagged the same structural issues: high capital intensity, heavy margin pressure from compute, and the question of whether the market will ever award a consistent SaaS‑like multiple to companies whose core input — compute — scales with sales. For retail investors and policy watchers, an Anthropic IPO at these valuations would finally lay bare the numbers behind frontier AI: true revenues, unit economics, customer concentration, and the actual cost of delivering the models that power Claude.

Ukrainian drones strike Russia’s largest data center, parts of 1.4M sq ft campus knocked offline

Why this matters now: Ukrainian drones reportedly damaged multiple modules at Yandex’s massive Kaluga data center, highlighting how physical attacks on server campuses can disrupt services and escalate cyber‑kinetic risk in modern conflicts.

Reports say Ukraine struck a Yandex data center in Kaluga — a sprawling campus with roughly 1.4 million square feet and some 63 megawatts of power capacity — taking several server modules offline. This followed a hit days earlier on Yandex’s Sasovo facility, where reporting indicated two of three supercomputers used for AI development were affected. Yandex told regulators and the press it was assessing damage; local authorities reported injuries and infrastructure impact. The sequence of strikes has already caused outages and stoked debate about whether large cloud and AI facilities are legitimate military targets.

“Several modules of the Yandex data center in Kaluga have been completely taken out of operation as a result of a UAV attack.”

A data‑center “module” usually means a discrete block of power, cooling and racks that can be isolated and, in theory, fail without collapsing the whole campus. Operators plan redundancy across modules and across geographic regions, but real‑world attacks test those contours. If multiple adjacent modules are disabled, a provider can still lose services if insufficient failover or cross‑site replication exists. For companies running critical services — search, payments, emergency comms — simultaneous physical outages can cascade into customer downtime, data loss, and financial penalties.

The attacks blur legal and ethical lines. International law generally protects civilian infrastructure, but adversaries argue that facilities supporting military or government communications, or that materially contribute to a country’s warfighting capacity, can be targeted. When a private firm hosts military workloads or national essential services, the distinction becomes fraught. Beyond law, there’s the operational lesson: modern resilience is as much about geography and logistics as it is about software. Mitigations include multi‑region replication, using providers in neutral jurisdictions, diversifying physical supply chains, and hardening site perimeter defenses. But these measures add cost and complexity — especially for compute‑heavy AI workloads that prefer concentrated, cheap power.

For listeners who watch cloud economics or national infrastructure, the Kaluga strike is a reminder that digital sovereignty and physical risk are inseparable. As states deploy drones and other low‑cost munitions, companies operating large facilities must factor kinetic threat into site selection, contractual SLAs, and disaster recovery plans — or accept that modern conflict can reach clean, corporate server rooms with disproportionate effects.

Closing Thought

The three stories converge on a single idea: the future of computing is where money, law and geography meet. Whether it’s a multibillion‑dollar AI IPO, government use of investigative databases, or a drone punching through a data‑center wall, the choices we make now — about valuations, oversight, and physical resilience — will determine whether digital growth is durable, dangerous, or both.

Sources