Billions more are dumped into federal AI programs while 95% of projects haven't returned a profit

Update your email or subscribe here!

Takeaway

Nathan Sanders and Bruce Schneier argue that DOGE’s Musk-aligned AI agenda prioritized power consolidation and deregulation over public benefit—and delivered few real efficiency gains. Despite the politicization and failed deployments, the authors contend government can still realize meaningful value from AI when guided by transparent principles, independent oversight, and civil-rights safeguards, citing practical use cases like translation, disability benefits processing, and immigration case management.

Why it matters

Public-sector AI is becoming a partisan battleground; abandoning the field risks ceding powerful tools to ideologues who would wield them to rewrite rules and erode safeguards. Buyers and operators should separate technology from corporate interests, require public use‑case disclosures, and enforce ethical frameworks through inspectors general and law. The analysis points to a looming political realignment around AI—and the costs of ‘unilateral disarmament’ by those wary of Big Tech.

Takeaway

The Trump administration is aggressively promoting AI through deregulation and a massive data-center buildout, with PwC estimating up to $1 trillion in investment by 2027 and states competing with tax incentives. Yet warning lights are flashing: MIT finds 95% of AI initiatives generate zero returns, Nvidia now represents roughly 8% of the S&P 500, and local resistance to data centers is growing. States such as California are moving ahead with light-touch transparency and whistleblower protections that add guardrails without halting innovation.

Why it matters

For enterprise and public-sector buyers, the gap between AI hype and proven productivity argues for milestone-based contracts, rigorous ROI tracking, and phased rollouts. State-level transparency rules and community pushback on data centers can impact vendor disclosures, timelines, and costs—factors that belong in procurement risk planning. Market concentration and potential valuation resets make diversification across vendors, models, and workloads a prudent hedge while the productivity dividend is still being validated.

Takeaway

Microsoft and the U.S. GSA inked a “OneGov” deal that gives a full year of Microsoft 365 Copilot at no extra cost to federal workers on the high‑security G5 license, with projected taxpayer savings of $3.1B in year one. The agreement also includes Azure discounts and removal of data transfer fees to accelerate cloud modernization. Core cloud/AI services are FedRAMP High, Copilot has provisional DoD authorization, and Microsoft is funding $20M for enablement—bringing total estimated value above $6B over three years.

Why it matters

Puts generative AI into millions of federal users at zero marginal cost, accelerating public‑sector adoption and reinforcing Microsoft’s footprint across productivity, cloud, and security. Agencies will still need change management, governance, and zero‑trust guardrails, creating demand for integrators, training partners, and AI safety tooling. The procurement model raises the bar for competitors on price, FedRAMP posture, and bundled value in government contracts.

Takeaway

Perplexity launched “Perplexity for Government,” which auto-detects .gov/.mil networks to apply default, enterprise-grade protections and lets federal users access top models (ChatGPT, Claude, Gemini, Grok, and Sonar) without creating an account. The company says federal workers’ usage will not be used to train models and is offering a government-tailored Enterprise Pro at $0.25 with a 15‑month runway. Perplexity is pursuing FedRAMP authorization and a GSA Multiple Award Schedule listing to formalize procurement.

Why it matters

Default, network-based protections and auto-enrollment lower adoption friction for federal teams already experimenting with genAI while reducing shadow IT risk. The move intensifies price and platform competition alongside GSA’s OneGov deals, giving agencies cross‑model optionality at commodity pricing. Procurement leaders should watch the FedRAMP timeline and MAS availability to align pilots with compliant acquisition paths.