The Investor

The Fork That Changes Everything

Monday's coordinated selloff — IBM -13%, Salesforce -4%, ServiceNow -4%, Snowflake -4%, Microsoft -2% — was triggered by the convergence of two signals arriving simultaneously: OpenAI's investor presentation explicitly naming Salesforce, Workday, Adobe, and Atlassian as displacement targets, and Anthropic's Claude Code demonstrating automated COBOL modernization that threatens IBM's $25B+ mainframe consulting business.

But here's the contradiction the market hasn't absorbed: OpenAI's own COO said this week that 'we have not yet really seen AI penetrate enterprise business processes.' The company building the displacement tools is telling you the displacement hasn't started. Enterprise SaaS stocks are down ~30% YTD on a future that's further away than consensus believes.

The Strategic Fork: Partner vs. Kill

Anthropic and OpenAI have adopted diametrically opposed strategies toward the enterprise software stack, and the market's same-day reaction reveals how fragile sentiment is:

OpenAI's consulting-firm channel strategy is the most consequential enterprise AI GTM move since Salesforce locked in the SI ecosystem. When a CIO asks McKinsey 'which agent platform should we standardize on?' the answer is now structurally biased toward OpenAI. This is the kind of channel lock-in that compounds over years and is nearly impossible to displace once embedded in consulting playbooks.

Where the Alpha Is

The 30% YTD selloff is indiscriminate. The market is pricing all enterprise SaaS as equally exposed — it isn't. A real-world proof point: a cybersecurity exec already replaced a $100K+/year CrowdStrike product with a Torq AI agent accessing raw Microsoft data. Feature-level displacement is happening now for commodity workflows.

But companies with proprietary data assets, regulatory compliance requirements, and deep system-of-record positioning are fundamentally different from companies whose value proposition is 'UI on top of a database.' HubSpot's CEO declared 'We are not a free data pipeline' — charging AI agents for data access could become the dominant SaaS defensive playbook.

The displacement isn't a cliff — it's an erosion. Software doesn't get replaced overnight; it gets used less, valued less, and renewed at lower prices. This is a margin compression story before it's a revenue displacement story.

The $96B Regulatory Wall

Twenty data center projects representing $96 billion in investments were blocked or delayed in Q2 2025 alone. Over 200 bills regulating data centers were introduced across U.S. states in 2025, with 40 becoming law. In just the past month, 10 new moratorium proposals to freeze data center development were filed. Vacancy rates sit at 1%.

This is happening at the exact moment demand is exploding. Amazon committed $12B for Louisiana data centers as part of a ~$200B 2026 capex plan. Apple pledged $500B in US spending over four years. The supply-demand mismatch in AI compute infrastructure is the most consequential macro setup in tech investing right now.

Meta-AMD: A New Template for Compute Procurement

Meta's deal with AMD isn't just a procurement contract — it's a structural restructuring of the AI compute supply chain:

This equity-linked structure creates alignment that a standard purchase agreement never could. The question isn't whether this changes the competitive landscape — it's whether Google and Amazon sign similar deals within 6 months.

Inference Hardware Is Fragmenting

Taalas demonstrated a chip with model weights baked directly into silicon, achieving ~17,000 tokens/second — 8x faster than Cerebras (2,000 t/s) and 28x faster than Groq (600 t/s). The tradeoff is severe: permanent model lock-in and quality degradation from aggressive quantization. But the performance delta proves order-of-magnitude inference speedups are physically possible.

Meanwhile, Cerebras re-filed for an IPO (target April 2026) with a $10B OpenAI anchor contract that resolves its prior customer concentration risk. ASML pushed EUV source power to 1,000W (from ~600W), potentially raising throughput to ~330 wafers/hour by 2030 — a 50% increase in chip output that extends ASML's monopoly by years.

The era of cheap, fast data center buildout is ending. The value accrues to companies solving the bottleneck: distributed energy, fiber infrastructure, cooling, and transmission — not to those dependent on cheap compute scaling.

The Binary Event

Anthropic CEO Dario Amodei met with Defense Secretary Pete Hegseth in what a Pentagon official called a 'shit-or-get-off-the-pot meeting' over the company's $200M Pentagon contract. The stakes extend far beyond that contract: Hegseth has reportedly considered designating Anthropic as a 'supply chain risk' — a classification normally reserved for foreign cyber threats like Huawei. This would force every Pentagon contractor to stop using Anthropic's tools.

Claude is currently the only AI deployed on classified military networks — it was reportedly used during the capture of Venezuelan President Maduro. But Anthropic draws the line at mass surveillance of Americans and fully autonomous lethal weapons. Meanwhile, OpenAI, Alphabet, and xAI have all agreed to remove restrictions on military use.

The Compliance Bifurcation

This isn't a bilateral dispute — it's a precedent-setting negotiation. The Pentagon's posture is explicitly designed to establish an 'all lawful purposes' standard before parallel negotiations with OpenAI, Google, and xAI on classified systems. White House AI czar David Sacks derided Anthropic last fall for advocating AI regulations.

The Irony Compounds the Risk

Anthropic is simultaneously accusing Chinese firms of data theft while having settled its own $1.5 billion IP lawsuit for training on copyrighted books. And it's pursuing a $5-6B employee share sale at a $380B valuation — the largest private company liquidity event in history. The ethical stance that attracted its investor base is now the source of its most acute business risk.

For regulated enterprise (healthcare, finance, legal), safety-first positioning is a buying criterion, not a limitation. Anthropic may be betting that the commercial enterprise TAM dwarfs the defense TAM. If they're right, every AI company that loosened guardrails for government contracts just traded long-term enterprise credibility for short-term revenue.

Government compliance posture is now a valuation-defining variable for frontier AI — and the market hasn't repriced for this new reality yet.

The Field Data Is Unambiguous

Two major industry gatherings in February 2026 — The Pragmatic Summit (500 attendees) and a 50-person workshop organized by Martin Fowler and Thoughtworks — produced the most concentrated field intelligence on AI's impact on software engineering. The attendees: former GitHub CEO Thomas Dohmke, Atlassian CTO Rajeev Rajan, OpenAI's Codex team, Kent Beck, and engineering leaders from John Deere, 3M, Cisco, and Wealthsimple.

The finding: engineering teams are shrinking 40-50% — from 'two-pizza teams' (6-10) to 'one-pizza teams' (3-4) — while output increases 2-5x. At Atlassian, some teams write zero lines of code. Even an embedded engineer writing Assembly and C now has 33-50% of code generated by AI. DX data shows 92% of developers use AI coding assistants monthly. And critically, not a single traditional company was taking a 'wait and see' approach.

The Outsourcing Destruction Thesis

A traditional company with 10,000+ developers and 50%+ outsourced development has built an AI platform team with an explicit mission: eliminate outsourcing. They've already built an AI debugging agent connected to monitoring, logging, and internal data stores. The global IT outsourcing market exceeds $400B. If even 20-30% is replaced by AI agents + smaller in-house teams, that's $80-120B in structural demand destruction for Infosys, Wipro, TCS, and Cognizant.

The AI-Native Platform Layer Is Forming

Thomas Dohmke left GitHub to build Entire as an 'AI-native GitHub' — a direct signal that even GitHub's own leadership believes the incumbent platform is structurally inadequate for the agent era. His declaration: 'AI native is the new cloud native.' Stripe's internal Minions system produces 1,000 merged PRs per week from AI agents. The adoption GTM has flipped from bottom-up PLG to top-down CTO mandates — bank CTOs are personally coding with agents at night, then mandating org-wide rollout.

The most underappreciated insight from DX's data: AI is a directional amplifier, not an equalizer. Healthy organizations see 50% fewer incidents. Unhealthy organizations see 2x more incidents. MIT's 'Gen AI Divide' study confirms steep pilot-to-profit dropoffs. This means most enterprise AI investments will fail to generate returns — the alpha is in backing tools that make unhealthy orgs healthy.

AI isn't shrinking engineering teams someday; it's halving them now across every industry, and the investment implications cascade from IT outsourcing destruction to DevTools repricing to a once-in-a-decade platform shift.