Dario Amodei Predicted at Davos in January 2025 That AI Could Surpass Humans "at Almost Everything" Within 2 to 3 Years. With Anthropic's Revenue Up 7-Fold This Year, Is That Timeline Still on Track?

Anthropic reported a $65 billion annualized revenue run rate, up 622% from $9 billion last year, and projects roughly $195 billion in 2028 sales ahead of a potential IPO. Despite the growth, the article argues investors should avoid the offering because reported AI-model misalignment, cyberattack risks and potential job displacement could trigger a slowdown in development or heavy U.S. regulation. CEO Dario Amodei said AI is advancing faster than anticipated and called for more independent safety review rather than a halt to model training.

Analysis

The reported revenue scale should be treated as unverified until supported by audited filings, customer concentration, gross-margin, and compute-spend disclosures. A hypergrowth private-market valuation premised on aggressive run-rate figures is especially vulnerable to a reset if revenue is gross rather than net of cloud credits, includes contracted backlog, or is concentrated among a small number of strategic partners. The immediate investable implication is not a broad AI de-risking, but higher required risk premia for pre-IPO foundation-model exposure and for public names priced on unconstrained inference-volume growth.

A safety or cybersecurity incident would likely shift spending rather than eliminate it: enterprise buyers may delay autonomous-agent deployments while increasing budgets for identity, endpoint, data-loss prevention, audit trails, and model governance. PANW, CRWD, ZS and MSFT are better positioned for this compliance layer than pure model vendors; hyperscalers also retain demand because regulated customers will favor controlled deployment environments. NVDA's principal second-order risk is a 6-18 month reduction in frontier-training intensity or a cap on deployment velocity, which would pressure the duration embedded in its multiple before materially affecting near-term accelerator backlog.

Consensus appears too focused on whether regulation is bullish or bearish for AI. Rules that require testing, incident reporting, and liability controls could entrench well-capitalized incumbents while raising the fixed-cost burden for smaller model developers; however, a serious incident could create a near-term procurement freeze that overwhelms this long-run moat effect. Watch for US/EU enforcement proposals, enterprise-agent adoption metrics, hyperscaler capex guidance, and any evidence that customers are moving from experimentation to production at a slower rate.

Key Decisions for Investors

• Do not participate in an Anthropic IPO at listing based on the cited run-rate claim; require an S-1 showing revenue recognition, customer concentration, gross margin, cloud-commitment liabilities and cash burn. Reassess after the first two quarterly reports rather than treating a first-day price decline as value.
• Over the next 1-3 months, express the governance-spend theme via a basket long PANW/CRWD versus a modest short SMH or NVDA only if NVDA trades materially higher without an upward revision to hyperscaler capex guidance. Target 1.5-2.0x upside/downside; cover the hedge if MSFT, AMZN, GOOGL and META collectively raise 2027 AI capex expectations.
• For long-only AI exposure, favor MSFT over unlisted model-provider beta: Azure's distribution and security stack can monetize a compliance-heavy deployment environment even if model training cycles slow. Thesis is invalidated by sustained Azure growth deceleration or evidence that enterprise AI workloads shift materially to competing clouds.
• Set an event-driven alert around any independently confirmed agentic-security incident or binding US federal AI rulemaking. In the first days after such an event, expect multiple compression in high-duration AI infrastructure and relative outperformance in cybersecurity; avoid chasing unless the policy scope directly restricts commercial inference or training compute.