Feb 4, 2026 · 24-Hour AI Briefing: Anthropic’s $350B Buyback Signal, Qwen3-Coder-Next Goes Open, and AMD’s AI GPU Revenue Still Looks Like Catch-Up
Over the past few months, the industry’s center of gravity has been shifting from “who has the strongest model” to “who can monetize, retain talent, scale distribution, and deliver reliably.” Today’s three stories map cleanly onto those three battlefields: capital + talent signaling, open-source engineering diffusion, and hardware revenue conversion in the AI cycle.
1. Anthropic plans an employee share buyback, valuing the company at $350B+
Commentary:
Anthropic’s valuation allegedly jumped from $170B in Aug 2025 to $350B by early 2026—doubling in roughly four months. That kind of step-function repricing only holds if the market believes the company has a durable path from model leadership to recurring revenue and enterprise-grade delivery.
Anthropic’s positioning is unusually reinforced by infrastructure alignment: strategic investment and deep cooperation across Microsoft Azure, NVIDIA H100, and Amazon AWS. Pair that with reported 2025 annualized revenue exceeding $5B, and the valuation story isn’t purely narrative—it’s anchored in actual monetization at scale.
In a world where OpenAI, xAI, and Google are fighting aggressively for top talent, a secondary-market buyback does two jobs at once: it gives early employees liquidity and sends a credibility signal to recruits and investors that the company expects to keep compounding. Whether $350B is “fair” and whether this is pre-IPO choreography comes down to the same hard questions: can revenue growth stay high, can inference economics keep improving, and can reliability/SLA delivery keep up as usage explodes?
2. Alibaba open-sources Qwen3-Coder-Next: 80B total parameters, ~3B active per inference
Commentary:
Qwen3-Coder-Next is clearly optimized for deployability: “usable coding capability” packaged as an enterprise component rather than a cloud-only behemoth.
Activating only ~3.75% of parameters yet hitting ~70% on SWE-Bench Verified suggests the routing strategy is paying off for code workloads. The bigger strategic move is licensing and distribution: free commercial use, already live on ModelScope and Hugging Face. For small teams that can’t justify Copilot Enterprise or CodeWhisperer, this becomes a credible, self-hostable alternative.
But the engineering reality remains: even with low active parameters, loading an 80B-class model still implies ~160GB storage footprint, and the plugin/tooling ecosystem is nowhere near Copilot’s deep VS Code integration. Adoption will hinge on whether deployment friction drops further, IDE/CI workflows become smoother, and the community builds repeatable best practices around it.
3. AMD Q4 2025 earnings: strong topline growth, but AI GPU revenue still reads as “chasing”
Commentary:
AMD posted Q4 2025 revenue of $10.27B (+34.1% YoY) and net profit of $1.5B. Client revenue ($3.9B) looks solid, and the broader PC market grew 11% YoY. The market, however, is judging AMD on AI.
AI GPU revenue was roughly ~$2.39B (about $390M from MI308 to China + ~$2.0B from other AI GPUs), under 23% of total revenue. Against NVIDIA’s ~$51.2B datacenter revenue in the same period, AMD’s AI business still looks like a catch-up story—and there’s a sense of “big R&D spend, modest shipment conversion.”
MI308’s $390M quarter is notably higher than expected and shows demand is real even under export constraints via down-binned SKUs. But the Q1 2026 outlook of only ~$100M China revenue implies that tailwind is not stable. Add memory price inflation and shifting enterprise procurement cycles, and 2026 execution risk increases.
The real question isn’t whether AMD “has AI,” but whether it can convert roadmap + R&D into sustained supply, software stickiness, and expanding hyperscaler deployments. Without those three, a true 2026 “breakout” is hard to underwrite.