NVIDIA’s “Structured” Groq Deal Still Faces Antitrust Risk, AMD RDNA 5 Locks in TSMC N3P, Tesla FSD Tests the EU Pathway: Dec 27, 2025 · 24-Hour AI Briefing
Over the past 24 hours, three narratives intensified at once: incumbents using deal structure to manage regulatory exposure, chipmakers choosing process nodes for yield and cost curves, and automakers probing EU approval mechanics for a “single-point breakthrough.” Different headlines, same underlying shift: the next phase of competition is becoming a combined battle across ecosystems, supply chains, and regulation.
1) NVIDIA’s ~$20B Groq asset deal could still trigger antitrust scrutiny
Although NVIDIA announced on Dec 24 a roughly $20B acquisition of Groq’s assets—paired with non-exclusive licensing to reduce antitrust exposure—new signals suggest the transaction may still face regulatory challenges.
Commentary:
Groq’s LPU architecture represents an inference path outside CUDA, and notably less dependent on HBM and CoWoS packaging. That makes it one of the more credible long-term threats to NVIDIA’s inference dominance, even if current share is small. Regulators may argue the deal eliminates future competitive pressure.
NVIDIA emphasizes Groq will continue as an independent company and that GroqCloud will be unaffected, but many in markets and policy circles view this as a highly structured acquisition of assets and talent in substance.
In short: the legal wrapper may be clever, but the strategic intent is obvious and the headline size is large—hard to keep off the antitrust radar.
Do you expect this deal to face a formal antitrust investigation?
2) Leak: AMD RDNA 5 targets mid-2027, stays with TSMC on N3P
A leak suggests AMD’s next-gen RDNA 5 GPUs will arrive around mid-2027, continuing with TSMC manufacturing and adopting the N3P node.
Commentary:
N3P is widely treated as a performance-enhanced variant within the N3 family, implying improved frequency/power characteristics and more mature production learning. For high-volume consumer GPUs, yield stability and cost curves often matter more than being first on a bleeding-edge node.
There were earlier rumors that AMD could pivot to Samsung 3nm for cost or diversification. This update points the other way: RDNA 5 remains TSMC-exclusive on N3P—at the expense of waiting roughly another 18 months.
If the leak holds, how much AI market share can AMD realistically capture by 2027?
3) Netherlands to test Tesla FSD in Feb 2026: a potential EU “single-point breakout”
Dutch regulators plan to test Tesla’s Full Self-Driving (FSD) system in February 2026. If the pathway works and approval follows, it could help open the door to broader EU rollout.
Commentary:
If the Dutch testing path proves viable, Tesla could try to convert a single-country win into wider access via EU type-approval and mutual recognition dynamics. But testing is not approval, and approval is not immediate large-scale deployment.
So far, the RDW has not promised approval—what’s on the table is a supervised on-road demonstration in Feb 2026.
The Netherlands is one of the key countries in the EU Whole Vehicle Type Approval (WVTA) landscape. A national approval at L3+ could, in theory, become leverage for expansion across member states—depending on how mutual recognition is applied in practice.
Will Tesla’s February FSD test succeed, in your view?
The most important AI events from the past 72 hours
if regulatory friction becomes the dominant marginal cost for AI incumbents, will the next cycle of AI competition look more like engineering—or a long campaign around ecosystems and rule-making?