AI Stocks Today (Dec. 22, 2025): Nvidia's China Chip Pivot, Micron's Memory Squeeze, and 2026 Forecasts Driving the AI Trade
Updated: Dec. 22, 2025 - 10:22 a.m. ET (15:22 UTC)
U.S. stocks opened the holiday-shortened week with AI back in the driver's seat. Semis are leading, cloud is selective, and the same question keeps getting louder: who keeps the revenue, and who keeps the margins?
Market snapshot (around 10:10 a.m. ET)
- Semis ETFs: SOXX +1.56%, SMH +1.0%
- Broad market: SPY +0.43%, QQQ +0.42%
- AI bellwethers: NVDA $183.47 (+1.37%), MU $271.73 (+2.18%), AMD $214.11 (+0.32%), AVGO $340.91 (+0.16%), TSM $293.29 (+1.50%), AMAT $259.03 (+1.02%), LRCX $173.59 (+0.77%)
- Platforms in focus: MSFT $484.30 (-0.33%), GOOGL $306.98 (flat to slightly lower)
Volume will likely stay muted into the midweek early close, with markets shut Thursday for Christmas.
1) Nvidia: China shipments are back on the table, with policy risk attached
Sources indicate Nvidia plans to start shipping H200 AI chips to China by mid-February 2026, initially 5,000-10,000 modules (roughly 40,000-80,000 chips). Beijing approval is still pending, and the timeline could shift based on decisions in China and U.S. license reviews.
The reported policy setup: sales allowed with a 25% fee and an interagency review process. Nvidia is also preparing to add capacity, with orders for that capacity opening in Q2 2026. The key investment angle: H200 performance is a step up from the China-tailored H20, which would be a meaningful upgrade for Chinese buyers if approvals land.
AMD sits in the same stream. Its MI308 was designed to comply with export rules, and policy changes keep reshaping the addressable market. "China optionality" is back-so is volatility.
2) Micron and the memory squeeze: AI demand is spilling into consumer hardware
Data center AI builds are pulling DRAM into a supply crunch. Memory makers are prioritizing high-margin data center demand, leaving PCs and consoles short and raising prices.
Estimates call for memory prices to jump roughly 30% in Q4 2025 and another 20% early in 2026. That math pushes console prices up 10%-15% and PCs as high as 30%, with console forecasts already getting trimmed.
No surprise Micron keeps catching a bid. Strong guidance tied to AI data center demand reframed MU as a pricing-power beneficiary, not just a cycle rider.
3) Broadcom and Oracle: demand is great, margins decide the multiple
Broadcom guided to strong AI revenue and disclosed a $73B backlog shipping over 18 months. The catch: a higher mix of lower-margin custom AI silicon has investors modeling pressure on profitability.
Oracle flagged sales and profit below expectations while lifting spending by $15B versus prior plans. The broader takeaway: the AI build is massive, cloud players are spending hard, and Wall Street wants clearer line-of-sight to returns-not just capacity.
4) Microsoft and Alphabet: platform premium hinges on monetization
Microsoft picked up a fresh Street call with a 2026-leaning upside case and a $625 target, arguing data center spend should convert into enterprise AI revenue. The debate has shifted from "does AI matter?" to "how fast do the dollars show up?"
Alphabet got a price-target bump to $350 on the "AI tools + cloud" thesis. A recent multi-year Google Cloud-Palo Alto Networks deal approaching $10B underscores where budgets are headed: AI-driven security services with clear, recurring use cases.
5) The underpriced constraint: power and physical infrastructure
AI isn't just chips and software. Data center power demand in the U.S. is projected to jump 22% this year to 61.8 GW and reach 134.4 GW by 2030, with grid reliability flagged as a growing risk.
That changes the equity story in three ways. Capex could run longer than expected, winners broaden into grid and clean-energy infrastructure, and companies that require huge compute to hit targets may face a higher risk premium.
For context on the grid strain, see coverage on data center power demand growth from Reuters here.
6) 2026 forecasts that keep semis in the lead
Gartner expects global AI spend to top $2T in 2026, including AI baked into mainstream devices and the infrastructure underneath. SEMI sees chipmaking equipment sales up about 9% to $126B in 2026 and another 7.3% to $135B in 2027.
These forecasts connect demand (models and apps), delivery (cloud and enterprise services), and buildout (fabs, memory, power, data centers). When the chain looks intact, semis tend to lead. Gartner's top-line view is outlined here.
7) Risk tape: policy, concentration, and efficiency doubts
Export controls can flip sentiment fast. Even a headline "yes" still needs licenses and approvals, which is why China-bound AI chip narratives trade on each incremental update.
Market concentration is a growing concern. The top U.S. tech names now rival entire national markets, which can magnify drawdowns and trigger odd passive flows when volatility spikes.
There's also a hardware efficiency debate. Some prominent investors argue the U.S. could lose ground if it leans into ever more power-hungry chips instead of more efficient, task-specific silicon. Add political attention on AI-adjacent contractors, and you get headline risk layered on top of fundamentals.
What to watch next (this week)
- Any clarity on U.S. licensing and China approvals for advanced AI chips.
- Fresh memory pricing signals from the supply chain; MU sensitivity stays high.
- Capex and margin commentary from hyperscalers and AI infrastructure vendors.
- Macro prints (GDP, consumer confidence, jobless claims) in thin holiday trading.
Practical takeaways for investors
- Semis lead on clean catalysts; memory has pricing power as AI soaks capacity.
- Separate "AI revenue" from "AI margins." Custom silicon and heavy cloud capex can drag mix.
- Don't ignore power and grid. Infrastructure exposures can participate while core AI names digest.
- Expect headline whiplash on export policy; size positions accordingly.
Want a quick scan of AI tools relevant to finance teams? Explore our curated list here.
Your membership also unlocks:
