Market Overview — December 30, 2026

Summary

• Risk assets should trend higher in 2026, but the road will be bumpy. The speaker sees a mix of stable growth, falling inflation, slightly softer employment, and coordinated monetary and fiscal easing, so the macro backdrop “points to a bull market”; the real source of volatility is the market’s enormous disagreement over the payback period for AI — “bumpy road ahead.”
• The “AI divide” is the year’s central fault line. Wall Street argues that digital productivity is not true productivity, worrying about insufficient demand, overinvestment, and excessive leverage; Silicon Valley sees token demand as so strong that supply may remain short for the next 2–3 years. Both sides agree that semiconductors will be in short supply in 2026, possibly from the start of the year; the longer-term token shortage thesis comes mainly from Silicon Valley.
• U.S. growth should hold up through the year. Third-quarter GDP came in at 4.3%; although the first and second quarters of 2026 may absorb a temporary hit from Trump’s policies last year, improving liquidity, government spending, and AI productivity should keep full-year growth “steady.” AI replacing some software-engineering jobs should also leave employment just soft enough to give the Fed room to stay dovish. “Train has left the station.”
• Inflation constraints should continue to ease. Leading wage indicators are falling, tariffs are a one-time price shock, and AI is improving efficiency while China continues exporting deflation; the speaker believes China’s property market has not bottomed, with high leverage and aging making it impossible to complete deleveraging in a single year. The result is a risk-asset-friendly mix of falling inflation, slightly weaker unemployment, and stable growth.
• Semiconductors are not a TPU-versus-GPU choice; the entire chain could face shortages. Wall Street prices memory and optical modules on production shortages while still debating whether Nvidia or TPU will win; the speaker’s rebuttal is that token demand is sufficient for both to grow. “If the ship sinks, how can the people on it survive?” If AI demand is disproved, MU, optical modules, and the rest of the chain will be unable to escape.
• The biggest gap between expectations is in GPUs and CPUs, not the already-loved back-end names. Micron is reportedly shifting phone production to data centers, while Honor is already struggling to secure supply and phone shipments could fall 20% next year; the shortages and allocation logic in memory and optical modules are already widely understood, but the scale of the memory shortage remains unclear. AI Agents will generate and store separate data for each user, and distributed storage ultimately still relies on CPUs for reads and writes, so “the logic for storage corresponds to the logic for CPUs.”
• Tactical exposure should stay around 60%—70%, using certainty to absorb volatility. The speaker sees Nvidia as offering the highest certainty or the largest expectation gap: the 2026 chip shortfall is about 4 million units, earnings are expected at $9.2—$9.5, and the stock trades at roughly 18—19x P/E; if the market corrects 10%—15%, high-valuation optical modules could fall 50% and memory 30%—50%. “This isn’t the time to go to cash; it’s the time to buy names that can take a hit.”

Deep dive

1. The Macro Mix Points to a Bull Market; AI Disagreement Determines the Ride

• The speaker’s one-line outlook is: “A bumpy road higher.” Growth, inflation, employment, and policy have yet to produce a major macro shock, but the market is sharply divided over the returns on AI investment.
• Wall Street worries about insufficient demand and overinvestment, so it shorts companies whose output has been pushed into 2027—2028 and drives up Oracle’s CDS; yet it is willing to chase storage and optical modules expected to run short immediately in 2026.
• Silicon Valley takes the opposite view: token supply may remain short for the next 2—3 years. Although the speaker comes from Wall Street, he explicitly puts more faith in Silicon Valley, especially in the medium-term window through the end of 2026.

2. Stable Growth and Slightly Softer Employment Keep Easing Policy Alive

• Third-quarter GDP reached 4.3%; even if the first half of 2026 sees a temporary dip caused by Trump’s policies last year, liquidity, fiscal stimulus, and AI productivity may still keep full-year growth “steady.”
• He sees AI Agents already entering enterprises, SaaS, and programming. OpenAI’s To B business has reportedly reached 6 categories, and the U.S. has onboarded a large number of companies with essentially no BD effort.
• AI will also put some people out of work, including software engineers, so employment will not be especially strong. That “slightly soft” backdrop actually supports the Fed in loosening constraints on banks, supplying liquidity, and maintaining a dovish bias.
• Stable bond yields are, in his view, a signal of policy certainty: “Train has left the station.” Whoever chairs the Fed, the broad monetary stance for 2026 is already locked in as accommodative.

3. Tariff Effects Fade as China Continues Exporting Deflation to the U.S.

• The speaker’s clear call for 2026 is falling inflation: wage surveys and other leading indicators are all moving lower, with no sign that the hardest force to suppress — the wage-price spiral — is reaccelerating.
• Tariffs may explain part of the earlier inflation, but they are a one-time price shock: “If the cup goes up 10%, it won’t go up another 10% next year.” As the base rolls over, their ongoing impact on PCE should diminish.
• The other two forces are higher efficiency from AI and deflation exported by China. He believes China’s property market has not bottomed; high leverage, an aging population, and the sheer scale of the property sector mean deleveraging will not end within a year.
• He uses the Chernobyl nuclear power plant as a metaphor: policymakers keep pulling out control rods to raise output, but old leverage prevents the effort from working. Until the excesses are cleared, the main spillover to the U.S. remains goods deflation.

4. Fiscal and Financial Conditions Provide a Floor; Policy Events Are the Main Risk

• In the Goldman data he cited, financial conditions and global growth each contribute about 0.2 percentage points; the fiscal impulse from the U.S., China, and Europe is broadly turning positive or remaining modestly positive in 2026.
• China’s first-quarter spending could contribute about 0.4 percentage points to GDP, with additional policy possible if the impact falls short. The effect of U.S. fourth-quarter spending is clouded by the government shutdown, while fiscal transmission itself has a lag, meaning announced support may not reach growth data until later quarters.
• The speaker does not rule out a social or policy shock resembling Black Lives Matter during the midterms. But looking only at inflation, growth, employment, and fiscal and monetary conditions, there is no massive macro tail risk comparable to the one seen in April.

5. The Real Compute Bottleneck Is Shifting from Process Nodes to Advanced Packaging

• Leading chips are increasingly moving from the edge to the cloud because phones cannot absorb doubled costs, while a roughly $3M NVL72 system can absorb higher chip and packaging costs.
• TSMC’s productivity gains are shifting from wafer processes such as N5, N3, and N2 toward advanced packaging. Whoever secures packaging capacity is the one able to ship more GPUs and convert them into revenue.
• Wall Street nevertheless applies two valuation frameworks to the chain: it prices memory on shortage-driven output but refuses to do the same for GPUs because of TPU competition. The speaker’s conclusion is that “TPU wins or Nvidia wins” is a false choice; both can win.

6. Memory’s Initial Rally Is Only Act One; AI Agents Will Eventually Push CPUs into Shortage

• Micron has reportedly shifted phone allocations to data centers because margins are higher there. Honor is still struggling to meet demand despite sourcing externally, and the speaker expects phone shipments could fall 20%.
• The allocation logic for optical modules and memory is already widely understood, but the size of the memory shortage remains unclear. Optical-module valuations are already high but could continue rising. MU currently trades at roughly 6x P/E, broadly in line with several investment banks’ estimates.
• AI Agents change the storage model: traditional platforms let every user call the same hotel data, while an Agent separately generates and stores results for users in Tsim Sha Tsui, Wan Chai, and elsewhere. The resulting data volume is “astronomical — truly frightening.”
• That distributed data still requires CPUs for reads and writes, while GPUs handle centralized training and API services. The speaker says only Intel and TSMC can make CPUs, while three companies can expand memory capacity, with NAND and optical modules easier to ramp. The market will therefore first chase the easier-to-expand names before the pressure reaches harder-to-expand CPUs and GPUs. He expects Intel chips to rise in price 3 times, and AMD is also reportedly preparing price increases.
• He also emphasizes that OpenAI already has roughly 800M monthly active To C users. Content generated through To B can later be accumulated as inventory and extended to To C; financial training, KYC, account opening, meeting minutes, and PPTs can already be generated with a single click.

7. Nvidia Offers the Largest Expectation Gap, but High Exposure Means Accepting High Volatility

• The speaker sees a larger expectation gap in front-end GPUs and CPUs than in the back-end shortage names that have already been heavily traded. The largest gap is in GPUs or TPUs: “The investment case is bigger on the left than the right, but today’s valuation has the right bigger than the left.”
• Nvidia is viewed as the most durable choice: it was reportedly short about 3M chips in 2025 and about 4M in 2026; 2026 earnings are expected at roughly $9.2—$9.5, implying a current P/E of about 18—19x.
• His falsification test is straightforward: if Nvidia falls to roughly $120—$130 and de-rates to 12—15x P/E, AI demand has been disproven. MU could then be cut in half, LITE could fall below $200 or even below $100, and AAOI could drop below $10; nobody on the ship would survive.
• Tactical exposure should therefore not be too low: keep 60%—70% invested and prioritize assets with greater certainty. A high-valuation winner can fall 30%—50% when the market starts asking questions; an asset that may ultimately rise 100% can still go through a violent de-rating.