Nvidia at a record, Micron feasts on earnings: what moved chip stocks on October 2

Nvidia shares struck an intraday record of 237.87 dollars on October 2, 2026, the first new high since May, ending the day up about 1.7 percent. Three separate drivers landed on the same day: a historic buyback authorization reaching 235 billion dollars, Morgan Stanley's restored top pick rating, and unabated AI infrastructure spending. The right lens is not the single-day move but all three firing at once. According to Investing.com analysis of the record, the breakout rests on the combination of three catalysts rather than one headline.

Morgan Stanley analyst Joseph Moore reinstated Nvidia as his favorite chip stock after investor meetings with chief executive Jensen Huang and finance chief Colette Kress in New York and Boston. The core of the note is valuation : the stock trades at roughly 15 times the bank's fiscal 2028 earnings estimate, a multiple that could expand if enthusiasm for AI returns. Even without expansion, rising estimates alone can carry the shares, the note argues. According to Investing.com coverage of the meetings and the 15-times math, the bank sees estimate revisions as sufficient on their own.

The new part of the note is gigawatt math. Morgan Stanley argues the AI bottleneck is moving from chip production to how fast data center space can be built and financed, which favors Nvidia's power-centric business. The bank estimates Nvidia earns about 40 billion dollars per deployed gigawatt today, rising well above 50 billion with the Feynman architecture due in 2028. That means at least 25 percent more revenue from the same power. According to TrendAndTicker analysis of the revenue-per-gigawatt projection and the 300 dollar target, the case rests on power efficiency.

The week's balance-sheet engineering came from Amazon. The company is weighing moving about 8 billion dollars of Grace Blackwell chips installed across American data centers into a special vehicle, funded with outside debt, and leasing them back. The goal is a more asset-light structure while AI spending inflates the balance sheet. This is not a chip sale but ownership transferred outward while usage continues. According to FinancialPost reporting that talks with investors ran in recent weeks, the structure is built around a debt-issuing vehicle.

Behind such financings sits a question over second-hand chip values. The Reuters inquiry summarized in the video puts the long-term value of the chips and the durability of these financings up for debate. The reply comes from CoreWeave: management argues GPUs the market prices for three-year lives actually run seven years and longer, with 2020-vintage A100 chips rentable at full rates under contracts stretching into 2029. Old chips earning profits for longer shrinks the scrap-value fear. According to CNBC analysis of the depreciation debate and the 2020 chips' contract horizon, the used market has revived too.

The hardest pricing data sits with AWS. Amazon Web Services plans a 15 percent price rise on EC2 Capacity Blocks reserved for machine learning from October 7, which would be a fourth straight quarterly hike. January 2026 brought 15 percent, July 2026 added about 20 percent; an H200-based machine went from 34.61 dollars per hour to around 39 dollars in most regions and 49.74 dollars in Northern California. The October move stems from a social media post and is not officially confirmed by AWS. According to Cryptobriefing compilation separating the confirmed calendar from the unconfirmed October claim, the rises cover only reserved GPU capacity.

Simple physics sits behind the hikes: demand exceeds supply. AWS monetizes new capacity the moment it comes online; even Nvidia GPUs launched six years ago got dearer on supply-demand grounds. Every hyperscaler described supply constraints on its latest earnings call, and clouds are building against signed contracts rather than blindly. Recalling the March 2026 plan to deploy over a million Nvidia GPUs in the coming year, Cryptobriefing assessment notes the company keeps scaling to meet demand.

The technology concept of the day was the Vera processor. Nvidia's Vera CPU, designed for agent-based workloads, targets maximum sustained per-core performance under full socket load. Olympus cores deliver 1.8 times single-thread speed; up to 85 percent of reinforcement learning evaluations fit in the same training window, while the monolithic 88-core design cuts peak loaded latency 40 percent versus x86 rivals. The inference explosion cited in the video confirms it: the train-versus-inference ratio is said to have moved from one-to-one into the twenty-to-forty-to-one band. According to the Nvidia technical blog publishing the developer detail, per-core memory bandwidth more than triples.

Micron was the earnings star of the week. Its fiscal fourth quarter reported September 30, 2026 brought 54.23 billion dollars of revenue, against 41.46 billion the prior quarter and 11.32 billion a year earlier. Adjusted earnings per share reached 33.42 dollars, GAAP net income 37.70 billion, and operating cash flow 43.97 billion. Full-year revenue of 133.19 billion dollars multiplied the prior year's 37.38 billion. Announcing the figures in its official release, the Micron investor relations page says the company enters a record fiscal 2027 on AI-driven demand.

The blemish was next-quarter gross margin guidance below the reported quarter, with a clean accounting explanation. The finance chief said fiscal 2026 compensation for every global employee was raised in the fourth quarter, with the manufacturing portion largely capitalized into inventory rather than hitting cost immediately. So margin pressure reflects the timing of a one-off pay decision, not operational weakness. Guidance sees 61.5 billion dollars of revenue next quarter, with customer prepayments at 12.7 billion. According to TheCompanyChronicle examination of the 14-week quarter detail and the 61.5 billion guide, part of the headline growth is calendar effect.

The buyback front is heating on two sides. Nvidia's board added 150 billion dollars of authorization, lifting the total program to 235 billion through fiscal 2028; the sum equals about 4.3 percent of the roughly 5.5 trillion dollar market value and is described as the largest program in corporate history. At Micron, a 100 billion dollar program is said to be on the table once buyback restrictions end this year; the company trades around 1.2 trillion with talk of up to 10 percent of the company per year. Comparing the math of both programs, CNBC reporting notes buybacks can lift earnings per share without any business growth by putting a big buyer into the market.

The twelfth thread is the Syracuse build and the competitive front. Micron is expanding its giant manufacturing campus near Syracuse, New York, the core of its plan to grow memory supply inside America. The same segment covers AMD's strong share gains and Nvidia stepping between Intel and AMD on processors with Vera. But the AMD leg gets only a few sentences in the video with no independent data, so read that part as an unconfirmed note.

A thirteenth piece sums up the hyperscalers' shared language: capacity constraint. Cloud giants lease new capacity the instant it goes live; contracted demand shows the investment is order-driven rather than speculative. Micron likewise stresses supply stays tight with new capacity absorbed immediately. The picture explains AWS hikes and Micron pricing power in one frame: physical capacity trails software demand.

Three indicators come into focus at the close. First, whether revenue per gigawatt truly clears 50 billion dollars with Feynman. Second, whether AWS reserved prices slow demand after October 7. Third, where Micron gross margin settles once the pay effect washes out. The trio will decide whether the chip-stock rally narrative carries into 2027. For investors the day condenses to one line: the hardware story is now written in power, capacity and balance sheet discipline rather than chip speed.