The true significance of GPT-6 Astra: It shifts the focus of "AI narrative" from "demand debate" to "physical bottlenecks."

The true significance of GPT-6 Astra: It shifts the focus of "AI narrative" from "demand debate" to "physical bottlenecks."

The release of a model is changing the direction of market debate.

For the past few months, the market has been debating one question: "How much infrastructure is needed to serve known AI needs?" The subtext of this question is: Has AI infrastructure already been over-invested in?

However, according to a report released by Morgan Stanley on September 7, the importance of GPT-6 Astra should not be simply understood as a continuation of another scaling narrative. This is not just another model upgrade.

Astra represents a substantial leap in the breadth of capabilities and interoperability, specifically in four areas: reasoning ability, engineering capabilities, computer usage, and physical world task execution. This means that AI is beginning to leap to a level capable of performing complex reasoning, specialized engineering tasks, directly manipulating computers, and even handling real-world tasks. With more things to do, there are more scenarios where it can generate revenue. OpenAI's latest publication also states that better models are opening up new "work domains."

The AI narrative has once again shifted from "demand debates" to "physical bottlenecks."

The emergence of Astra may shift the core market question from "How much infrastructure is needed to serve known AI needs?" to: "As model intelligence improves, how many new workloads will become economically viable?"

This is a question with completely opposite directions. The former is a question from the demand side, while the latter is pressure from the supply side.

  • Elasticity effect: Buy more smart devices for every dollar

The analysts presented a core economic logic in their report: better models can create a resilience effect—the more intelligence and utility AI gains for every dollar spent, the greater the quantity, duration, and complexity of inference workloads will be.

This is similar to the classic "Jevens Paradox": efficiency improvements do not reduce consumption, but rather increase total consumption. Increased intelligent output per unit cost → previously uneconomical application scenarios become feasible → new workloads emerge → the demand for computing power and infrastructure actually increases.

  • Significantly expanded market reach

Analysts believe that if Astra's capabilities translate into commercially viable applications, the pool of accessible AI revenue will expand dramatically, far exceeding today's use cases, which are primarily focused on chat and coding.

Morgan Stanley estimates that the global knowledge work TAM is approximately $22.5 trillion (based on an estimate of 900 million knowledge workers worldwide with an average annual salary of approximately $25,200); the consumer spending TAM is approximately $30 trillion, covering retail + travel ($16 trillion), autonomous driving/mobility ($4 trillion), food delivery ($4 trillion), and advertising ($3 trillion).

  • The real bottleneck lies in the physical constraints on the supply side.

Analysts concluded that Astra's emergence shifted the bottleneck narrative from "demand formation" back to "physical supply constraints"—that is, whether these intelligences can be delivered at scale.

What does this specifically refer to? Computing power, electricity, materials, labor, etc.

Where exactly are the physical bottlenecks: ABF and HBM4E

There are two specific measures to address the supply-side tensions.

The first one is the ABF carrier board.

Morgan Stanley predicts that ABF substrates will be in short supply starting in 2027, and the gap will continue to widen until 2030. The key constraint is that new capacity will take at least two years to come online.

The second is the complexity of the downstream process (BEOL) of HBM4E.

The report points out that the back-end processes of HBM4E represent a major paradigm shift in semiconductor manufacturing—HBM has evolved from dedicated 3D memory stacks to highly integrated custom chiplet logic systems.

Specific impact chain:

  • With the increase in the number of HBM4E interconnect layers (such as SK Hynix's introduction of dummy bumps), a large amount of DRAM capital expenditure will be forced to be used for BEOL capacity expansion.
  • Capital expenditures for front-end process (FEOL) migration and accelerated DRAM GB shipments are not expected to materialize until the second half of 2027.
  • DRAM manufacturers are prioritizing equipment capacity, which may delay NAND capacity expansion.

Both bottlenecks share a common characteristic: they are verifiable and traceable physical constraints, not market sentiment.

Electricity: The Most Realistic Physical Bottleneck

With increasing regulatory pressure on data centers, power supply has become another key physical bottleneck. Analysts predict that the total computing power of hyperscale cloud vendors will grow from approximately 35GW in 2025 to approximately 145GW in 2028, an increase of about four times.

Analysts point out that the United States faces a 38-gigawatt power shortage, and data centers will increasingly adopt behind-the-meter self-generating solutions.

The bank estimates that off-meter power solutions will increase capital expenditure by approximately $3 billion per gigawatt. For example, the all-inclusive cost of the Rubin Ultra 1 Nvidia chip, including off-meter power, is approximately $50 billion per gigawatt.

What has the market underestimated?

Analysts believe the market is currently underestimating three things:

First, the global technology beneficiaries of GPT-6 Astra have not yet been fully priced in.

Second, the supply shortage may last longer. The constraints of ABF and HBM4E BEOL cannot be resolved in the short term.

Third, some stocks that can rise without relying on AI are quietly gaining strength. Analog chips (STM, NXP, Renesas) are in the early stages of a cycle recovery after more than three years of L-shaped bottoming out – inventory reduction, pricing stabilization, and improved industrial orders.

The conclusion is not "buy more AI".

Morgan Stanley's investment priorities:

AI computing power (highest priority) > network (second priority) > memory (selective) + analog chips (early cycle hedging), including:

AI computing power: GPUs (NVIDIA), ASICs (MediaTek, GUC), ABF substrates (Visionox, Ibiden), MLCCs (Murata, Samsung Electro-Mechanics), back-end packaging and testing (Advanced Semiconductor, Tokyo Electron, Huafeng, ASE, KYEC), power supplies (Delta).Network: GLW, LITE, COHR, KEYS, Furukawa Electric, FujikuraMemory chips: Prioritize structural market share growth and localization (Changxin Memory); SK Hynix, Samsung, and Kioxia have tactical upside potential given the continued tight supply.Non-AI: Analog chips (STMicroelectronics, NXP, Renesas)

The bank stated, "We prefer to invest in companies where a broader inference cycle intersects with limited physical capacity, rising content intensity, and increased manufacturing complexity."

Places where you need to stay alert

Morgan Stanley's report was not one-sidedly optimistic, and it clearly pointed out three points that need attention:

First, AI expectations are already very high. The market's tolerance for AI companies has shifted from "good performance" to "perfect performance is a must." Even if Astra makes substantial progress, if its performance does not significantly exceed expectations, the stock price reaction may still be limited.

Second, capital expenditure growth is expected to slow in 2028. Stock pricing reflects the direction of change in growth rate, and a slowdown could continue to suppress valuations, even if the absolute value is still increasing.

Third, macroeconomic headwinds remain. The report mentions uncertainties such as oil prices, inflation, the Federal Reserve's interest rate path, and the 2028 US election, with particular attention to potential political resistance to data center expansion should the Democrats win the executive branch.

~~~~~~~~~~~~~~~~~~~~~~~

The above content is from Zhuifeng Trading Platform .

For more detailed analysis, including real-time updates and firsthand research, please join the [ Trading Channel Annual Membership ].

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.