As hyperscale cloud providers continue to increase their capital expenditure and AI compute demand expands from training to new frontiers like inference and Agentic AI, investors are seeking the next critical answer: How long can this AI infrastructure cycle persist, and which segments will continue to benefit?

From September 8th to 11th, Goldman Sachs will host the Communacopia and Technology Conference in San Francisco, featuring 32 global companies spanning digital/AI chips, EDA software, analog semiconductors, semiconductor equipment, memory, and IT services. Goldman Sachs analysts have outlined five crucial debates for the conference, addressing AI compute demand, semiconductor equipment cycles, memory supply-demand dynamics, analog chip recovery, and the commercialization of AI within EDA software.

Based on Goldman Sachs' latest assessment, the resilience of AI capital expenditure remains the central pillar across the industry chain, while cyclical sectors such as equipment and memory are also seeing structural demand support. Concurrently, custom AI chips and Agentic AI are carving out new growth avenues, indicating that the growth narrative for the tech supply chain is evolving from purely expanding compute power to encompass a broader array of hardware and software segments.

Debate One: AI Compute Demand - How Will Drivers and Competitive Dynamics Evolve?

The AI compute market is entering a new phase. Historically, the focus was on GPU demand from large model training, but investors now are probing deeper: How much more capital can cloud providers inject? Will data center land, power, and facility resources become bottlenecks? Can Agentic AI generate new compute demand? How will the competition between general-purpose and custom chips play out?

Goldman Sachs anticipates that companies will largely maintain an optimistic outlook on AI demand. The capital expenditure environment for hyperscale cloud providers remains robust, with Agentic AI poised to be a new demand catalyst. As inference costs decline and applications like code generation start to demonstrate quantifiable returns on investment, the economic viability of AI infrastructure investment is being further validated.

On chip architecture, Goldman Sachs believes general-purpose merchant silicon will remain dominant in the near term, but the market share of ASICs and custom accelerator chips will gradually increase. As customers demand lower per-unit compute costs and better energy efficiency, AI chip competition will shift from purely performance-focused to a holistic optimization of performance, cost, and power consumption.

Meanwhile, NVIDIA is expected to further discuss its Rubin product cycle and Physical AI, Broadcom will likely focus on AI networking and custom XPU development, and Advanced Micro Devices is set to elaborate on its competitive position in AI infrastructure with its MI4XX rack-scale solutions and ROCm software stack.

Debate Two: Semiconductor Equipment - Can the WFE Upswing Extend to 2028?

AI capital expenditure is filtering through to the semiconductor equipment market via advanced logic, foundry, DRAM, and advanced packaging. The market's core debate centers on whether WFE growth in 2027 can outpace the projected ~35% growth in 2026, and if the current equipment up-cycle can sustain itself beyond 2028.

Goldman Sachs' outlook is relatively optimistic, suggesting the WFE up-cycle has the potential to last at least until 2028. DRAM, leading-edge logic/foundry, and advanced packaging will be primary growth engines, with capital expenditure in NAND and mature-node logic expected to gradually pick up as well.

Beyond traditional chipmakers, the entry of non-traditional customers like Terafab and a re-accelerated investment cycle from Intel could provide additional demand for equipment.

From a process perspective, deposition and etch remain key focus areas for Goldman Sachs. Additionally, advanced packaging, inspection, and metrology are likely to benefit from the growing complexity of AI chips and the expansion of advanced manufacturing nodes.

Debate Three: Memory and SSDs - Can Tight Supply and Demand Justify a Valuation Reset?

AI server demand is repositioning memory from a traditional cyclical commodity to a critical component of AI infrastructure. Current market disputes revolve around how much DRAM and NAND supply will be added by 2028, whether manufacturers' expansion will alter the global supply-demand balance, and if memory optimization by AI server makers could weaken demand growth.

Goldman Sachs' supply-demand model projects DRAM supply-demand gaps of 5.0%, 5.9%, and 3.9% for 2026, 2027, and 2028, respectively. For NAND, the corresponding gaps are projected at 4.4%, 4.6%, and 3.0%. This indicates that even with new capacity coming online over the next few years, the DRAM and NAND markets could remain undersupplied.

Beyond supply and demand, industry capital returns are also worth attention. Goldman Sachs expects most memory companies to return between 50% and 100% of their excess free cash flow to shareholders. If memory prices remain strong, improved cash flow and enhanced shareholder returns could further alter how the market prices the sector.

Technology progress is another key track to watch. In hard drives, the progress of HAMR (Heat-Assisted Magnetic Recording) mass production is notable, while in NAND, the commercialization of new technologies like HBF could influence future supply structures.

Debate Four: Analog Semiconductors - Can the Cyclical Recovery Outlast Expectations?

The market narrative for analog semiconductors is shifting from "cycle bottom" to "how much room is left for recovery."

Goldman Sachs data shows that as of June, analog chip shipments were approximately 5% above their long-term trend line. However, analysts believe this does not signal the end of the current cycle. Over the past four years, industry shipments have consistently remained below trend, implying significant pent-up demand for recovery.

Meanwhile, traditional end-markets like automotive are gradually improving, and AI data centers provide new incremental demand. If price recovery translates into revenue and margin improvements, the combination of AI demand and traditional end-market recovery could make this analog cycle more prolonged than the market currently expects.

Consequently, the market should focus not just on shipment changes, but on whether price, inventory, and end-demand improvements can form a positive feedback loop, and whether AI data center demand can emerge as a new structural variable that extends the industry's cycle.

Debate Five: EDA Software - Can Agentic AI Unlock a New Growth Curve?

If AI compute, equipment, and memory are the immediate beneficiaries of infrastructure investment, EDA software could be the next growth area.

Goldman Sachs is focused on whether the custom AI chip wave can accelerate EDA industry growth, and if Agentic AI can convert design efficiency gains into actual revenue.

With rising customization in AI chips and increasing design complexity, the shortage of engineers is becoming more acute. Goldman Sachs suggests that EDA companies can leverage Agentic AI to automate parts of design, verification, and optimization processes, converting efficiency gains into commercial value.

Goldman Sachs estimates that by 2030, Agentic AI could create an incremental market of approximately $3.7 billion annually for the EDA industry. This opportunity is not yet fully reflected in sell-side earnings estimates and could start to materialize as early as the second half of 2026.