EP692 | 🍺

A Volatile Market Is Beginning to Show the Conditions for the Next Bull Run

The host believes that although the market remains volatile, several recent corporate partnerships and financial figures have given it the conditions needed to lay the groundwork for the next bull run. The key point is not simply that individual companies have reported positive news, but that value distribution across the AI industry is being reshuffled: models, interfaces, enterprise applications, and supply chains may each retain different portions of the profits.

Take the partnership between Salesforce$CRM and Anthropic as an example. Cloudforce allows users to access external models within Salesforce's enterprise work interface, while also bringing Salesforce's 37 skills and plug-ins into AI tools. This is not simply a matter of one side replacing the other. It allows model capabilities to connect with existing enterprise workflows. The host therefore believes that software will not disappear as a whole just because AI has emerged; what is really happening is a redistribution of value.

The host particularly values Salesforce's enterprise moat. Although the market often complains about its products, the company has accumulated extensive enterprise data, workflows, and industry knowledge over many years, while also building customer trust. Enterprises may not be willing to hand critical data to AI startups without a long track record, even if those companies show impressive revenue growth and computing scale. Data residency, governance rights, and customer relationships may determine long-term pricing power more than a model's score in a single test.

AI Partnerships May Look Harmonious, but the Profit Battle Over Two Separate Bills Continues

After reviewing the public materials from both sides, the host points out that Cloudforce's business model may create two bills: customers pay one set of fees when they use an AI model, submit questions, or consume usage credits; if they also use the Salesforce API or call Anthropic models within the Salesforce interface, they may pay additional platform and service fees. On the surface this is a partnership, but in practice the two sides are still competing for a larger share of the revenue.

He speculates that model companies may capture higher margins at the model layer, but if enterprise data and customer relationships remain with Salesforce, the ultimate value may still flow back to the platform provider. This depends on whether enterprises are willing to hand closed-off data to a new frontier lab, and whether model providers can maintain stable products, pricing, and services in the future. A system built on Anthropic today might be switched to OpenAI tomorrow if OpenAI performs better. By comparison, an enterprise platform that can connect to different models may be the more flexible choice.

This also leads to the host's rebuttal of the claim that “AI will destroy all software.” Existing software companies can move beyond simply providing an interface and redirect resources toward enterprise actions, data, and workflows; new AI companies may gain advantages in models and user interfaces. It is like a single building whose entrances, utilities, and tenant data are controlled by different companies: the revenue will not naturally all flow to just one layer.

Marvell's Upward Revision Puts Optical Interconnects in the Spotlight

Marvell$MRVL raised its financial outlook, and the host believes the biggest contribution may come from XPU-related projects. Early design wins do not necessarily show up in revenue immediately. Clearer results tend to appear only once products begin shipping in volume. The current upward revision may therefore indicate that some products have moved closer to mass production and market delivery.

Another important growth driver is optical communications, especially optical interconnect switching chips used in scale-up architectures. The host notes that the market had already been betting that Marvell's DSP and optical communications businesses would receive an upward revision during the earnings call. That direction has now been confirmed, becoming a major catalyst for the stock's rebound. If XPUs and GPUs are to be deployed in server racks at scale, they need more than computing chips: optical components, switches, packaging, substrates, and assembly capabilities are also required. All of these areas may benefit from the expansion of AI infrastructure.

However, the host does not equate the upward revision with an absence of risk. After correcting for some time, optical communications stocks have begun strengthening again. The market still needs to watch whether the earnings call can become an ongoing catalyst, and whether demand growth will be constrained by supply, pricing, and delivery capacity. If the entire industry competes for limited capacity at the same time, revenue opportunities may exist, but margins and delivery schedules will still need to be reassessed.

NVIDIA's Moat Lies in the Supply Chain, Not Just in Chip Benchmarks

NVIDIA's$NVDA new inference chips have shown strong tokens-per-watt efficiency and inference performance in testing, outperforming some older GPU systems. The host cautions, however, that these products are primarily designed for inference. Training workloads that require enormous computing power cannot be judged by inference benchmarks alone.

When new models or chips appear every month or two, investors can easily mistake “fastest in the current test” for “certain to win in the long term. ”

Even if other XPUs, custom chips, or models perform impressively in specific tests, NVIDIA's advantage still comes from its more complete system and supply chain. The host believes the company purchases large quantities of optical components, lasers, substrates, and other parts while locking in capacity in advance. Competitors may design excellent chips but still be unable to deliver on schedule because they cannot obtain packaging, substrates, switches, or assembly capacity.

This makes supply-chain management itself a competitive barrier, rather than merely a matter of purchasing at lower cost.