Why Open-Source AI Could Increase Demand for Compute

Open-source AI could increase rather than reduce demand for computing infrastructure as more capable models become widely available. Amit Kukreja's market commentary connects that theme with an AMD-Anthropic deal involving a $5 billion investment and plans for up to two gigawatts of next-generation MI450 capacity beginning in 2027.

His much larger revenue figures are illustrative calculations based on assumed spending per gigawatt, not booked AMD revenue, but they show why compute capacity has become central to the AI investment thesis.2:08

After a week in which semiconductor stocks shed 20% following the emergence of Kimmy K3, Nvidia’s CEO finally spoke. In an interview published that morning, Jensen Huang argued that the market had misunderstood the Chinese model just as it had DeepSeek. He was emphatic: American companies should be allowed to use Chinese AI models. “You download the models, you can fine-tune it, you can guardrail it as you desire,” he said.

Far from being a threat, great open models lead to more usage, which in turn requires more Nvidia computers, more data centers, and more services. Huang’s message was a direct rebuttal to those who saw Chinese open-source models as a reason to cut AI capex forecasts.15:15

The logic is straightforward: open-source AI lowers the barrier to experimentation and deployment, expanding the total addressable market for compute. Huang explicitly called Kimmy K3 “a Deep Seek 2.0” and predicted the same pattern of growth that followed the earlier Chinese model. Elon Musk agreed, replying “true” to a summary of Huang’s comments.

For investors in Nvidia and the broader AI infrastructure ecosystem, Huang’s defense of open-source models was the most important signal of the day.17:29

Beyond the $5 billion investment, the AMD-Anthropic deal includes a financial backstop for Anthropic’s future data center leases, a structure that critics call “circular financing.” The bear case is familiar: Anthropic takes money from AMD, uses it to buy AMD chips, and AMD records the revenue. But the host argued that from an AMD shareholder’s perspective, the arrangement is a net positive.

Anthropic’s commitment to purchase up to two gigawatts of capacity by 2027 guarantees demand for AMD’s next-generation MI450 at a time when the company is still trying to break Nvidia’s stranglehold on AI training. AMD will also use Anthropic’s cloud models to improve its own chip design, a rare feedback loop that could accelerate product development.2:56

Super Micro Computer stunned the market with preliminary guidance that gross margins would nearly double from 8.7% to 17% in just 90 days. The stock surged 24% in early trading, and the ripple effects lifted Dell (up 11%) and HPE. The usual narrative around commodity server racks is that pricing power is nil, but SMCI’s guidance suggests something else: demand for compute is so intense that even rack integrators can command higher prices.

The host noted that the move implies SMCI has gained leverage over hyperscaler customers who are desperate for capacity. If true, it validates the thesis that the entire AI compute stack (from chips to enclosures to cooling) is shifting into a seller’s market.23:47

While semiconductors and neoclouds rallied, enterprise software stocks were getting crushed. The catalyst was a pair of earnings reports from IBM and Pegasystems, both of which cited the same phenomenon: customers are reallocating IT budgets toward cyber security hardware in response to the Mythos threat, leaving less room for workflow and SaaS tools. Pegasystems fell 18% after its quarter, and the fear spread to ServiceNow, Salesforce, and Palantir.

The host warned that if ServiceNow’s after-hours report confirms the trend, it could trigger a broader software selloff. The market is effectively punishing any software company that cannot prove it is integrated into the AI processing stack rather than sitting on the application layer.10:1340:18

Reddit’s stock dropped 9% after reports that it may cancel its $60 million data-licensing deal with Google. The market interpreted the move as a loss of a revenue stream, but the host argued the opposite: Reddit’s user-generated content is among the most valuable training data for AI models, and $60 million is a bargain. By walking away, Reddit signals it expects to command a much higher price, perhaps hundreds of millions.

The data has become a walled garden, and the AI labs that need it will have to pay up. However, the broader software selloff also dragged Reddit down, illustrating how difficult it is for any software company to avoid the sector’s negative sentiment.27:39

All eyes were on Google’s after-hours report, not for its search business but for its cloud revenue growth, capex guidance, and remaining performance obligations (RPO). Google had $486 billion in RPO last quarter, and a significant increase could signal sustained AI demand. The market is also watching whether Google’s cloud operating margin is improving and how much of its RPO is concentrated in a single customer like Anthropic.

The host noted that Google’s capex consensus was around $190 billion for 2026; any reduction would be catastrophic for semiconductor stocks, while an increase could confirm that the AI buildout cycle still has room to run. A poll of the stream’s audience showed most expected capex to rise.32:4051:10

President Trump’s tweet that morning escalated the Iran conflict: any ship attacked in the Strait of Hormuz would be met with the destruction of an Iranian bridge or power plant. Iran retaliated in kind. Oil prices jumped above $87, and the 10-year Treasury yield pushed toward 4.7%.

The host recalled that such a simultaneous rise in equities, gold, Bitcoin, oil, and bond yields has historically ended with a stock market break, as it did in April 2025. The cost of the Iran war has already reached $37.5 billion, and the Pakistanis are seeking a $10 billion fund for their mediation efforts. The macro backdrop is becoming increasingly difficult for risk assets, especially high-beta semiconductor stocks.12:3922:411:58:53

Tesla’s earnings were also due after the bell, and the host speculated about which Elon would show up: enthusiastic, bored, or melancholy. “Excited Elon is really fun because he’s the best storyteller on the planet,” the host said, “but when he’s bored on a call, it sucks the wind out of the room.”

Tesla’s stock has been range-bound, and the market is looking for signs that the company’s AI ambitions (Dojo, Optimus, FSD) can justify the valuation. With semis showing signs of life, a positive Tesla call could add fuel to the AI trade, but a lackluster one could reinforce the rotation out of high-multiple names.2:18:42

A surprising data point emerged: short interest across U.S. equities is at its highest since 2010. For the Russell 3000, it stands at 6.1%; for the S&P 500, 3.7%. Among NYSE-listed stocks, short interest reached a record 9% in late June, higher than during the 2008 crisis or the 2020 pandemic. The host noted that when everyone is hedging, a crash is less likely; it’s when put protection is cheap that the market is vulnerable.

The high level of bearishness may itself provide a contrarian floor. Combined with the strong consumer data from Capital One and AT&T, the economy is not flashing recession signals yet. But with oil, yields, and geopolitics all rising, the path forward is narrow.52:241:18:22