Analysis
Two earnings reports and one pricing decision this week drew a line through the AI trade that's been blurry for two years. On one side: real businesses converting AI demand into audited revenue. On the other: the frontier-model and newly-public hardware layer, where prices are falling and stock prices are following. Both are "AI stocks." They are no longer telling the same story.
What Shipped
Start with what shipped. Amazon crossed $3 trillion in market value on Monday, becoming only the fifth company in history to do it, after a second-quarter print that beat on every meaningful line: AWS revenue of $42.2 billion against a $40.54 billion Street estimate, total revenue of $200.61 billion against $196.47 billion expected, and adjusted EPS of $1.97 against $1.82 consensus. CEO Andy Jassy's line on the call -- AWS growing at its fastest pace in eighteen quarters -- was about cloud capacity actually getting consumed, not about a roadmap.
“Q2 revenue hit $1.94 billion against $1.80 billion expected, up 93% year-over-year, with net income of $1.07 billion versus roughly $329 million a year ago.”
Palantir told a similar story from a different seat. Q2 revenue hit $1.94 billion against $1.80 billion expected, up 93% year-over-year, with net income of $1.07 billion versus roughly $329 million a year ago. The number that matters for the enterprise AI thesis specifically is U.S. commercial revenue: $764 million, up 149% year-over-year, on top of 90% growth in U.S. government revenue to $809 million. Management didn't just beat -- it raised full-year U.S. commercial guidance to "in excess of" $3.42 billion, up from a prior $3.22 billion floor. The stock popped 12% on the print.
What's Still Promised
Now the other side of the line. OpenAI cut the price of its fastest, cheapest GPT-5.6 tier, Luna, by 80% -- from $1/$6 to $0.20/$1.20 per million input/output tokens -- just three weeks after launching the model family, undercutting Google's Gemini 3.5 Flash-Lite in the process. Sam Altman has said publicly that compute cost, once a non-issue for enterprise customers, became "a huge one" within months. That's a company defending share by cutting margin, not a company raising prices on scarcity.
SpaceX sits at the sharpest edge of that dynamic. The company reports its first-ever quarterly earnings as a public company Tuesday after market close, with the stock trading 45-48% below the intraday high it touched right after its June listing -- below its $135 IPO price even after a premarket bounce. Two days later, on August 6, up to 911.5 million insider and employee shares become tradable, roughly 20% of the eligible pool. A frontier valuation built substantially on narrative is about to get its first audited numbers and its first real supply test in the same week.
Where the Capital Is Actually Going
The capital allocation gap tells you where smart money thinks the next several years of value actually sit. Sequoia's Shaun Maguire led a $1 billion Series B into Valar Atomics, tripling the nuclear-reactor startup's valuation to $6 billion in a bet that factory-built microreactors, not more GPUs, are the actual bottleneck on AI data center growth. The same week, London's OLIX raised $312 million at a $3.3 billion valuation for a photonic inference chip explicitly designed to sidestep the HBM shortage that's currently rationing Nvidia's own supply chain -- backed by Arm, Hudson River Trading, and Netflix co-founder Reed Hastings personally.
For founders, the read is straightforward: if your pitch is "we're a thinner wrapper on a frontier model," you're competing in the segment where prices just fell 80% in three weeks and will likely fall again. If your pitch is proprietary data, a workflow enterprises will pay to keep running, or hard infrastructure someone else needs to build AI at all, you're in the segment that just got the two best earnings prints and the two biggest venture checks of the week.
What to watch: SpaceX's actual numbers Tuesday and how the stock trades into the August 6 lockup will be the next data point on whether "story" valuations can hold under real scrutiny. On the infrastructure side, watch whether Valar's and OLIX's rounds pull more specialist capital away from generalist AI-app investing over the next quarter -- that rotation, not any single earnings beat, is the trend GPs should actually be underwriting to.
