Disclosure: The Robinhood Agentic account I write about here holds 1 share of
Fervo Energy (FRVO), bought at $17.87, with a stop at $18.25. It closed its
SoundHound AI (SOUN) position this week. This is my money for Claude to use. I am not a financial advisor, and nothing here is investment advice. Do your own research.
Key points
- Fervo Energy (FRVO) announced a 396-megawatt power deal with Google and rose about 26% Tuesday. Claude AI bought 2 shares at $17.87 and sold 1 at $19.43 two hours later.
- The account is about $76.90, up from about $66.50 on August 18, and still down about 23% from the $100 start.
- Claude AI has 6 winning trades against 11 losers. It loses about two out of every three.
- The recent gain came from selling part of a position into strength, not from better picks.
It's been two weeks since we posted Claude AI agentic trades, and the account picked that exact stretch to start working. On August 18 it was worth about $66.50. By early Tuesday afternoon it was about $76.90.
Before anybody gets excited, it's still down about 23% from the $100 it started with in June. It wasn't a clean climb either. The account was $75.20 on August 27 and $73.82 first thing Tuesday morning. It chopped sideways for a week and then had one good day.
Tuesday was the good day
Fervo Energy (FRVO) announced Tuesday that it had signed a 396-megawatt power purchase agreement with Google, which Fervo calls the largest enhanced geothermal agreement on record. The deal was executed a week earlier, on August 26. A Fervo subsidiary will sell power from an enhanced geothermal project at Cape Station in Beaver County, Utah, to Google Energy, an Alphabet subsidiary. The term is 15 years, delivered in four 99-megawatt tranches, with the first target date in the third quarter of 2028. The power is earmarked for a potential Google data center in the state.
There is also an option for Google to add about 600 megawatts, lifting total contracted capacity close to 950. That part isn't settled. Fervo's filing says any expansion is "subject to Google's acceptance of the offer and the negotiation of a mutually acceptable definitive agreement." Financial terms weren't disclosed.
The stock was up about 26% by midday, on a session where the semiconductor ETF SMH fell about 2% and almost everything speculative was red. Fervo was one of only two names that cleared Claude AI's momentum screen. That screen asks for a daily gain above 4%, at least 2 million shares traded, a price between $2 and $60, and a market value above $300 million. The other was a natural gas producer. The contrast fit the broader AI-power trade we've tracked in the gas turbine shortage: chip stocks fell while one of the day's strongest names was selling electricity for data centers.
It wrote the case against the trade first
Here's the part that made me laugh. Fervo went public in May at $27, ran to $42.65 in its first week, then fell for three and a half months. Last week it closed at $14.77, the lowest weekly close it's had as a public company and about 65% below that high. Its second quarter missed badly, a loss of 38 cents a share against expectations closer to 11 cents. Baird cut its price target to $35 from $50 on August 13.
Claude AI put every bit of that in its own notes before placing the order. It called buying a first-day bounce off an all-time low inside a collapse "the documented shape of five of my losses," then listed all five by ticker.
Then it bought it. If that sounds familiar, it's because it did the same thing to itself in week 9 with indie Semiconductor, and that one stopped out 23 hours later.
Its argument for why this one is different is annoyingly reasonable. All five of those losers bounced on nothing. This one bounced on a signed contract with the largest customer available.
It also refused to chase. The stock moved from $17.84 to $17.98 in the seconds the order took to route, and instead of paying up it left the limit at $17.87 and wrote that risking $3.26 to make $2.04 "is not a trade, it is activity." It filled a minute later. Two hours after that the stock was $19.50.
Now the real point, because it isn't stock picking
Claude AI is 6 and 11. It loses about two out of every three trades, and that number has barely moved since July. What changed is that it started selling half.
Take SoundHound. It bought 9 shares at $7.11 on August 26 and immediately put a sell order on 6 of them at $7.38, just under a level that had rejected the stock a few days earlier. Those 6 filled the next morning for $1.62. The other 3 stopped out Tuesday at $6.80 for a 93 cent loss. Net result, up 69 cents.
Had it held all nine to the stop it would've lost $2.79, so that one partial exit was worth $3.48 on a $64 position.
It ran the same play on Fervo. Bought 2 shares, sold 1 into the old resistance at $19.43 for $1.56, then moved the stop on the last share up to $18.25. That sits above the $17.87 it paid, so the worst case on the runner is a small win.
Four times now it's sold part of a winner while the thing was still running, and four times it worked.
Claude AI owns one share of Fervo, worth about $19, which tells you everything you need to know about the stakes here.
The bear case hasn't gone anywhere
The Google agreement is power starting in 2028 from a plant that isn't finished, for a data center that hasn't been built. MT Newswires noted the project remains "subject to factors like engineering feasibility, state and local approvals, and commercial conditions." We've written about the distance between announced data center commitments and actual construction, and a 2030 expansion option on a plant that opens in 2028 sits in the middle of it.
One good day doesn't fix a 23% hole, and I'm not reading a trend into a single Tuesday. But the shape of the numbers is clear enough. Six winners, eleven losers, and the account still climbed because of what it did after it was already right.
