SharkWater Trading • Desk Notes • Rates • AI Infrastructure • Data Gaps
5.33 Percent On The Long Bond, And Three Setups With No Paper Behind Them
August 19, 2026
Bottom Line Up Front
The thirty year Treasury yield touched 5.33 percent on August 18, the highest since June 2007, after clearing 5.311 percent on the 17th. The space, nuclear, and data center complex sold off five to sixteen percent across the board. Almost none of those companies reported anything. They were repriced, not re-rated.
Three names had their own news. Nebius (Nasdaq: NBIS) launched a $4.50 billion convertible offering this morning and fell hard. Fundrise Innovation Fund (NYSE: VCX) ran about 19 percent on August 17 against a NAV last struck on March 31. Cerebras (Nasdaq: CBRS) shipped its CS-4 and lost 12.7 percent. I am not writing a trade on any of them today, because in all three cases the document that decides the trade has not been published. That is the post.
The Only Thing That Actually Happened This Week
Strip out the single stock headlines and the tape has one driver.
| Instrument | Level | Note |
|---|---|---|
| 30 year Treasury, Aug 17 | 5.311% | Highest since June 2007 |
| 30 year Treasury, Aug 18 | ~5.285% settle | Touched a fresh 19 year high intraday |
| 10 year Treasury, Aug 18 | ~4.706% | Down about 1bp on the session |
| 2 year Treasury, Aug 18 | ~4.175% | Essentially unchanged |
| $25B 30 year auction, prior week | 5.216% | Highest auction stop since 2001 |
Source: CNBC and Bloomberg reporting on Treasury market levels, August 17 and 18, 2026. Intraday levels, not official close prints. The 5.33 percent figure is a secondary-source intraday high and has not been checked against the H.15 release.
Read the shape of that table, not the levels. The two year did not move. The ten year did not move. The long bond did. This is not the market repricing Fed policy. It is the market repricing the willingness to fund thirty year paper, against a June TIC report showing the United Kingdom, China, and Japan all trimming Treasury holdings, and against a calendar of long dated issuance that is not getting smaller.
Think of the discount rate as depth of water under the keel. Nothing about any of these boats changed on Tuesday. The tide dropped roughly sixty basis points at the long end over a month, and the hulls sitting deepest, the ones whose cash flow does not arrive until 2030, are the ones that touched bottom first.
Every name on this list is a duration asset wearing an equity ticker. A nuclear developer with first power in the early 2030s. A data center landlord commissioning in 2027. An inference company with a backlog that converts over eight years. When the long end moves, the present value of 2032 moves more than the present value of next quarter. That is arithmetic, not sentiment, and it explains a five to sixteen percent bucket move better than any single company story does.
NBIS: The Selling Is In The Press Release, Three Paragraphs Down
Nebius announced its intention this morning to offer $4.50 billion of convertible senior notes in two series: $2.75 billion due February 15, 2030 and $1.75 billion due February 15, 2034. Initial purchasers get an option on up to another $375 million of 2030 notes and $300 million of 2034 notes, so the deal can reach $5.175 billion. Proceeds are earmarked for data center construction, buildout, GPU procurement, and general corporate purposes.
The stock was quoted down between roughly 7.6 and 9.9 percent intraday depending on when you looked, on top of the prior session. Two day drawdown in the neighborhood of sixteen percent. Every wire story I read attributed that to dilution from the convert. That is the lazy read.
Read the exchange paragraph. Concurrently with pricing, Nebius expects to enter privately negotiated agreements with holders of its existing 2.00 percent notes due 2029 and 3.00 percent notes due 2031, exchanging a portion of that paper for Class A ordinary shares. The company then states, in its own release, that participating holders may sell those shares in the open market and enter into or unwind derivative hedge positions, and that this could decrease the market price of the Class A shares and may also affect the initial conversion price of the new notes.
That is the issuer telling you, in advance, that mechanical share supply and hedge unwinds are coming. Some material portion of today's move is that, not a verdict on the business. It is the same species of thing as a leveraged ETF rebalance. The print is not information.
What Is Known And What Is Not
| Term | Status as of Aug 19 |
|---|---|
| Size | $4.50B, up to $5.175B with the option. Known. |
| Maturities | Feb 15, 2030 and Feb 15, 2034. Known. |
| Coupon | Set at pricing. Not known. |
| Initial conversion rate and price | Set at pricing. Not known. |
| Accretion schedule | Set at pricing. Not known. |
| Call protection | No optional redemption before Feb 21, 2028 (2030s) or Aug 21, 2028 (2034s), then subject to a 130% price trigger, or 150% for 2034s redeemed between Aug 21 2028 and Aug 21 2029. Known. |
| Size of the note-for-share exchange | Individually negotiated. Not disclosed, may not happen at all. |
Source: Nebius Group N.V. press release, August 19, 2026, issued via Business Wire. Terms are as proposed and the offering is not guaranteed to complete.
Note the accretion structure, because it is unusual enough to matter. Principal accretes to a premium over the original amount by maturity, and the redemption price is the accreted amount. But conversion is computed off the original principal, not the accreted principal. That splits the economics: bondholders get a compounding floor, equity holders get a conversion threshold pinned to the smaller number. Whether that is cheap or expensive capital depends entirely on the coupon and conversion premium, and neither exists yet.
One more thing worth putting in front of you. In March 2026 Nebius announced $3.75 billion of converts and priced upsized at $4.0 billion. Add today's deal and this company has brought up to roughly $9.2 billion of convertible paper to market in five months, and is simultaneously retiring some of the earlier tranches into stock. That is not a capital raise. That is a capital structure being rebuilt in public, at speed, while the long end sells off.
VCX: Two Corrections, And A NAV That Is Four And A Half Months Old
Fundrise Innovation Fund did run about 19 percent on August 17. The framing in my notes was wrong twice, so let me fix it before anything else.
Correction one: this was not an insider lockup. The post listing lockup covered restricted shares held by roughly 100,000 pre-listing retail investors, with an average cost basis reported around $10 to $11 a share. Fundrise accelerated the expiration on July 24 from September 14 to August 13, which made those shares tradable on August 14. Calling it an insider unlock gets the supply profile backwards. This is a hundred thousand retail accounts sitting on three to four times money, not a handful of affiliates.
Correction two: the August 17 move was almost certainly not the unlock. The unlock already traded, on the 13th and 14th. Here is the actual tape.
| Session | Close | Move | What happened |
|---|---|---|---|
| Aug 12 | $31.89 | n/a | Day before expiration |
| Aug 13 | $38.50 | +20.73% | Lockup expires. Supply still locked. 1,402,447 shares. |
| Aug 14 | $34.40 | -10.6% | First tradable session. ~3.8M shares, roughly 17x median. All-time low $28.71 intraday. |
| Aug 17 | $41.00 | ~+19% | Anthropic valuation headlines circulate |
| Aug 18 | $40.10 | -2.2% | Range $37.26 to $42.95 |
Source: session data compiled from Investing.com quote pages and third-party unlock analysis, August 2026. Not exchange-official. The August 17 attribution is inference from the timing of press coverage, not a company disclosure.
Anthropic is the fund's largest position at roughly 20.7 percent of the portfolio. On August 17 a wave of coverage put Anthropic's 2026 revenue somewhere between $100 billion and $120 billion, with public market speculation attaching a valuation well north of a trillion. A fund with a fifth of its book in one private company moved 19 percent on a headline about that company's hypothetical listing price. That is coherent. It has nothing to do with the lockup.
The Number Everything Hangs On, And Why I Will Not Publish It Yet
The last NAV I can find attached to a filing is $18.97 per share, for the period ended March 31, 2026. At the August 18 close of $40.10 that is a premium of roughly 111 percent. Which would be the entire story, except for three problems.
One. That NAV is four and a half months old, and it predates a materially higher Anthropic mark. If Anthropic's carrying value rose between March 31 and June 30, the premium the market is paying is smaller than 111 percent, possibly much smaller.
Two. The $18.97 figure has already been the subject of a public correction elsewhere. At least one widely read analysis originally published a NAV of $18.43, could not substantiate it against a primary source, and replaced it. When two versions of the anchor number are circulating, the anchor is not verified until you have the filing open in front of you.
Three, and this one is mine. A reported market cap of about $1.16 billion against a $40.10 price implies roughly 29 million shares. The August 14 volume, reported at about 3.8 million shares and described as 10.8 percent of shares outstanding, implies roughly 35 million. Those do not reconcile. Share count is an input to every premium calculation on this page, and I do not currently have a version of it I would defend.
The June 30 NAV print is the trade. Everything else is noise around a stale photograph. On the sister vehicle at Robinhood the quarterly update landed about sixty days after quarter end, which would put a Fundrise June 30 disclosure somewhere in the back half of August. That is now. Set the alert, pull the filing, then do the arithmetic. Not before.
CBRS: The Benzinga Miss Is The Warrant Footnote Again
Short answer to the question in my notes: the revenue miss is almost certainly not a business miss. It is the customer warrant amortization, and we wrote the mechanic up on August 17.
| Line | Q2 2026 | Comment |
|---|---|---|
| Core revenue | $209.9M | Up 103% year over year |
| Reported sales, as cited by Benzinga | $180.11M | Against $194.20M consensus |
| Gap between the two | ~$29.8M | Author's arithmetic, not a company-disclosed line |
| Full year 2026 core revenue guide | $880M to $890M | Raised from $855M to $865M |
| Q3 revenue guide | ~$215M | Consensus ~$216M |
| Adjusted gross margin | 40.6% | Down from 46.5% sequentially |
Source: Benzinga, Stocktwits, and Investing.com coverage of the August 12, 2026 Q2 report. The reconciliation below has not been checked line by line against the Form 10-Q. EDGAR was reachable only indirectly during this session.
Cerebras carried roughly $1.128 billion of customer warrant assets as of June 30, arising from warrants issued to OpenAI as part of the commercial agreement, amortizing against revenue through October 2031. Under ASC 606 that amortization is a reduction of reported revenue, not an expense line. A roughly $29.8 million wedge between core revenue and reported sales in a single quarter is the right order of magnitude for that schedule.
Order of magnitude is not confirmation. Pull the 10-Q and tie the number to the disclosed amortization line before quoting me on it. What I will state without hedging is the structural point: as long as your data feed compares management's core revenue guidance to a GAAP consensus, this company will manufacture a phantom miss every quarter for another five years.
The Wire Services Cannot Agree On The EPS Either
One outlet reported adjusted earnings per share of negative $2.98 against an estimate of negative $0.18, framed as a large miss. Another reported an adjusted loss of $0.05 against the same negative $0.18 estimate, framed as a beat. Both were published within a week of each other about the same quarter.
Those cannot both be adjusted figures. The likeliest explanation is that one is GAAP, carrying post-IPO stock compensation, and got mislabeled somewhere in the chain. I have not resolved it. If you are trading this name off a screener, understand that the screener is being fed one of those two numbers and you do not know which.
What Actually Deserved The Selloff
Adjusted gross margin fell roughly five points sequentially, to 40.6 percent from 46.5 percent. The CFO attributed it to higher costs to rent compute capacity the company had previously deployed with other customers. That is a real margin story on a company scaling into a very large contract, and it is a better reason to sell the stock than a revenue line that was never going to print where consensus expected.
The CS-4 itself, launched at Supernova on August 18, is three WSE-3 Turbo engines per system on TSMC five nanometer, claimed at six times CS-3 system performance, ten times throughput per watt, up to thirty times tokens per second versus GPU alternatives, and fifty percent fewer components. Shipping this quarter, CS-5 targeted for 2027, 600 megawatts of capacity by end of 2027. The stock fell 12.69 percent on the day, then extended lower, roughly sixteen percent on the week. Needham and Rosenblatt reiterated Buy at $300, UBS at $330, Mizuho trimmed to $300 from $310. ARK bought about 35,000 shares.
XE: Right Correction, Wrong Timing, And Read The Qualifiers
The ticker correction holds. XE is X-Energy (Nasdaq: XE), the small modular reactor developer that listed in late April 2026 with about $1.1 billion of net IPO proceeds. It is not Expand Energy. Any aggregator still serving natural gas fundamentals under that symbol is wrong.
The award timing in my notes was off. The Department of Energy funding was not fresh news on August 18. CEO Clay Sell disclosed it on the Q2 earnings call on August 13, describing a formal notification received the prior day. By the 18th it was a week old and already in the price.
More to the point, read what the award actually is.
| Headline | What the transcript says |
|---|---|
| "$1 billion DOE award" | Up to an additional $1 billion on the ARDP cooperative agreement |
| Free federal money | Subject to the same 50:50 cost share as the original award. X-Energy matches it. |
| Booked | "Expected to be obligated" through the normal contractual process. Not obligated yet. |
| Total program | Would take DOE cost share contribution up to $2.115 billion |
Source: X-Energy Q2 2026 earnings call transcript, August 13, 2026, as published by Benzinga and others, with DOE confirmation reported by Reuters the same day. Transcript text, not a filed document.
The rest of the quarter: total revenues and grant income of $54.6 million, up 154 percent, driven by ARDP engineering work. Liquidity around $1.9 billion. TX-1 fuel fabrication about 80 percent vertically complete, holding the first new NRC Part 70 fuel fabrication license in over fifty years. Management points to 414 million fully diluted shares as the right denominator, comprising 280 million Class A, 119 million Class B, and 15 million from options and restricted stock.
So a 7.82 percent decline on August 18 is not the market ignoring a billion dollars. It is a company whose first revenue-generating reactor is targeted for the early 2030s, marked down alongside everything else with that maturity date, one week after a conditional cost-shared grant that its own CEO called its largest current revenue source.
Watchlist Ticker Map, With Verification Status
| Ticker | Maps to | Stale mapping to kill | Verified this session |
|---|---|---|---|
| XE | X-Energy Inc (Nasdaq) | Expand Energy | Yes, via Q2 call and Reuters |
| RVI | Robinhood Ventures Fund I (NYSE) | Retail Value Inc, the shopping center REIT | Yes, prior desk work |
| VCX | Fundrise Innovation Fund (NYSE) | None known | Yes, via 8-K coverage and issuer release |
| FLY | Firefly Aerospace | Fly Leasing, delisted aircraft lessor | No. Not checked this session. |
| CBRS | Cerebras Systems (Nasdaq) | None known | Yes |
One note on VCX that is worth carrying forward. Yahoo currently labels the entity "Fundrise Growth Tech Fund, LLC" on the quote page while the fund markets itself as the Fundrise Innovation Fund. That is a naming inconsistency, not necessarily a stale-ticker problem, but it is exactly the sort of thing that turns into a wrong holdings table three months from now.
Tonight: Wolfspeed
Wolfspeed (NYSE: WOLF) reports fiscal Q4 and full year 2026 today at 5:00 pm Eastern. The stock fell about ten percent into the print and sits near $31.46.
| Item | Figure |
|---|---|
| Consensus revenue, June quarter | ~$223.6M |
| Consensus EPS | -$0.52 |
| Prior quarter actual revenue | $150.2M, a 28% miss versus forecast |
| Company Q4 guidance given in May | $140M to $160M |
| Prior quarter stock reaction | Rose 7.26% on a 28% revenue miss |
| Consensus rating and mean target | Neutral, ~$30 |
Source: Investing.com earnings preview and quote data, August 18, 2026, and Wolfspeed's Q3 FY2026 guidance. Note the conflict: consensus revenue of $223.6M sits above the company's own $140M to $160M guidance range. I have not reconciled whether guidance was revised in the interim. Do not trade this spread without checking.
That last row is the whole setup. A stock that rallied seven percent on a twenty-eight percent revenue miss last quarter is not trading on the print. It is trading on the balance sheet repair, and the balance sheet is where I would look tonight: cash, debt retired, and whether the AI data center revenue line is still compounding after the March convertible and equity placement.
Bull Case On The Bucket
The move was rates, not fundamentals. Two year and ten year yields barely budged. Nothing in the demand picture for compute, power, or interconnect changed between Monday and Tuesday.
The operating news was mostly good. Cerebras raised full year guidance and margin guidance and shipped a product claiming six times the prior generation. X-Energy grew revenue 154 percent and got a federal cost share increase. Nebius is raising capital because it has somewhere to deploy it.
Forced selling creates dislocations. Convertible hedge unwinds, lockup supply, and leveraged ETF rebalancing all produce prints that are mechanical rather than informational. Those are the prints worth buying into if you know which is which.
Bear Case On The Bucket
A higher long end is a permanent haircut on 2030 cash flow. If the thirty year settles above five percent rather than visiting it, every discounted cash flow underwriting this complex needs rebuilding, and the answer comes out lower.
This complex funds itself with paper that gets more expensive as rates rise. Nebius, Wolfspeed, and Keel have all come to the convertible market in the last six months. Rising rates and falling equity prices raise the cost of that instrument from both directions at once.
The information quality is degrading, not improving. A ticker that still resolves to a natural gas producer. Two incompatible EPS figures for the same quarter. A NAV number that has already been publicly corrected once. Share counts that will not reconcile. When the data layer is this noisy, position sizing has to absorb the error, and most people are not sizing for it.
The SharkWater Take
I am not putting on any of these three trades today, and I want to be specific about why, because "wait for more information" is usually a cop-out and here it is not.
On NBIS, the coupon and conversion premium are the trade. A 4.5 billion dollar convert priced at one and a quarter percent with a forty percent conversion premium is a triumph. The same size priced at four percent with a fifteen percent premium is a distress signal. Those two outcomes lead to opposite positions in the equity, and the release explicitly says both numbers get set at pricing. Anyone with a view on Nebius this afternoon has a view on a document they have not read.
On VCX, the June 30 NAV is the anchor and it does not exist publicly yet. Trading an eleven-percent-or-a-hundred-and-eleven-percent premium is not a trade, it is a coin flip with a spread attached. And I cannot even fix the denominator right now.
On CBRS, I am confident enough in the warrant explanation to say the revenue miss is an artifact. I am not confident enough to publish the reconciliation, because I could not get EDGAR to answer cleanly today and I am not going to source a 10-Q claim to a wire story. The margin compression is the real issue anyway, and that number I do trust.
Here is what I actually think about the bucket. The rate move is the event and the single stock headlines are decoration. The long end going to a nineteen year high is a structural repricing of everything whose payoff sits past 2030, and this group is nothing but assets whose payoff sits past 2030. If you want to be long this complex, you are making a rates call whether you know it or not, and you should say so out loud before you size it.
The names I would look at first when I do act are the ones where the cash flow is nearer and the counterparty is diversified. That is the same conclusion I reached on Marvell two days ago and it survived contact with a bond selloff, which is more than the backlog stories can say.
Execution Notes
Structure: No position recommended and no option ladder published in this post. I did not pull a live chain on any of these five names. Any premium I printed here would be a Black-Scholes guess and this is not a post where a guess is good enough.
NBIS trigger: the pricing release and the corresponding 8-K or 6-K carrying coupon, initial conversion rate, conversion premium, and accretion schedule. Expected within days of the August 19 launch.
VCX trigger: the June 30, 2026 NAV and the corresponding portfolio schedule. On the comparable vehicle the lag ran about sixty days from quarter end, which puts this in the back half of August.
CBRS trigger: the Q2 Form 10-Q, specifically the customer warrant asset amortization line, tied against the $209.9 million core to $180.11 million reported gap.
XE trigger: formal obligation of the incremental DOE funds to the ARDP agreement, and the announcement of the utility agreement for the next gigawatt project that management said was being finalized.
Invalidation on the rates thesis: the long bond retracing back under roughly five percent with the two year unchanged would say this was a positioning event, not a repricing, and most of this post ages badly within a month.
Data Gaps In This Post
Publishing these rather than burying them, because a reader who does not know what I could not verify cannot weight what I wrote.
- SEC EDGAR was reachable only indirectly for this entire session. Every filing referenced here came through a secondary carrier. Spot-check the primary documents before acting on any figure in this post.
- The CBRS core-to-reported revenue reconciliation is arithmetic plus a disclosed mechanic. It is not tied to a line item in the 10-Q.
- The VCX NAV of $18.97 is secondhand and has a competing figure of $18.43 in circulation. The share count does not reconcile across sources.
- The two conflicting CBRS adjusted EPS figures, negative $2.98 and negative $0.05, are unresolved.
- Wolfspeed consensus revenue of $223.6 million sits above the company's own May guidance range of $140 to $160 million. Unreconciled.
- The 5.33 percent thirty year print is a secondary-source intraday high, not an H.15 close.
- FLY mapping to Firefly Aerospace is carried forward from prior notes and was not re-verified.
- All intraday percentage moves cited for August 18 and 19 are as-quoted and unsettled.
Fair winds and following seas.
Disclaimer: This post is for informational and educational purposes only and does not constitute investment advice, a recommendation, or an offer to buy or sell any security. Prices, yields, and percentage moves cited are approximate, drawn from secondary market data providers and press reporting dated August 13 through August 19, 2026, and in several cases reflect unsettled intraday quotes. SEC EDGAR was accessible only indirectly during the preparation of this post, and figures attributed to filings arrived through secondary carriers rather than primary retrieval. Specific unverified items are enumerated in the Data Gaps section above and should be treated as unconfirmed. Reconciliations described as the author's arithmetic are estimates derived from public disclosures, not company-reported metrics. The Nebius convertible offering described here is proposed, its economic terms are undetermined as of publication, and it may not be completed. Forward-looking guidance and program funding expectations are management's or the relevant agency's, not the author's. All securities discussed are volatile and several are recent public listings with limited trading history. Verify every figure against primary filings before acting. Past performance does not guarantee future results. Consult a licensed financial professional regarding your specific situation.
No comments:
Post a Comment