Pinned Tweet
One thing I’ve noticed from trading more actively is how quickly market narratives turn over.
Themes used to have real staying power. Now the market can move on within weeks. I’m not sure how much of this is driven by AI, but the pattern is hard to miss: software gets sold at a record pace, attention shifts to high-short-interest names, and everyone is already searching for the next bottleneck or story.
According to J.P. Morgan, weekly retail outflows from software and services recently hit their highest level since the data series began. At the same time, social-media interest was clustering around high short interest names like GRND, RH, NBIS, and CAKE, alongside memory, AI infrastructure, and space names like SNDK, MU, VRT, and RKLB.
It doesn’t feel like capital is rotating neatly from one theme to the next. It feels more like several stories are lighting up and burning out at once. Sometimes the entire market seems to trade like a meme. Positions don’t stay in one place for long, while social media and AI have shortened the gap between discovering a narrative, spreading it, and exhausting it.
But a discarded theme bouncing back doesn’t mean it has reclaimed the center of the market. If earnings estimates keep rising and capital continues to flow in, it may still be a trending asset. If the move is driven mainly by short covering, an event, or a sudden spike in attention, it is probably still a trade.
SpaceX holding up despite a large lockup expiry may be another example. Maybe less stock was sold than expected. Or maybe the market had already moved past the unlock and started pricing in the next story.
In a market where narratives have shorter shelf lives, finding a good story is no longer enough. The harder question is whether it is still a trend, or whether it has already become just another trade.
The Shortening Shelf Life of Market Narratives
Looking at this week's @QFEX earnings calendar, there are quite a few important dates lined up. Hood, MSFT, Google, and others.
qfex.com/earnings-calendar
GPU Prices Are Hitting Record Highs. So Is the Cost of Financing AI.
Two indicators that seem contradictory are rising at the same time across the AI infrastructure market.
According to Silicon Data, on-demand GPU rental prices across neocloud providers have reached record highs:
- B300: $7.35 per GPU-hour
- B200: $6.29
- H200: $3.51
- H100: $2.88
- A100: $1.55
Even the A100 index, which had recently shown some weakness, appears to be turning higher. At least in the spot-like GPU rental market, demand for AI compute remains strong.
At the same time, 5 year CDS spreads across hyperscalers, semiconductor companies, and other AI linked names are widening rapidly. Oracle’s spread has climbed to roughly 300 basis points, SpaceX is above 200 basis points, and names such as Broadcom and Meta have also moved higher.
A CDS is essentially the market price of insuring against default on a company’s debt. A rising spread does not mean default is imminent, but it does indicate that investors are assigning a higher cost and greater risk to lending that company money.
The next AI bottleneck may therefore be less about securing GPUs and more about financing them on sustainable terms.
Higher GPU rental prices create more revenue potential for data-center operators. But expanding supply still requires enormous upfront investment in accelerators, power, land, networking, and cooling infrastructure.
As credit spreads widen, the return generated by those GPUs must clear an increasingly demanding hurdle: interest expenses, electricity costs, depreciation, and other operating costs.
This means GPU ownership alone is no longer enough to evaluate an AI infrastructure business.
Rental price × utilization − electricity − depreciation − cost of capital
The gap is likely to widen between operators that can keep this equation positive and those that must continually borrow at expensive rates to expand their GPU fleets.
AI compute remains scarce. But turning that scarcity into sustainable cash flow will increasingly depend on one critical variable: the price of capital.
Sources:
xitter.cf/ParadisLabs/status/211…
xitter.cf/silicondata/status/211…
Cl over $100
Many market commentators expect oil prices to fall back to prewar levels once the conflict in the Middle East ends and the Strait of Hormuz reopens. The 1990 Gulf War is often cited as a precedent. But today’s energy market is different in several important ways.
During earlier crises, major producers had enough spare capacity to replace much of the disrupted supply. Some spare capacity still exists today, but it may not be large enough or accessible enough to offset a serious disruption around the Strait. Much of the available capacity is concentrated in the same region, which limits its value when shipping routes themselves are at risk.
The United States shale industry may also be a less powerful buffer than it once was. Production remains strong, but companies are now more focused on financial discipline and shareholder returns. Even when prices rise, new supply may take time to reach the market.
The end of the conflict would certainly remove part of the war premium from oil prices. However, shipping, insurance, and security costs are unlikely to normalize immediately. Buyers may also need to rebuild inventories, while the possibility of another disruption could remain in the market for some time.
Oil prices could therefore fall sharply after a ceasefire without returning all the way to their previous level. The Gulf War remains a useful historical reference, but it should not be treated as a direct roadmap. The structure of the market has changed, and the price floor after the conflict may be higher than many investors currently expect.
Nebius Is Turning Idle GPU Capacity Into a Real Time Market
The easiest way to understand Nebius is to think of it as operating a GPU hotel
- Longterm contracts: Customers reserve large blocks of rooms for an extended period.
- On-demand: Customers pay a fixed rate for reliable access whenever they need it.
- Spot GPUs: Customers rent temporarily idle capacity at a discount, but Nebius can reclaim it with 60 seconds’ notice.
Historically, spot capacity was always priced below on-demand capacity because it could be interrupted at any time.
That model is now changing in three steps:
1. October 1: Nebius increased on demand prices for H100, H200, B200, and B300 GPUs by roughly 17–21%.
2. October 8: It introduced dynamic spot pricing, with prices recalculated as often as every 15 minutes based on available capacity and customer demand.
3. November 2: Nebius will remove the rule requiring spot prices to remain below on-demand rates.
This means customers may soon pay more for interruptible GPUs than for regular on-demand capacity if they urgently need compute and availability is limited.
Why this matters
1. Nebius can monetize idle GPUs more efficiently
Previously, spot prices could not exceed on-demand rates, regardless of how much urgent demand existed. Removing the ceiling allows Nebius to capture more revenue from the same hardware when demand spikes.
However, removing the cap does not automatically mean prices will rise above on-demand rates. That will only happen if real customer demand supports it.
2. GPU scarcity is becoming visible through pricing
According to @OrnnExchange, Nebius’s B200 spot price in us-central1 moved from approximately $0.99 per GPU-hour between October 14 and 18 to around $4.00–$4.70 over the following few days.
That suggests increasing demand for available capacity. But the price was still below Nebius’s B200 on-demand rate of $8.50 per GPU-hour.
So far, the data confirms rising spot demand, not that spot GPUs are already trading above on-demand prices.
3. GPU capacity is becoming easier to value as a financial asset
Historically, lenders struggled to estimate what a GPU cluster might earn after its long-term customer contracts expired.
A transparent spot price gives lenders a better way to estimate:
- The revenue potential of idle GPUs
- The collateral value of GPU infrastructure
- How much debt a data center can reasonably support
If GPU futures develop alongside this market:
- AI companies could hedge against future compute-price increases.
- Neocloud providers could hedge against falling rental rates.
- Lenders could better assess the risks of financing GPU data centers.
In other words, GPU time is beginning to resemble electricity or oil: a commodity with observable prices that can support collateral, financing, and derivatives.
Bottom line
Nebius has started selling idle GPU capacity through an auction like pricing system that can update every 15 minutes. By removing the price ceiling, it is moving toward a market where GPU time receives a real-time price based on immediate supply and demand.
Ornn is building the pricing indices and financial infrastructure intended to turn that compute capacity into something that can be valued, financed, and eventually hedged through futures.
This policy change does not yet prove that Nebius’s financial performance has improved. In the upcoming earnings report, the key indicators will be utilization, realized revenue per GPU-hour, and management’s comments on spot demand.
The following quarter may be even more important because the removal of the spot-price ceiling does not take effect until November 2.
Sources:
nebius.com/blog/posts/introd…
xitter.cf/OrnnExchange/status/21…
$NBIS’s Problem Isn’t Demand. It’s Turning 5 GW Into Revenue.
$NBIS closed at $165.47 on October 16, down 25% over five sessions and 42% from its recent high. The key issue is not whether AI demand exists, but how quickly Nebius can convert contracted power into active, revenue generating capacity.
Nebius expects to end 2026 with 5 GW of contracted power. William Blair estimates only about 0.6 GW will be active this year, meaning nearly 90% of that capacity has yet to generate revenue. That gap represents both the upside and the main execution risk.
Demand is well supported:
- RPO as of June 30: $37.49B
- Microsoft: $17.4B base contract, potentially rising to $19.4B
- Meta: $12B of dedicated capacity plus up to $15B for eligible capacity Nebius cannot sell elsewhere
- Q2 revenue: $582M, up 454% YoY
- AI Cloud ARR: $3.0B
- AI Cloud adjusted EBITDA margin: 50%
Recent contracts generated more than $20M in annual contract value per MW, while management estimated a 22 month payback period for Q2 deals. Nebius argues that its full stack software and infrastructure model can produce more revenue per MW than simply renting GPUs.
But the buildout is extremely capital intensive. Q2 capex was $5.7B, while first-half purchases of property, equipment, and intangible assets reached $8.13B. Nebius had $8.04B in cash and expects more than $9B of customer prepayments in 2026, but it also raised roughly $2.8B by selling 12.7M shares through its ATM program.
The central question is whether Nebius can activate its 5 GW pipeline on schedule and convert $37.49B of RPO into revenue and cash flow without major cost overruns or excessive dilution.
The numbers to watch are active power, capex per MW, customer prepayments, and margins.
If you're looking for a good place to trade NBIS, consider @QFEX :
qfex.com/@hyeon__dev
Why @GodelTerminal chose only QFEX and Hyperliquid?
Excited to announce that QFEX is now live on @GodelTerminal.
→ 230+ stocks
→ One terminal
→ 24/7/365
Powered by QFEX
$NBIS’s Problem Isn’t Demand. It’s Turning 5 GW Into Revenue.
$NBIS closed at $165.47 on October 16, down 25% over five sessions and 42% from its recent high. The key issue is not whether AI demand exists, but how quickly Nebius can convert contracted power into active, revenue generating capacity.
Nebius expects to end 2026 with 5 GW of contracted power. William Blair estimates only about 0.6 GW will be active this year, meaning nearly 90% of that capacity has yet to generate revenue. That gap represents both the upside and the main execution risk.
Demand is well supported:
- RPO as of June 30: $37.49B
- Microsoft: $17.4B base contract, potentially rising to $19.4B
- Meta: $12B of dedicated capacity plus up to $15B for eligible capacity Nebius cannot sell elsewhere
- Q2 revenue: $582M, up 454% YoY
- AI Cloud ARR: $3.0B
- AI Cloud adjusted EBITDA margin: 50%
Recent contracts generated more than $20M in annual contract value per MW, while management estimated a 22 month payback period for Q2 deals. Nebius argues that its full stack software and infrastructure model can produce more revenue per MW than simply renting GPUs.
But the buildout is extremely capital intensive. Q2 capex was $5.7B, while first-half purchases of property, equipment, and intangible assets reached $8.13B. Nebius had $8.04B in cash and expects more than $9B of customer prepayments in 2026, but it also raised roughly $2.8B by selling 12.7M shares through its ATM program.
The central question is whether Nebius can activate its 5 GW pipeline on schedule and convert $37.49B of RPO into revenue and cash flow without major cost overruns or excessive dilution.
The numbers to watch are active power, capex per MW, customer prepayments, and margins.
If you're looking for a good place to trade NBIS, consider @QFEX :
qfex.com/@hyeon__dev
One of the reasons I've been adding more $NBIS is that Nebius is suppose to have 5 GW of contracted power by end of 2026 but only ~0.6 GW active meaning nearly 90% of that secured capacity has yet to come online.
Whats even better is that Nebius already has so much of the power needed for its next phase of growth with active capacity projected to more than triple to 2 GW by 2028 and reach 3.5 GW by 2030 without needing to secure additional power through 2028.
And I think bigger opportunity is how much revenue Nebius can generate from every MW as its full stack infrastructure and Token Factory move it beyond simply renting GPUs with older GPU prices already up ~30%, contract values exceeding $20M per MW and some deals paying back in under two years.
.@koolkrypto223, one of the key figures behind the $DRV narrative, has unstaked his holdings.
debank.com/profile/0xaf0fdd3…
The tokens will become unlocked on October 24.
KoolKrypto played a meaningful role in driving $DRV’s rally and building momentum. Given his influence on the narrative, the unstaking could still be a noteworthy signal.
Source : t.me/Web3LearningWithInger/2…
Based on the latest closing prices, the equal-weighted basket is now down roughly 20.5% since September 8, rather than 18.6%. If the exposure was maintained through leveraged call options, the losses could be substantially larger depending on the strikes and expirations.
However, this does not mean the entire fund is down that much. Position weights, hedges, cash, and any sales remain unknown, while some short-dated calls reportedly expired profitably. The November 16 13F may confirm the September 30 positions, but it will not reveal strikes, expirations, premiums paid, or the fund’s exact P&L.
Leopold Aschenbrenner might genuinely be the unluckiest motherfucker alive.
After getting absolutely destroyed in July and being forced to liquidate much of his AI portfolio, he reportedly came back swinging in early September, buying hundreds of millions of dollars worth of options premium.
Look at how those stocks have performed since September 8:
$CRWV: -37.4%
$SKHY: -25.8%
$SNDK: -23.8%
$DRAM: -22.9%
$BE: -18.1%
$INTC: -9.4%
$AMD: +9.1%
Six out of seven are underwater, with an average decline of roughly 18.6%.
And remember, we're talking about someone who reportedly rebuilt exposure through leveraged options, not just common shares.
Imagine surviving a massive forced liquidation, watching your favorite stocks recover, deciding it's finally safe to get back in, and then getting hit by ANOTHER brutal selloff barely a month later.
The worst part about options is that you can be absolutely right about the long-term AI infrastructure thesis and still lose a fortune because your timing was wrong.
I genuinely feel for the guy.
The market has absolutely no mercy.
Today I came across an interesting theory that says, “the less money a startup makes, the more aggressively its founder describes the company.”
And the post I happened to see was one by @LC.
- lens today remains 100% focused on the same core vision we had in 2022.
- Let's build a new internet together.
=> It's clear that they aren't making any money.
Strong support for this theory
lens today remains 100% focused on the same core vision we had in 2022.
it all started with this:
Lens Protocol Open Letter
Social media platforms have created communities and allowed voices to be heard that might otherwise have been silenced. They have given us the power to connect across continents and oceans.
But these platforms are centralized. They own our data. They control our monetization. They dictate our algorithmic distribution.
It’s time to build a better way.
The Lens Protocol is a decentralized, open-source social graph that returns control of information and links to the people. It allows content creators to own their digital roots and enables individuals to fully possess their connections with their communities.
By signing this letter, you support a future where:
1. Creators own their content and distribution.
2. Users own their social graph and data.
3. Developers have open access to build innovative social experiences without the risk of being de-platformed.
Let's build a new internet together.
Onchain options have been getting a lot more attention lately, and DeFiLlama does show a clear rise in TVL and trading activity. But the recent jump should not be interpreted simply as a wave of new capital entering the market. A major driver was Derive’s migration to V3, which changed both the protocol architecture and how its assets and activity are indexed.
The distinction between premium volume and notional volume is important here.
- Premium volume measures the gross value of option premiums exchanged. If a trader buys one BTC call for $500, the trade adds roughly $500 to premium volume.
- Notional volume measures the underlying exposure: if BTC is trading at $60,000, the same contract adds roughly $60,000 to notional volume.
Premium volume is therefore closer to the amount paid for options and the platform’s fee generating activity, while notional volume reflects the scale of exposure traded. Neither represents net new capital on its own.
Derive V3 was more than a routine software upgrade. V2 balances and positions were imported into a V3 genesis state, bridge funds were moved to Ethereum L1, and Derive Chain was wound down. Collateral previously distributed across multiple chains, bridges and vaults is now recognized through a single Ethereum SPOT_VAULT. DeFiLlama also switched to the V3 address and adapter structure, so the chart briefly lost visibility into some assets before recovering as the migrated collateral was fully recognized. Much of the TVL jump is therefore better understood as asset relocation and re-indexing rather than entirely new deposits.
The volume data shows a similar gap. In the fixed seven-day windows around the cutover, September 30 to October 6 versus October 7 to 13 UTC, premium volume increased by roughly 69%, while notional volume fell by about 30%. This suggests that total underlying exposure did not suddenly explode. Higher option prices or implied volatility, a different strike and expiry mix, multi leg RFQs, and repeated position rolls can all increase premium volume. Multi leg strategies also count each option leg as gross volume even when the legs form a single combined trade.
The supporting indicators were much calmer: fees increased by roughly 7% and unique taker wallets by about 9% over the same period. That still points to improving engagement and growing interest in onchain options.
The options narrative is real, and Derive is seeing genuine activity. Still, the recent step change is best understood as a combination of organic growth and the accounting effects of the V3 migration. Premium volume, notional volume, fees, unique users, open interest and net deposits should be tracked separately before concluding that the market itself has expanded as dramatically as the headline charts suggest.
defillama.com/protocols/opti…
What You Can Gain from @QFEX This Week
Loss Refund Program
- Connect your X account and apply for the refund program. (qfex.com/rewards-hub)
- Check your dashboard to see how well you’ve met the weekly taker volume requirements.
- Keep your weekly P&L in the negative (though, of course, it’s great if you can make a profit).
- If you win, you'll have the chance to receive a refund of up to $2,000.
Check out our new referral program.
- From now on, please use referral links instead of the cumbersome invitation code system.
- Earn up to 20% from direct referrals, plus 4% and 2% from indirect referrals. Referred traders get 10% off on fees, and rewards are settled daily.
- docs.qfex.com/qfex/referral
If you have any questions, feel free to DM me anytime.
qfex.com/@hyeon__dev
BTW ran a sale on @KaitoAI in January at an average price of $0.008 ($80M FDV).
Terms were 50% unlocked at TGE with the remaining 50% locked for 3 months post-TGE, meaning sale participants’ tokens were fully unlocked around June.
At the current price, that’s roughly a 280–290x return.
Is anyone still here?
The @bitwayofficial Public Sale is now live on our Capital Launchpad!
Key information on the sale:
Valuation: $80M FDV
Previous Valuation: $100M
Investors include: YZi Labs, Hashkey Capital, KR1, Symbolic Capital, Continue Capital
Target Raise: $1M
Maximum: 3% of total supply
Vesting: 50% at TGE, + 50% at TGE + 3 Months
Min/Max Size: $1,000-$250,000
Estimated TGE: January 2026
Bitway is the internet capital gateway that connects on-chain liquidity with global opportunities.
At its core, Bitway offers a suite of products, including Bitway Earn, an on-chain wealth management platform, and Bitway Chain, a Bitcoin-compatible PoS Layer 1 built to support native BTC financing and enterprise-grade applications.
The sale is open to all jurisdictions permitted under our Terms of Use.
Head to our Capital Launchpad to get involved!
TL;DR:
- At the TOKEN2049 event in Singapore, there were team, Apex Accelerator and Stone Venture, that posed as a security event to trick people into installing remote control software.
- They are reaching out to cryptocurrency professionals and various other people under the pretext of business collaboration.
- When my friend, who had been on a call with them, sent a link related to the event that had become a hot topic, they kicked him out of the group chat without any explanation.
Please stay alert to news like this and stay safe.
DPRK going ham these days. These guys (Apex Accelerator) kept asking for a call so I had one booked this week. Sent them what this was about and they just deleted the group.
Special thanks to the goat @hyeon__dev for raising this in his TG channel.
