DEGENARIO retweeted
Trion update for today:
Our beta testers sent us a lot of mobile feedback, so today is making Trion feel great on a phone.
Shipped:
• New Charts / Trade tabs, so the chart gets the whole screen and the order ticket gets its own
• The price chart now fills about three quarters of the screen on an iPhone
• Faster mobile loads: the options page ships half the code, and the perp chart opens about 30% faster
• Deposit and Withdraw right from Portfolio, plus a one-tap fix when your wallet is on the wrong network
In progress:
• The Trion iOS app.
The CMPT-USD perpetual is still in closed beta. Thanks to everyone testing and sending feedback. Keep it coming.
DEGENARIO retweeted
GM everyone,
The Trion Discord is now open.
Signed up for the beta or have any questions? Join the server. Invites go out by DM from our only official account, Discord username: trion.market
We'll never ask for keys, seed phrases, passwords or funds. Anyone else is a scammer.
discord.gg/htgufeGZHZ
DEGENARIO retweeted
Trion Trading Competition starts next week.
$3,000 prize pool for the top 5 traders on the CMPT-USD perp beta:
1st $1,200 · 2nd $750 · 3rd $500 · 4th $300 · 5th $250
You can even hook up your AI to trade for you.
DEGENARIO retweeted
Compute is the commodity that world runs on now. Big Tech companies are spending over $700B this year, most of it on AI data centers, and CME is getting ready to list futures on the GPU-hour.
A potential problem is when price goes up. The H100-hour on our index went from $2.06 in January to $2.68 in July. That's 30%. If your whole business runs on GPUs, that's your margin getting eaten, and there's been basically nothing you could do about it... until now.
So we're building the market for it. On Trion one contract is one H100 GPU-hour. Run a GPU for a month and that's equal to 730 contracts.
This our compute market you can lock your bill, put a cap on it with options, or just trade it if you have a directional view. If you own GPUs you can do the opposite and lock in what you earn.
The part we're most excited about is agents. An agent knows exactly how much compute it's burning, all day, every day.
This allows agents to hedge their own compute without anyone babysitting them.
The chart shows what a 20% move does under each approach.
The CMPT-USD perp is still in testing. We don't think people realize yet how big that gets over the next couple of years.
DEGENARIO retweeted
Trion Agent Kit now live.
Let your code or AI trade inside limits you set. Agent keys can't withdraw. Revoke anytime.
TypeScript SDK · Python SDK · MCP connector for Claude, ChatGPT and more.
Docs: trion.market/docs
DEGENARIO retweeted
I will be purchasing the first GPU tomorrow.
NVIDIA RTX PRO 6000 Blackwell Workstation Edition.
Exciting to get start our GPU fleet. We will continue to grow it with fees collected from trading.
DEGENARIO retweeted
Don't forget to sign up for beta.
I need you guys to abuse this so we can fix anything that comes up during testing.
That is just the nature of things. Testing underpressure and usage is the best way to improve and harden the exchange.
trion.market/beta/
DEGENARIO retweeted
Compute is the most valuable resource in the world right now, and I believe it only goes up from here.
Every model, every agent, every AI product runs on it. Demand is exploding. Supply is selling out everywhere.
Here's my thesis, and our plan for Trion.
Demand is exploding
• The biggest cloud providers are projected to raise capex 98% in 2026, then another 50% in 2027 (TrendForce).
• Goldman Sachs estimates more than $5 trillion of cumulative AI capex from 2025 to 2030 (cited by BlackRock).
• AI tools deliver value an order of magnitude greater than their cost (SemiAnalysis). At that return, demand doesn't blink at price.
Supply is sold out
• By spring 2026, on-demand GPU capacity was sold out across every GPU type, and capacity coming online through September was booked in advance (SemiAnalysis).
• H100 contracts signed 2-3 years ago are renewing at the same rates, some for 4 more years.
• On AWS, buyers were fighting to pay $14 an hour per GPU for B200 spot capacity.
The inputs are repricing
• Server DRAM contract prices rose 64% in 2H25, with a further ~270% jump expected in 2026 (TrendForce).
• HBM could climb another 70-140% in 2027.
• Every GPU is built on memory. When memory reprices, compute follows.
The price has turned
A year ago, the consensus was that GPU rental prices would keep falling. The opposite happened. H100 1-year rental contracts climbed almost 40%, from $1.70 an hour in October 2025 to $2.35 by March 2026 (SemiAnalysis).
My thesis
For the next two years, compute gets scarcer and more expensive. Demand is growing faster than fabs, memory lines, power and data centers can be built, and new capacity is spoken for before it arrives.
Compute is the defining commodity of this era, and every great commodity gets a great market. CME plans to list GPU futures on October 5, pending CFTC approval. BlackRock says compute is becoming "a distinct, large, and increasingly investable economic resource."
Wall Street is starting to price compute. Trion already trades it.
Live on Trion today
• CMPT: a public price for one H100 GPU-hour, from six named venues
• Calls and puts on CMPT, on Robinhood Chain
• Perps are next.
The plan: own the compute itself
• The $TRION tax buys GPUs
• We rent them out
• Holders get discounted access
• Rental income buys more GPUs
The fleet grows one GPU at a time, and every GPU helps buy the next.
The endgame: GPU-backed perps
Perpetuals on the price of compute, backed by hardware we actually own.
The world is going to run on compute. I'm building Trion to be where it gets priced, traded and owned.
DEGENARIO retweeted
the woman in both these pictures is my mother. 😢
one is before cancer. the other is what this disease has done to her.
my mother, Babita Singh, has already fought stage 4 colon cancer once. unfortunately, we are fighting cancer again.
over the years, we’ve spent ₹38L+ ($38000) on her treatment, pooling everything we could from family and relatives. we’ve now reached a point where we’ve exhausted almost everything we had.
[attaching bills + medicines]
we need to raise another ₹10 lakh( $10000) to keep her treatment going. the fundraiser covers 5 lac part of this, and we’re trying to arrange the rest however we can.
if you can help with any amount, i’ll be forever grateful.
👉🏻 fundraiser: (gpay and all, only 5 lac)
impactguru.com/fundraiser/he…
👉🏻 Fundraising through crypto (only $5k)
solana: [ HwJMgrHnGnDLvWhD8wVPMh6FVCmsHiGJrVRywrLp1hsH
ethereum: [ 0x99Fd581d47213b00b035b95b32deD0cE241902B2 ]
if you can’t donate, please repost this. it might reach someone who can.
feel free to dm me to ask any for any proof, thank you ❤️
DEGENARIO retweeted
READ THIS
AUTOCOMMIT turns every ACMT trade into an action inside a shared on-chain program. A buy adds a new instruction, while a sell edits part of the existing sequence. Over time, these actions build a complete program that can eventually be sealed and stored as a finished build. GitHub mirrors this process so anyone can see and verify how the program changed. In simple terms, people trade the token, and the market itself writes the code
DEGENARIO retweeted
这几天在香港忙到脚后跟打着后脑勺。
昨晚通宵写脚本,白天连轴转录课程,刚才整个人已经累傻掉了。
晚上坐轮渡来澳门,刚上船,立刻昏昏沉沉睡了过去。
醒来之后,发现 $BEM 起飞了,原来是跑通了Linux……
我基于 @BNBCHAIN ,实现了历史上第一次在L1主网上跑Linux。这是第一个每条指令都由 L1 主网交易执行、全部状态都存在 L1 上的 Linux。
此前这个领域公认的先行者是 Cartesi:它的 RISC-V 虚拟机能启动完整的 Linux,但 Linux 是在链下节点上跑的,以太坊 L1 只负责结算,只有当参与者对结果有争议时,才把争议的那一步拿到链上重算。正常情况下,L1 从头到尾没有执行过一条 Linux 指令。它是 L2 框架,靠欺诈证明和至少一个诚实的挑战者。
而TapeOut做的是真正的,Linux 的每一条指令都在 BNBChain 主网的交易里由验证者算出来,并且拥有32 个寄存器、全部 CSR以及16 MB 内存的每一页都在BNBChain主网上,没有任何链下执行。也就是说Cartesi 上让 L1 能验证 Linux,我们让BNBChain 真实的链上运行完整的 Linux。
它是由TapeOut 的电路+容器组成,第一次组成了一台完整的链上计算机。一片电路(#4246)开通了容器,容器成了这台机器的主人,谁持有这片电路的 NFT,谁就是它的管理员。CPU 是一个 RISC-V 元件,内存是 16 MB、按 1 KB 分页的 16,384 个页合约,写时复制。主板合约用交易跑指令,把改过的内存页写回链上,把串口输出发成事件。
装的是 kernelorg 主线 Linux 6.5.12,它从复位向量 0x80000000 上电,在主网上安拍跑完整个内核启动,直到 root 登录:一共44,988 拍,9,000 余笔交易(花了我好多bnb啊,有点肉疼,另外我还找bnbchain的小伙伴申请了10个tBNB做的测试网测试,感谢 @0xSandy1913 帮助协调测试网的tBNB)。
另外呢,任何人都可以用它。打开 tapeout.net/linux.html,像用真实终端一样直接敲命令:uname -a 会告诉你这是 riscv32 上的 Linux 6.5.12;factor 1000000007 在链上做一次质因数分解;echo 写一个文件,这个文件就住在主网上某一页合约的字节码里,你能在 BscScan 上看见它,谁使用谁自己付 gas(我可支付不起这么多人测试,哈哈)。
这件事说明: @BNBCHAIN 不只是一个公共账本,它可以基于TapeOut承载可验证的生产工具,一台计算机是所有生产工具的底座。
接下来:用门级 RISC-V 核做这台机器的裁判,让每一条指令都有晶体管级的定义;让 Linux 能直接读链上数据;让更多电路拥有自己的不同类型的机器。BNB Chain 的延展性比大多数人想象的大得多,TapeOut在为此全力以赴。
测试用例:tapeout.net/linux.html
@cz_binance @heyibinance @BNBCHAIN @fourdotmemezh
我基于 @BNBCHAIN ,实现了历史上第一次在L1主网上跑Linux。这是第一个每条指令都由 L1 主网交易执行、全部状态都存在 L1 上的 Linux。
此前这个领域公认的先行者是 Cartesi:它的 RISC-V 虚拟机能启动完整的 Linux,但 Linux 是在链下节点上跑的,以太坊 L1 只负责结算,只有当参与者对结果有争议时,才把争议的那一步拿到链上重算。正常情况下,L1 从头到尾没有执行过一条 Linux 指令。它是 L2 框架,靠欺诈证明和至少一个诚实的挑战者。
而TapeOut做的是真正的,Linux 的每一条指令都在 BNBChain 主网的交易里由验证者算出来,并且拥有32 个寄存器、全部 CSR以及16 MB 内存的每一页都在BNBChain主网上,没有任何链下执行。也就是说Cartesi 上让 L1 能验证 Linux,我们让BNBChain 真实的链上运行完整的 Linux。
它是由TapeOut 的电路+容器组成,第一次组成了一台完整的链上计算机。一片电路(#4246)开通了容器,容器成了这台机器的主人,谁持有这片电路的 NFT,谁就是它的管理员。CPU 是一个 RISC-V 元件,内存是 16 MB、按 1 KB 分页的 16,384 个页合约,写时复制。主板合约用交易跑指令,把改过的内存页写回链上,把串口输出发成事件。
装的是 kernelorg 主线 Linux 6.5.12,它从复位向量 0x80000000 上电,在主网上安拍跑完整个内核启动,直到 root 登录:一共44,988 拍,9,000 余笔交易(花了我好多bnb啊,有点肉疼,另外我还找bnbchain的小伙伴申请了10个tBNB做的测试网测试,感谢 @0xSandy1913 帮助协调测试网的tBNB)。
另外呢,任何人都可以用它。打开 tapeout.net/linux.html,像用真实终端一样直接敲命令:uname -a 会告诉你这是 riscv32 上的 Linux 6.5.12;factor 1000000007 在链上做一次质因数分解;echo 写一个文件,这个文件就住在主网上某一页合约的字节码里,你能在 BscScan 上看见它,谁使用谁自己付 gas(我可支付不起这么多人测试,哈哈)。
这件事说明: @BNBCHAIN 不只是一个公共账本,它可以基于TapeOut承载可验证的生产工具,一台计算机是所有生产工具的底座。
接下来:用门级 RISC-V 核做这台机器的裁判,让每一条指令都有晶体管级的定义;让 Linux 能直接读链上数据;让更多电路拥有自己的不同类型的机器。BNB Chain 的延展性比大多数人想象的大得多,TapeOut在为此全力以赴。
测试用例:tapeout.net/linux.html
@cz_binance @heyibinance @BNBCHAIN @fourdotmemezh
DEGENARIO retweeted
stocks is like bundling your own company and then farming it long term or am i tripping
DEGENARIO retweeted
when bitcoin & solana put in a +10% weekly candle we can really get this shit crackin
DEGENARIO retweeted
A black-box AI you can't audit is a promise, not infrastructure.
The connector, the node client, and the contracts are all public on GitHub. Read them, run them, verify them.
github.com/parallelixnetwork
#ParalleliX #AI #OpenSource
DEGENARIO retweeted
COMING SOON: PARALLELIX PODS
Your laptop can't fit a 70B model BUT your group chat can.
Pool your team's machines into one private AI mesh, shard models across them, get one `pk_pod_` endpoint.
Private routing, never dispatched to public-network operators.
In active development.
CA: 0x93FF39f65cC1D21067939961993ADF3f36BBF893
$PRLX $ETH #AI #ParallelixPODS #NodeCLIv3





























