SCOOP: US officials have had talks about having government acquire shares in AI giants, sources say
Altman has discussed w/ senior officials, including Trump. Did again recently
May be *ceding* shares to USG - not a purchase
Shares could go 2 dividend
https://t.co/ks2Yzgr2W0
South Korean chipmaker SK Hynix, a major Nvidia supplier, said investor feedback on its proposed US listing was 'tremendously positive,' with a source saying the deal could raise $14 billion as its market value topped $1 trillion last week https://t.co/9W8VPBqwDp
Let me give some "behind the scenes" as to why AI ROI is so elusive. Even if the AI works, you have to navigate the "Seven Gates of Software Hell".
I ran AI for a company that managed a huge portion of the world's communications data for financial services companies. This is an excruciating read but the realities are tough.
Let's get started. Suppose you want to scan all of your communications for customer complaints and respond quickly. Here's your journey:
Gate 1: Data Controls
Various geographies require the data to be stored in-region, and in some cases, only accessed in-region. You may need separate AI deployments for each one.
The data may need to be scrubbed for PHI/PII and will need to be scrubbed for Material Non-Public Information. If it leaves the System of Record you'll need to ensure there is a way to selectively delete data so that you can adhere to GDPR, CCPA or PIPL.
Gate 2: Data Quality
Even if you get controls in place, you discover that your data is coming from 8 different vendors. Some are real-time, others T+1 and they all have different APIs. To boot, your corporate directory has 4 identities for Brandon Carl that have never been merged so you can't properly query even a single person's data.
Gate 3: Security + Controls
Given the sensitivity of the data you're sending, you'll need to go through an extensive security audit. Since this is an LLM you'll need to look beyond SOC2 and into OWASP Top 10 LLM risks and Gen AI risks too.
Gate 4: SLAs
Your AI Agent calls are taxing your system with bursty volumes and risking your mission-critical production workloads. You may need to set up read-only replicas, throttling and overage billing.
Gate 5: Vendor Risk
Your vendor will be assessed for their financial viability as well as the controls they put in place. This may go as far as analyzing the vendor's software development processes.
Gate 6: Legal + Procurement
You've almost made it, but procurement needs to demonstrate that they are saving the firm money. Negotiations come down to the end of the quarter. Redlines are flying everywhere to meet your firm's AI policies and to ensure there's no training on your data.
Gate 7: Model Governance
The AI/ML models need to be assessed versus your firm's Responsible AI Policy. And if you're going to automate things get really tricky. The model needs to be assessed for Materiality, Autonomy and Complexity. You may need documented evaluations, extensive model documentation and champion challenger comparisons performed by your own internal AI teams.
–––
You've made it this far, congratulations!
While you've been working through the "7 Gates of Hell" you've had to manage a team of workers you know you're going to fire to justify the AI spend. This requires coordination with HR, one-time separation costs and managing team morale for those employees that stay.
Thanks @emollick for posing the question. Also see https://t.co/ISNlNgDfvH
The Hormuz crisis is serious. But this isn't 1973, writes @DanielYergin
Today's world has shale, LNG, strategic reserves, bypass pipelines & far more diversified supply.
Markets have learned how to absorb shocks.
That may explain why comparative inventory data isn't confirming the most extreme "tank bottom" forecasts.
#OilMarkets #WTI #Hormuz #EnergySecurity #CrudeOil #EnergyMarkets #OilPrices #InventoryData #DanielYergin #Geopolitics
https://t.co/swIXDhmR1R
RAY DALIO JUST SAID HE THINKS THE AI BUBBLE IS GOING TO POP
On Bloomberg, he laid out exactly when, why, and how he sees it happening.
He also compared the current market to two specific years:
"We are right now rising closer to, not at, the same level in 2000 and the same level in 1929, a specific level where you say, oh no, here's the one that we really need to worry about."
On all tech revolutions:
"All great technology changes produce bubbles. And the reason they produce bubbles is because nobody can get it exactly right."
On the mechanism that pops bubbles:
"There's a bubble, and then there's the pricking of the bubble. The pricking of the bubble happens when there's a need for wealth to be sold to get the money."
On the wealth gap that comes with it:
"A very small percentage of the population is going to do unbelievably and a lot of people won't."
On a political solution:
"I'm not optimistic on us working together to solve."
He says the technology is real. The prices and the debt are the problem.
Welp, that happened faster than I predicted. Thought it would be end of 2027, then early 2027, but agentic traffic growing so fast that bots have now passed human traffic online for the first time in the Internet's history. https://t.co/2zX5bHdhsa
Google Cloud revenue showed a +63% y/y growth this past quarter. Microsoft Intelligence Cloud revenue showed a +30% y/y growth this past quarter. AWS revenue showed a +28% y/y growth. Despite this, AWS' margins increased 213bps q/q while the other CSPs lagged behind. How you sell tokens is become equally important to how much of it you sell. Bedrock's TaaS (token-as-a-service) business model with Anthropic has 3 parts:
🟠 fixed IaaS fee,
🟠 revenue share of the tokens,
🟠 and performance hurdles that trigger outperformance payments above certain token/spend thresholds.
The risk with this business model is that there's no guaranteed take-or-pay floor so revenue can miss if adoption stalls but their bet paid off, primarily driven by Anthropic's addition of $21B net new ARR in a single quarter.
▶ Marvell CEO says copper wall is moving inside the rack, and copackaged optics is the only way through
• Marvell CEO Matt Murphy emphasized at Computex 2026 that the next bottleneck in AI infrastructure is not compute or memory but connectivity.
• The shift from copper to optical interconnect is already underway and is expected to trigger a large scale demand cycle within the semiconductor industry.
• He highlighted Marvell’s sophisticated engineering capability, integrating advanced CMOS DSP, fourth generation SiPh, and SiGe based broadband analog technology through its Coherent optical modules.
• Marvell’s first 102.4T switch dedicated to AI data centers, the Teralynx T100, is built on a 3nm process, draws under 1,000W, and delivers up to 25% lower power than competing solutions.
• The T100 routes signals through copper traces on the PCB to optical modules on the front panel, whereas a CPO switch connects optical fiber directly to the package and removes copper wiring entirely.
• The reach of copper cable is inversely proportional to bandwidth: at 100Gbps per lane it can carry signals about 5m, but at 200Gbps this shortens to roughly 2.5m, and at 400Gbps copper can no longer make connections even within the rack.
• Each time the “copper wall” moves one step, the number of connections that must shift to optical increases at least tenfold, which is expected to drive explosive demand across the optics industry, and the Taiwan supply chain is already expanding to respond.
• The number of connections inside a rack is roughly ten times the number of connections between racks, so conventional pluggable optical modules alone cannot address the power and space limits; CPO solves the connectivity problem by integrating the optical engine directly into the switch and compute package.
• Nvidia’s Vera Rubin platform has already adopted Spectrum-X Ethernet Photonics, the first CPO based switch to enter mass production, a case showing that the CPO transition has moved beyond proof of concept into actual commercialization.
• As optical connectivity extends into the server itself, compute, memory, and network resources can be disaggregated and dynamically configured per workload, enabling a shift from a fixed server architecture toward operating the entire data center as a single integrated system.
• He stressed that the CPO transition is impossible without Taiwan’s manufacturing ecosystem, explaining that Marvell has accumulated high volume PAM4 production experience, field data, and supply chain infrastructure including ASE.
• Over the past decade Marvell has invested a total of $36 billion to acquire companies such as Inphi, Cavium, and Celestial AI, expanding its connectivity portfolio.
• He emphasized that Marvell is the only company able to address the full connectivity stack of an AI data center, from millimeter scale inside the package to kilometer scale between data centers.
$MRVL
1/6
Financial Times: "A company-level OECD analysis of government subsidies across 15 key industrial sectors found that nearly 60 per cent of Chinese firms’ global market share gains since 2005 could be attributed to subsidies."
https://t.co/o2mcZeaagi
Short seller Andrew Left, founder of Citron Research, found guilty of securities fraud by a federal jury in Los Angeles.
Left was accused of using tweets about dozens of companies to illegally influence their share prices and profit from the moves. Prosecutors said he made ~$20M from such trades between 2018 and 2023.
Left took the rare step of testifying in his own defense. Trial lasted three weeks, jury deliberated two days.
Wow, Berkshire Hathaway investing $10 billion into $GOOGL in a private placement as part of a broader $80 billion equity capital raise by the company to expand its AI infrastructure
OpenAI frontier models and Codex are now generally available on AWS, giving enterprises a new way to build on Amazon Bedrock with OpenAI through the security, compliance, and governance workflows they already use.
This is also the beginning of a broader expansion of OpenAI capabilities on AWS, including future availability for cybersecurity capabilities like Daybreak.
https://t.co/vMws0YU6Q3
Nvidia is entering the PC market with a new chip aimed at loosening the stranglehold of Intel technology in that arena and modernizing the machines for the AI era https://t.co/OsyA9C4wi0
Joel Greenblatt (Gotham Asset Management) 13F largest holdings as of March 31, 2026:
(1) SPY (S&P 500 ETF) -- 18% of portfolio
(2) Apple -- 2% of portfolio
(3) Nvidia -- 2% of portfolio
Large-cap US ETF’s hold $5 trillion in assets.
SpaceX, OpenAI, and Anthropic’s will drop about $1 trillion on the market (ballpark guess).
I normally ignore flow data, but after @vaneckpk ‘s nudging, this deserves attention.