@LucasSacerdote_ Oh I see. But so then your $1b Ebitda in ‘27 estimate must be for just the US operations? Seems steep. Do you have this modelled out somewhere that I can check out? I also get to just over $1b in ‘27 but that includes CSI and US (excl Recurrent)
OpenAI / Altman just got a body blow from $MSFT Satya Nadella:
Microsoft plans to move Copilot to usage-based pricing with open-source AI model DeepSeek instead of OpenAI due to costs $NVDA $AMD $MU $ORCL
Even though this represents <5% in pro forma revenue for OpenAI, others could follow suit, creating a bigger issue...
Microsoft CEO Satya Nadella's shift to offer DeepSeek-V4 alongside a move to usage-based token pricing for Copilot Cowork is an absolute watershed moment for the AI industry.
The era of "all-you-can-eat" flat-rate corporate subscriptions is ending because agentic AI (AI that continuously plans, codes, and loops in the background) is dramatically more expensive to run than simple chatbots. The financial and strategic ripple effects across tech, hardware, and venture capital break down as follows:
- AI Models Cost Prohibitive: OpenAI (GPT-5.5) and Anthropic (Claude 4.8) have priced themselves into a corner. Reportedly, DeepSeek V4 Pro costs roughly one-third as much as Claude for inputs and one-seventh as much for outputs.
- Taking Away Pricing Leverage: Microsoft is OpenAI’s biggest backer, but Nadella is explicitly executing a "multi-model platform strategy." He noted on X that "we do not want a world where enterprises cede all value to a small number of models." By introducing a hyper-efficient open-source competitor, Microsoft is stripping OpenAI of its premium pricing leverage.
- What this Could Mean for AI Hardware: The shift to open-source, highly optimized "Mixture-of-Experts" (MoE) architectures like DeepSeek means enterprise software can achieve the same results using significantly fewer compute cycles. If the industry shifts from bloated, massive frontier models to lightweight, fine-tuned open-source models, the frantic, exponential panic-buying of premium AI chips could cool down longer term.
o Bull Interpretation: AI is moving to usage-based enterprise billing. This transforms AI into a metered utility (like electricity or AWS cloud storage). As enterprises optimize their costs via cheaper models, total volume and actual utilization of AI agents will skyrocket. The sheer scale of agentic workflows running continuously will still require an immense, unyielding baseline of infrastructure.
- This development explains exactly why the public market has started applying a heavy discount to proxy investment vehicles holding pre-IPO tech blocks.
o Closed-end venture vehicles like $DXYZ, $VCX, and Powerlaw Corp. $PWRL hold concentrated secondary allocations in OpenAI and Anthropic, both of which are currently in the confidential stages of prepping trillion-dollar IPOs. With Microsoft proving that it will happily swap out closed-source models to preserve its own corporate margins, the public markets are realizing that the long-term enterprise software moats of OpenAI and Anthropic are far less secure than originally hyped.
- Political Issue: This move introduces a massive geopolitical paradox. The Trump administration has been actively tightening restrictions on advanced AI models leaking out of the United States. Now, Microsoft—one of America's most foundational tech institutions—is adopting a model born out of a Chinese lab (DeepSeek) simply because the raw economics of American frontier models are unsustainable. To protect itself from political blowback, Microsoft has heavily modified the model, layered it with Western safety/bias overrides, and restricted it entirely to self-hosted Azure environments.
@InvestNorthwise They actually announced it in March already, but thats neither here nor there. At $10m capex per MW, theres no way the current stock price is justified.
@InvestNorthwise I believe your statement about Moses Lake to be incorrect as 3 weeks ago they said it is not going to be GPUs as a service.
On your CIFR analogy.. CIFR is indeed doing powered shells but their capex is closer to $10m/MW, which seems much more realistic.
Firstly, leaving out the capex makes the PV less conservative, not more. Secondly, the $750m is for the natgas and helium expansion, not nuclear. Thirdly, whether the capex is funded by equity or debt doesn't change the fact that it raises the EV. Lastly I regret to have engaged with this pump account.
Pls explain to us simpletons how this is a problem when global inventories is >6 billion barrels. The outage is like 8mmbls/d? At this rate we can survive 2 years? Understand that flow rates from inventories is capped at like 4mmbls/d. So everyone just reduce consumption by <4% and we’re good?
Keith the discounts only apply to their open market home sales, which historically have a 12-15% margin. So discounting 10% still leaves them in profit while unlocking working capital which alleviates any liquidity concerns. The 8% margin you reference is the group wide margin including partnerships.
@antoniovelardo_@SharpInvesting1@Frederik_1234 Their gearing covenant is calculated on period end net debt. Not avg daily net debt. Very hard to see VTY breaking any of their covenants.
@Floebertus The incentive is not to rebuild your org’s saas internally. It’s to replace the humans clicking around in the app.
Rebuild saas internally, save $100k;
Replace 3 employees with agents doing the clicking, save $400k.
@markcboon Well at least he didnt say "next quarter" once during this call. Instead changed his wording for commercial delivery timelines to "imminently". He's probably been told people are calling him Paul "next quarter" Mann.
@broheim777 Don't you think MSFT is more likely expanding to the Tract property right adjacent to them as opposed to the SSOF land?. Tract site (1,063 acres and 700MW capacity) is specifically being developed for a datacenter with no mentioned customer yet: https://t.co/xlh9QZEkBD https://t.co/XZzzBh4Zvt