Great to see the @Meta x @oklo partnership announced this morning.
Meta is expanding its data center footprint in Ohio and exploring nuclear power to support future AI infrastructure. Oklo develops small modular nuclear reactors designed for long term, on site energy generation.
AI scale is an energy problem, not just a compute problem. I expect to see many more of these types of partnerships.
@jake_ledner@PalantirTech@Dyson The Dyson Institute also pays students $45k/yr.
Learn by doing, get paid, avoid student debt. Love to see it.
James Dyson explained his thinking behind this on the @davidsenra show recently - great listen.
@Alisa__miss@OzloSleep@WHOOP Rarely got more than 7 hours before Ozlo. Now, I’m consistently at 7-8, and sometimes over 8 which previously felt near impossible for me.
Like many, I struggle to get enough sleep. @OzloSleep helps me add about 30-60min each night, measured by my @WHOOP.
I shout from the mountaintops about Ozlo because it improves both the front and back end of sleep.
Front end: I used to scroll endlessly. Now I doze off listening to a podcast that kicks over to white noise the moment Ozlo detects that I'm asleep.
Back end: I used to wake too early unable to return to sleep. Now I tap to resume my podcast, drown out the morning thoughts, and actually fall back asleep.
Game changer.
I'm a day one @OzloSleep customer that's dealt with the bugs that early adopters usually do, and the initially limited sleep content library. It's been exciting to see Ozlo continuously improve the customer experience through consistent software/firmware updates and new partnerships like this.
Ozlo recently announced a partnership with @calm to provide a more complete sleep experience.
Wins for Ozlo: Adds a huge sleep content library + a new distribution channel to potentially reach Calm's 4M+ paid subscribers
Wins for Calm: Breaks into physical sleep products for the first time + access to a high value customers that perfectly match their ICP (people willing to spend $300+ on sleep products)
Wins for consumers: One stop shop sleep system + a free one year subscription to Calm
@friedberg points to Google’s head of AI and infra citing that 7-8 year old TPUs are still fully utilized and that justifies a 6 year useful life. This is true, but it’s important to also examine the method of depreciation used and whether that aligns with the actual value provided by the assets within the period.
So the question is: Do GPUs actually depreciate on a straight line basis?
Meta, Amazon, Google, Microsoft and CoreWeave clearly think so.
SDH100RT index which tracks GPU prices is down 28% this year.
https://t.co/G0fjpiyNxg
That’s meaningfully steeper than the ~16–20% annual hit you’d take on a 5- or 6-year straight-line schedule.
Price declines of that magnitude are worth monitoring. If they persist, it would make a strong case for accelerated depreciation, like 150% declining balance.
To be clear: even if Meta or Google use a near-term friendly method like straight-line over six years, this is not “cooking the books.” GAAP allows management to choose any reasonable method if supported by justification.
It just means they are punting greater expenses to a future period. And in that future period, they will have lesser economic value provided relative to book expenses.
Valid accounting but it’s a fair basis for people like Burry to be skeptical of future performance.
In fact, Jensen has even joked he’s NVIDIA’s “chief revenue destroyer” because they actively try to obsolete old chips with new ones that are orders of magnitude better. So it’s plausible that Burry isn’t totally off base about scrutinizing depreciation.
Oddly, though, the price decay and obsolescence risk hits hyperscalers harder than NVIDIA or application companies like Palantir. They are the ones holding billions of dollars of hardware on their balance sheets.
This leads to an interesting scenario:
NVIDIA keeps releasing dramatically better chips, Palantir and other app companies see massive demand for inference, all the while, hyperscalers get squeezed as their older GPU fleets lose value faster than expected.
The $100M OpenAI and Intuit partnership is another sign of the shift to an AI-native UI. Users could upload financial documents and have ChatGPT extract and input the details for tax preparation or other Intuit workflows.
Two issues stand out: data privacy and AI mistakes
How comfortable will people be uploading their most sensitive financial information to ChatGPT? And will we see more mistakes or compliance problems as people start blindly trusting AI with their financial tasks?
Competitors Unity and Epic Games just announced a major partnership to “advance an open, interoperable future for gaming"
Translation: Fortnite is becoming a UGC platform and millions of Unity devs can now build and monetize inside it (a direct shot at Roblox)
Reminiscent of Sony and Microsoft teaming up in 2019 on cloud gaming when the market opportunity and doing what's best for the dev community outweighed the rivalry
Simple summary of the new NVIDIA, Microsoft, Anthropic partnerships (multiple 2 way deals forming one 3 way alliance):
• Anthropic committing $30B of spend on Azure
• NVIDIA investing $10B in Anthropic
• Microsoft investing $5B in Anthropic
Super high level wins for each:
• Microsoft: A $30B cloud customer + Claude on Azure
• NVIDIA: Guaranteed long term demand as Anthropic continues building its models on NVIDIA chips
• Anthropic: Funding, distribution, massive compute access, and optimized hardware for future models
Tried to read an article on NVIDIA’s website about the newly announced NVIDIA, Anthropic, Microsoft partnership and ran straight into the Cloudfare outage. Talk about bad timing lol 🤦♂️
ProSafe just partnered with TruePath Vision to bring computer vision into hotel safety systems. Instead of relying only on panic buttons, hotels can now use existing cameras to detect risks like weapons and suspicious behavior in real time.
The idea is promising, but the real test comes down to accuracy and latency. False positives can be highly disruptive, and delayed or missed alerts during active threats are far worse.
If they get it right, hospitality could see something closer to what ZeroEyes has delivered for schools.
With Waymo now offering fully driverless freeway rides, the gap in autonomous capability continues to widen.
Last week, @AMD and @STRADVISION announced a multi-year partnership that brings STRADVISION's perception software together with AMD’s automotive SoCs in a co-optimized, production ready stack.
Why it matters: this gives automakers and Tier 1 suppliers a clear, viable path toward L3 capability without having to redesign the entire electronic architecture, and helps narrow the distance with AV leaders.