Only 4% Are Cutting Jobs, Pancreatic Cancer Survival Doubles, and NVIDIA’s $96.2B Quarter | EP #284
Nvidia's record $96.2B quarter reopens the circular-financing/bubble debate, with Alex Wissner-Gross flagging Nvidia's vendor financing of its own customers as the real 'elephant in the room' while Dave Blundin frames the whole AI economy as a self-contained Alpha-Centauri-style civilization indifferent to the legacy economy. OpenAI's new Jalapeno inference chip and Apple's 512GB-memory Mac Studio both signal the industry fragmenting away from pure Nvidia dependence, even as Nvidia's interconnect (post-Mellanox) becomes the new moat. A Principal Financial Group survey demolishes the AI-jobs-apocalypse narrative (only 4% of SMEs expect AI to cut staff/wages; 52% have added staff), while X's takedown of a 200,000-account Chinese bot farm seeding anti-data-center sentiment collides with Quincy, WA's real data-center-driven poverty turnaround. The panel also covers a first-in-class RAS-inhibitor pancreatic cancer drug nearly doubling median survival, tooth-enamel regeneration, weather-modification drones, lunar manufacturing economics, Tesla's Cyber Cab rollout, and a robotic-intimacy startup, closing on China's new crackdown on AI companion apps for minors.
Nvidia's $96.2B quarter and circular-financing bubble fearsEconomy▶ 0:00
Nvidia posted $96.2B quarterly revenue (up 106% YoY, 80-85% gross margins) with Jensen Huang guiding 70% growth into 2028; Alex Wissner-Gross flags Nvidia financing its own customers' data-center buildouts as an unpriced 'elephant in the room,' while Dave Blundin argues the AI economy is a self-sustaining closed loop (like an alien civilization) that doesn't need the legacy economy to validate it.
SpaceX's $3.5 trillion revenue projection by 2033Space▶ 14:19
Elon Musk projects SpaceX could hit $3.5T revenue by 2033 (vs. Amazon's current $828B); Alex Wissner-Gross attributes the path mainly to Optimus robots and Starlink/hyperscaler cloud plus an eventual Dyson swarm, not applications or cars, which he calls 'rounding errors.'
China's video-gen boom vs. America's LLM/RSI focusAI▶ 24:53
Alibaba's Wan 3.0 generates cheap 30-second videos; video generation is ~70% of China's AI token consumption. Alex frames US labs as 'revenue maxing' on code/enterprise while Chinese labs give away weights and burn tokens on video; Dave argues the real race is recursive self-improvement (RSI), which both US and Chinese labs are quietly chasing regardless of surface-level product focus.
China regulates AI companions amid demographic collapseAI▶ 35:00
China is banning AI companion services for minors and restricting them for adults, citing loneliness and falling birth rates (China ~1 child/family, South Korea 0.75). Alex predicts the CCP will eventually reverse course once it realizes party-aligned AI companions are a control tool, not a threat.
Flock Safety license-plate surveillance and civil libertiesGeopolitics▶ 45:01
Flock Safety's AI license-plate camera network (6,000+ communities, 20B+ scans/month) draws scrutiny after reports of police misusing access to stalk ex-partners; Salim Ismail argues mass surveillance's cost has dropped to zero, eliminating the practical friction that used to protect civil liberties.
Job apocalypse myth-busting: only 4% of SMEs cutting staff over AIEconomy▶ 1:07:43
A Principal Financial Group survey of 130,000 small/medium businesses found only 4% expect AI to reduce staffing/wages, 31% expect increases, and just 1.4% of any staff reductions were attributed to AI; 52% increased staff in the last 3 months. Salim contrasts this with 'doomsday' commentary from Bill Gates and Ray Dalio.
OpenAI's custom inference chip (built with Broadcom) delivers 1.5-1.9x more work per watt and up to 3.6x lower latency than Nvidia's GB200/GB300 at half the power draw, with a claimed ~54x throughput gain hosting GPT-OSS; Alex speculates OpenAI could become its own hyperscaler/cloud provider.
Nvidia's CUDA moat fading, interconnect becomes the new moatCompute▶ 1:31:03
Dave Blundin explains CUDA is already a dead moat at inference time (easy to port to AMD/Intel) but still dominant for frontier training research; Nvidia's real remaining moat is Mellanox-derived interconnect needed to cluster hundreds of thousands to millions of GPUs coherently.
Apple's Mac Studio (M5 Ultra, 512GB) enables local/on-prem AICompute▶ 1:33:17
Apple's new Mac Studio with M5 Ultra and up to 512GB unified memory can run large open models locally; Salim sees big wins for law firms and HIPAA-bound healthcare that can't use the cloud, while Alex/Dave criticize Apple for having 'no AI strategy' despite ideal hardware lineage (unified memory descended from the Apple Car project).
Uranium enrichment breakthrough (Actinide's HALEU) and China's solar/nuclear surgeEnergy▶ 1:41:03
Startup Actinide demonstrated uranium enrichment to 15.38% purity (5x traditional levels), potentially solving the fuel bottleneck for SMRs; separately, China's monthly solar generation grew 8x since 2020 while US nuclear generation stayed flat, and Dave notes most US solar panels are made in China.
Rainmaker's cloud-seeding drones and programmable weatherOther▶ 1:49:53
Rainmaker's drones generated an estimated 19M gallons of rain over Alaska's Kenai Peninsula in 3 hours via cloud seeding; Alex envisions a permanent solar-powered drone swarm under AI control enabling hurricane dispersal and a tradeable global 'rain market.'
First RAS-inhibitor pancreatic cancer drug nearly doubles survivalHealth▶ 1:56:44
FDA approved daraxonrasib, the first RAS inhibitor for metastatic pancreatic adenocarcinoma, nearly doubling median survival (6.7 to 13.2 months) and tripling tumor response rate (11% to 32%); RAS mutations drive ~30% of all human cancers, suggesting broader applicability.
Biomimetic gel regenerates tooth enamelBiotech▶ 1:59:19
University of Nottingham researchers developed a fluoride-free, elastin-like protein gel that regrows true tooth enamel (not a synthetic filling) by directing calcium/phosphate mineralization; still tested ex-vivo on extracted teeth but expected to move to human trials soon.
Lunar manufacturing economics and Star Mine orbital computeSpace▶ 2:05:24
Deloitte estimates $566B cumulative lunar economic value by 2050, which the panel calls a lowball; Elon Musk's Star Mine concept (manufacturing AI satellites/data centers on the moon) could theoretically generate ~$2 quadrillion/year in gross revenue at 100 terawatts of orbital compute.
Predictions made
openAlex Wissner-Gross: The CCP will reverse its stance and embrace AI companions once it recognizes them as a party-control mechanism over youth, rather than continuing to restrict them.
“I think the party decides actually having... a sizable fraction of the youth of China deeply engaged with party deeply controlled AI companions is actually from the party's perspective net helpful not net harmful.”
Your call:
openAlex Wissner-Gross: At least some US states will legalize AI companionship, including romantic and robotic romantic companionship, even as other states try to ban it.
“My bet is that at least in some places in the US AI companionship including romantic companionship including robotic romantic companionship will be legal.”
Your call:
openAlex Wissner-Gross: OpenAI will launch its own hyperscaler cloud offering (hosted on its own chips), potentially even hosting rival models like Anthropic's, similar to how Elon Musk's SpaceX hosts others' compute.
“I could imagine a world where say an open AI compute cloud hosts an anthropic model and OpenAI and Anthropic win at the same time.”
Your call:
openAlex Wissner-Gross: SpaceX/Tesla will merge within the next year or two, giving the combined entity Optimus robot revenue alongside hyperscaler cloud and Dyson swarm business lines.
“My bet is Japan, Singapore, Korea is going to follow closely.”
Your call:
openDave Blundin: OpenAI's inference chip advances mean inference workloads are permanently moving off Nvidia hardware, even as Nvidia retains the training market.
“There's no doubt inference is moving off of Nvidia for sure... this is the first 2x of 100x performance gains.”
Your call:
openAlex Wissner-Gross: Daraxonrasib will be among the last generation of pancreatic cancer drugs developed without AI-driven tools (AlphaFold-style protein folding, virtual cell models), with AI-designed successors pushing response rates toward near 100%.
“I think this is probably slash hopefully one of the last generation of pancreatic cancer medications that almost precedes the AI wave.”
Your call:
openPeter Diamandis: His son will never need to get a driver's license because robotaxis (Cyber Cab) will be widely available before he reaches driving age.
“My kid will never get a driver's license and I've got just about a year for this cyber cab to roll out to get out there before... he wouldn't have to.”
Your call:
openAlex Wissner-Gross: Human motion-capture-trained intimacy robots (like Somnia's Model L) won't scale as an approach compared to pre-training robots directly on internet video for physical/embodied tasks.
“It doesn't scale nearly as well as just pre-training off of internet video is my guess.”
Your call:
Numbers that matter
$96.2 billionNvidia's revenue in a single quarter, up 106% year-over-year.
108 billionNvidia's guidance for the following quarter's revenue (about 10% higher).
$3.5 trillionElon Musk's projected SpaceX revenue by 2033, versus Amazon's current $828 billion (highest-revenue company today).
200,000 accountsSize of the Chinese bot farm X identified, with 200 accounts specifically posting anti-data-center narratives.
29.4% to 6.2%Poverty rate drop in Quincy, Washington, attributed to data center tax revenue.
4%Share of surveyed small/medium businesses (Principal Financial Group, 130,000 customers) that anticipate AI will reduce staffing and wages.
1.4%Share of businesses that reduced staff who attributed it to AI/automation.
52%Share of surveyed SMEs that increased staff over the last 3 months (vs. only 12% reduced).
1.5-1.9x / 3.6xOpenAI's Jalapeno chip's efficiency (AI work per watt) and latency improvement vs. Nvidia GB200/GB300.
~54xClaimed throughput-per-second-per-user increase OpenAI got hosting GPT-OSS on Jalapeno vs. existing (presumably Nvidia) hardware.
512 GBUnified memory capacity of Apple's new Mac Studio with M5 Ultra, enabling large open AI models to run locally.
15.38%Purity of high-assay low-enriched uranium (HALEU) Actinide demonstrated, 5x the traditional 3-5% enrichment level.
800% (8x)Growth in China's monthly solar generation from 2020 (20 TWh/month) to mid-2026 (~160 TWh/month).
19 million gallonsAdditional rainfall generated by Rainmaker's 10 cloud-seeding drones over Alaska's Kenai Peninsula in 3 hours.
6.7 to 13.2 monthsMedian survival increase from daraxonrasib in previously treated metastatic pancreatic cancer, nearly doubling survival.
Worth digging into
🕳️ Nvidia's vendor-financing exposure and circular revenue
Alex Wissner-Gross flags that markets may not have priced in how much of Nvidia's demand is directly or indirectly financed/backstopped by Nvidia itself through intermediaries - a structural risk distinct from normal 'circular economy' trade.
🕳️ Actinide's HALEU enrichment method vs. calutron alternatives
Alex claims US uranium enrichment is still stuck on WWII-era centrifuge/gas technology while better methods (electromagnetic calutron-style separation using modern power electronics) have been technically feasible for 80 years but unused.
🕳️ Star Mine lunar manufacturing economics
Musk's claim of $2 quadrillion/year in orbital compute revenue from 100 TW/year of lunar-manufactured compute is an order-of-magnitude jump from Deloitte's $566B lunar economy estimate - worth independently modeling the assumptions.
🕳️ OpenAI's Jalapeno chip and potential OpenAI Compute cloud
A ~54x throughput claim on GPT-OSS is extraordinary and, if OpenAI moves to hosting third-party models (per Alex's speculation), would reshape the inference market and Nvidia's customer base.
🕳️ Principal Financial Group SME AI-jobs survey
The 4%-expect-cuts / 52%-increased-staff data directly contradicts prevailing AI-jobs-apocalypse narratives from figures like Bill Gates and Goldman Sachs, and Salim offered to share the full report.
🕳️ X's bot-farm takedown of anti-data-center influence ops
The claim that a 200,000-account Chinese bot farm specifically targeted American data-center sentiment is a significant geopolitical claim that would benefit from independent verification beyond X's own safety team post.