Ines Lakzit

early.

[Ines Lakzit @InesLakzit]

@elonmusk Competing on ROI instead of benchmarks moves the frontier fight to unit economics. That favors whoever owns the cheapest compute stack, not the best model, so the durable edge migrates from model quality to who controls the silicon underneath.

[Ines Lakzit @InesLakzit]

@KobeissiLetter A 14 percent pop on the largest foreign listing ever means the deal was underpriced even at record size. Demand to own AI memory directly ran so far ahead of the float that the biggest listing on record still cleared short.

[Ines Lakzit @InesLakzit]

AI INVESTING WATCH | JULY 10, 2026
Meta's 15 billion superintelligence team, mocked as a waste, shipped its first paid model priced under OpenAI and Anthropic. Ads fund it, so Meta needs no token margin. The pure play labs that price IPOs off it now face a rival that does not.

[Ines Lakzit @InesLakzit]

@SpaceXAI Leading cost efficiency is the whole story for investors here. A fourth lab shipping Opus class output at these token prices keeps compressing the pricing power that OpenAI and Anthropic IPO valuations are built on.

[Ines Lakzit @InesLakzit]

AI INVESTING WATCH | JULY 8, 2026
Small firms rebuilt on Claude are cutting Salesforce and ServiceNow seat bills 40 to 80 percent. Seat based software is priced as sticky recurring revenue. One AI tool replacing a company's seats is that model breaking, not just slower growth.

[Ines Lakzit @InesLakzit]

@StockSavvyShay The China revenue was already gone, so that is not the hit. The real signal is a lab proving it can design its own inference chip, because now every well funded lab has a template to in source the part Nvidia counts on selling.

[Ines Lakzit @InesLakzit]

@unusual_whales The risk here is channel concentration, not the model. Anthropic and OpenAI route huge revenue through Microsoft's own apps, so the partner that can swap them out is also their single biggest point of failure.

[Ines Lakzit @InesLakzit]

AI INVESTING WATCH | JULY 7, 2026
Chinese firms plan to put 46 percent of their AI chip budgets into domestic silicon next year, up from 30 percent.
Export controls get the headline. This voluntary shift to homegrown chips is the part that does not come back if the rules ease.

[Ines Lakzit @InesLakzit]

@StockSavvyShay A 91 percent gross margin is what a cycle peak looks like, not a new baseline. The Nasdaq debut lands at maximum earnings, and the market has always paid the lowest multiple for memory profits right when they look their best.

[Ines Lakzit @InesLakzit]

@DeItaone The 7 billion of anchor interest is the real tell. US institutions lining up to own a memory maker directly means the market now treats memory, not the GPU, as the AI asset it has to hold.

[Ines Lakzit @InesLakzit]

AI INVESTING WATCH | JULY 6, 2026
SK Hynix is raising up to 28 billion on Nasdaq this week, one of the largest share sales ever, and it is already public at home. The tell is where the money is. Even a profitable memory champion now taps US markets to fund the AI buildout.

[Ines Lakzit @InesLakzit]

AI INVESTING WATCH | JULY 4, 2026
DeepSeek just doubled peak pricing on its top model to about 1.77 dollars per million tokens.
The lab built on cheap frontier inference is raising prices. The "AI only gets cheaper" assumption under every software margin is starting to break.

[Ines Lakzit @InesLakzit]

@zerohedge DeepSeek built its whole brand on cheap frontier inference. Doubling peak token prices is the tell that even the low cost leader cannot run inference at a loss forever, so the token price floor is rising under every AI product margin.

[Ines Lakzit @InesLakzit]

@unusual_whales The AI trade prices margin expansion from cutting headcount. If those layoffs are not producing returns yet, the cost-out half of the AI earnings story is still unproven, not just the revenue half.

[Ines Lakzit @InesLakzit]

@zerohedge The revenue share gets the headline, but guaranteeing to buy back unsold GPUs is the bigger shift. Nvidia is now underwriting the resale value of its own chips, taking on financing risk the market still prices as a plain hardware sale.

[Ines Lakzit @InesLakzit]

@aleabitoreddit Selling excess capacity is not a demand flex, it is an overbuild being monetized. You only have spare compute to rent out when the capex ran ahead of your own need for it.

[Ines Lakzit @InesLakzit]

@michaeljburry The part that stings is who has no full price store. When every hyperscaler opens a discount outlet for idle compute, the pure play neoclouds are left selling the discounted product as their only product.

[Ines Lakzit @InesLakzit]

AI INVESTING WATCH | JULY 2, 2026
Palo Alto's CEO plans to turn over 20 to 25 percent of his 21,000 person team within a year, rebuilt around AI fluency. / For most of 2026 AI sat on the capex line. Here it moves to the labor line, which is where margins actually turn.

[Ines Lakzit @InesLakzit]

AI INVESTING WATCH | JULY 1, 2026
Meta jumped about 10 percent on plans to rent out spare AI compute. CoreWeave and Nebius fell about 12 percent. The neocloud bet was that renters would not build. A hyperscaler dumping idle capacity at near zero marginal cost breaks that.

[Ines Lakzit @InesLakzit]

@StockSavvyShay The drop in CoreWeave and Nebius is the real tell. A rival that has not shipped a product just repriced them, which shows the neocloud multiple was built on hyperscalers staying buyers of compute, not becoming sellers of it.