
Morgan Stanely says cheaper AI could mean more compute, not less
The bearish case for AI infrastructure often starts with efficiency: If models become cheaper to run, the industry should need fewer GPUs, fewer data centers and less power.
Morgan Stanley argues the opposite.
In a research report released Monday, the bank applied Jevons paradox to AI. As inference becomes cheaper, businesses can run more tasks, take more agentic steps and deploy models across more workflows. Efficiency reduces the cost of each unit of compute, but the resulting increase in usage could push total compute demand higher.
Morgan Stanley estimates that an average AI task creates about $55 of value while costing only $2 to $5. That gap leaves plenty of room for adoption to expand as prices fall.
Open-weight models could also change where the compute runs. Instead of concentrating every workload inside a handful of hyperscale clouds, enterprises can deploy models through public clouds, private data centers, on-premises clusters, or APIs.
A McKinsey survey cited by Morgan Stanley found that 63% of respondents already use open-source models, often alongside closed systems. The likely result is a hybrid market: frontier models for complex reasoning and open models for high-volume or specialized work.
The bank sees a few companies benefitting from the lift, namely neo-cloud providers like CoreWeave and Nebius.

Power providers including Bloom Energy, Innio, Solaris Energy Infrastructure, Williams Companies and Liberty Energy gain from the need for reliable on-site generation.
Cheaper tokens may pressure model pricing, but cheaper AI could ultimately mean more GPUs, more facilities, and more power demand.
Today on The Blockspace Podcast
We kick off earnings week with a market update and a look at investment banks raising their AI infrastructure stock targets.
DeepSeek is reportedly targeting 1 GW of computing capacity in Inner Mongolia. Fortitude Mining CFO Erik Ellingson joins the show to discuss the company’s Zcash mining strategy, and analyst and investor Ben Pouladian brings questions about AMD’s latest chip.

BLOCKSPACE
Earnings Week Kickoff, DeepSeek Taps Inner Mongolia for 1 GW, Banks Up AI Stock Targets

In the News
Texas Governor Abbott orders data center audit as ERCOT queue reaches 474 GW
Texas Governor Greg Abbott directed state regulators Monday to audit every data center advancing through ERCOT’s interconnection process, preventing affected projects from moving forward until the review is complete.
The Aug. 3 directive requires the Public Utility Commission of Texas and the Electric Reliability Council of Texas to verify each project’s power demand, water use, public financial assistance, community protections and ownership. Projects that fail to meet PUCT, ERCOT or state requirements must be denied a grid connection.
DeepSeek targets 1 GW of AI compute in Inner Mongolia
DeepSeek is planning roughly 1 GW of computing capacity in Ulanqab through a combination of company-owned infrastructure and leased capacity, according to Bloomberg. The Chinese AI startup wants part of the footprint operating by late 2027 or early 2028. DeepSeek has not disclosed whether the gigawatt figure represents critical IT load or total facility power, nor has it identified the chip mix, construction partners, financing structure or full buildout cost.
Morgan Stanley says open-weight AI could expand compute and power demand
Morgan Stanley says lower inference costs could increase AI usage enough to offset efficiency gains. Open-weight models may distribute workloads across public clouds, private facilities and on-premises systems rather than reducing total infrastructure requirements. NVIDIA appears as a leading beneficiary across the bank’s scenarios, alongside CoreWeave, Nebius, and several on-site power providers.
B. Riley raises TeraWulf’s price target to $40
B. Riley Securities raised its TeraWulf (NASDAQ: WULF) price target from $32 to $40 while maintaining a Buy rating. The target implies roughly 124% upside from the $17.82 reference price in the bank’s note. B. Riley pointed to TeraWulf’s 20-year Anthropic lease at its Kentucky campus, covering approximately 401 MW of critical IT load and an estimated $19 billion of contracted revenue.
B. Riley lifts CleanSpark’s target to $26 on HPC potential
B. Riley raised its CleanSpark (NASDAQ: CLSK) price target from $19 to $26, implying about 81% upside from Monday morning’s share price. The valuation gives substantial weight to CleanSpark’s contracted 175 MW Sandersville lease and prospective AI/HPC agreements at its Sealy and Brazoria County campuses. Turning the Texas exclusivity arrangement into signed leases remains the central catalyst.
Matt Prusak leaves American Bitcoin for Giga Energy
Matt Prusak is joining Giga Energy as chief business officer and interim CFO. He will leave his roles as president and interim CFO of American Bitcoin (NASDAQ: ABTC) effective August 4. American Bitcoin appointed derivatives head Paul Sacks as interim CFO as the company reported $67 million in second-quarter revenue and a $57.2 million net loss.

The Daily Meme
Before you go, do you mind filling out this quick survey? It truly helps us understand who is reading and listening to Blockspace!
